The document describes a two-layer classification technique for online handwritten Gujarati character recognition. In the first layer, an SVM classifier with an RBF kernel is used. In the second layer, a k-NN classifier is used for characters identified as similar by the first layer. The system achieves an average accuracy of 94.65% and processing time of 0.095 seconds per stroke using a hybrid feature set including derivatives, zoning, and normalized chain codes.
Automated Bangla sign language translation system for alphabets by means of M...TELKOMNIKA JOURNAL
Individuals with hearing and speaking impairment communicate using sign language. The movement of hand, body and expressions of face are the means by which the people, who are unable to hear and speak, can communicate. Bangla sign alphabets are formed with one or two hand movements. There are some features which differentiates the signs. To detect and recognize the signs, analyzing its shape and comparing its features is necessary. This paper aims to propose a model and build a computer systemthat can recognize Bangla Sign Lanugage alphabets and translate them to corresponding Bangla letters by means of deep convolutional neural network (CNN). CNN has been introduced in this model in form of a pre-trained model called “MobileNet” which produced an average accuracy of 95.71% in recognizing 36 Bangla Sign Language alphabets.
An Optical Character Recognition for Handwritten Devanagari ScriptIJERA Editor
Optical Character Recognition is process of recognition of character from scanned document and lots of OCR now available in the market. But most of these systems work for Roman, Chinese, Japanese and Arabic characters . There are no sufficient number of work on Indian language script like Devanagari so this paper present a review on optical character recognition on handwritten Devanagari script
Spoken language identification using i-vectors, x-vectors, PLDA and logistic ...journalBEEI
This document discusses spoken language identification using i-vectors and x-vectors for feature extraction, and PLDA and logistic regression for classification. It examines extracting features from Javanese, Sundanese, and Minangkabau languages, then classifying the languages using various parameters. The study finds that x-vector outperforms i-vector when using PLDA classification, except when using logistic regression, where i-vector performs better. It tunes parameters for i-vector UBM size, i-vector dimension, x-vector max frame size, and num repeats, reporting equal error rates to evaluate performance on test segments of 3, 10 and 30 seconds.
Design and Development of a 2D-Convolution CNN model for Recognition of Handw...CSCJournals
Owing to the innumerable appearances due to different writers, their writing styles, technical environment differences and noise, the handwritten character recognition has always been one of the most challenging task in pattern recognition. The emergence of deep learning has provided a new direction to break the limits of decades old traditional methods. There exist many scripts in the world which are being used by millions of people. Handwritten character recognition studies of several of these scripts are found in the literature. Different hand-crafted feature sets have been used in these recognition studies. Feature based approaches derive important properties from the test patterns and employ them in a more sophisticated classification model. Feature extraction using Zernike moment and Polar harmonic transformation techniques was also performed and a moderate classification accuracy was also achieved. The problems faced while using these techniques led us to use CNN based recognition approach which is capable of learning the feature vector from the training character image samples in an unsupervised manner in the sense that no hand-crafting is employed to determine the feature vector. This paper presents a deep learning paradigm using a Convolution Neural Network (CNN) which is implemented for handwritten Gurumukhi and devanagari character recognition (HGDCR). In the present experiment, the training of a 34-layer CNN for a 35 class self-generated handwritten Gurumukhi and 60 class (50 alphabet and 10 digits) handwritten Devanagari character dataset was performed on a GPU (Graphic Processing Unit) machine. The experiment resulted with an average recognition accuracy of more than 92% in case of Handwritten Gurumukhi Character dataset and 97.25% in case of Handwritten Devanagari Character dataset. It was also concluded that the training and classification through our network design performed about 10 times faster than on a moderately fast CPU. The advantage of this framework is proved by the experimental results.
Handwriting character recognition (HCR) is the ability of a computer to receive and interpret handwritten input. Handwritten Character Recognition is one of the active and challenging research areas in the field of Pattern Recognition. Pattern recognition is a process that taking in raw data and making an action based on the category of the pattern. HCR is one of the well-known applications of pattern recognition. Handwriting recognition especially for Indian languages is still in infant stage because not much work has been done it. This paper discuss about an idea to recognize Kannada vowels using chain code features. Kannada is a South Indian language. For any recognition system, an important part is feature extraction. A proper feature extraction method can increase the recognition ratio. In this paper, a chain code based feature extraction method is investigated for developing HCR system. Chain code is working based on 4-neighborhood or 8–neighborhood methods. Chain code is a sequence of code directions of a character and connection to a starting point which is often used in image processing. In this paper, 8–neighborhood method has been implemented which allows generation of eight different codes for each character. These codes have been used as features of the character image, which have been later on used for training and testing for K-Nearest Neighbor (KNN) classifiers. The level of accuracy reached to 100%.
Dimension Reduction for Script Classification - Printed Indian Documentsijait
Automatic identification of a script in a given document image facilitates many important applications such as automatic archiving of multilingual documents, searching online archives of document images and for the selection of script specific OCR in a multilingual environment. This paper provides a comparison study of three dimension reduction techniques, namely partial least squares (PLS), sliced inverse regression (SIR)
and principal component analysis (PCA), and evaluates the relative performance of classification procedures incorporating those methods. For given script we extracted different features like Gray Level Co-occurrence Method (GLCM) and Scale invariant feature transform (SIFT) features. The features are
extracted globally from a given text block which does not require any complex and reliable segmentation of the document image into lines and characters. Extracted features are reduced using various dimension reduction techniques. The reduced features are fed into Nearest Neighbor classifier. Thus the proposed
scheme is efficient and can be used for many practical
pplications which require processing large volumes
of data. The scheme has been tested on 10 Indian scripts and found to be robust in the process of scanning and relatively insensitive to change in font size. This proposed system achieves good classification accuracy on a large testing data set.
Bangla handwritten character recognition using MobileNet V1 architecturejournalBEEI
Handwritten character recognition is a very tough task in case of complex shaped alphabet set like Bangla script. As optical character recognition (OCR) has a huge application in mobile devices, model needs to be suitable for mobile applications. Many researches have been performed in this arena but none of them achieved satisfactory accuracy or could not detect more than 200 characters. MobileNet is a state of art (convolutional neural network) CNN architecture which is designed for mobile devices as it requires less computing power. In this paper, we used MobileNet for handwritten character recognition. It has achieved 96.46% accuracy in recognizing 231 classes (171 compound, 50 basic and 10 numerals), 96.17% accuracy in 171 compound character classes, 98.37% accuracy in 50 basic character classes and 99.56% accuracy in 10 numeral character classes.
Character recognition of Devanagari characters using Artificial Neural Networkijceronline
This document summarizes a paper on character recognition of Devanagari script using artificial neural networks. It discusses the challenges in recognizing handwritten Devanagari characters due to variations in writing style and speed. It presents the methodology used which includes data acquisition, preprocessing, segmentation, feature extraction and classification. In feature extraction, techniques like Gabor transform are used to extract features. A probabilistic neural network is then used for classification. The methodology achieved a recognition rate of 80% for handwritten Devanagari characters.
Automated Bangla sign language translation system for alphabets by means of M...TELKOMNIKA JOURNAL
Individuals with hearing and speaking impairment communicate using sign language. The movement of hand, body and expressions of face are the means by which the people, who are unable to hear and speak, can communicate. Bangla sign alphabets are formed with one or two hand movements. There are some features which differentiates the signs. To detect and recognize the signs, analyzing its shape and comparing its features is necessary. This paper aims to propose a model and build a computer systemthat can recognize Bangla Sign Lanugage alphabets and translate them to corresponding Bangla letters by means of deep convolutional neural network (CNN). CNN has been introduced in this model in form of a pre-trained model called “MobileNet” which produced an average accuracy of 95.71% in recognizing 36 Bangla Sign Language alphabets.
An Optical Character Recognition for Handwritten Devanagari ScriptIJERA Editor
Optical Character Recognition is process of recognition of character from scanned document and lots of OCR now available in the market. But most of these systems work for Roman, Chinese, Japanese and Arabic characters . There are no sufficient number of work on Indian language script like Devanagari so this paper present a review on optical character recognition on handwritten Devanagari script
Spoken language identification using i-vectors, x-vectors, PLDA and logistic ...journalBEEI
This document discusses spoken language identification using i-vectors and x-vectors for feature extraction, and PLDA and logistic regression for classification. It examines extracting features from Javanese, Sundanese, and Minangkabau languages, then classifying the languages using various parameters. The study finds that x-vector outperforms i-vector when using PLDA classification, except when using logistic regression, where i-vector performs better. It tunes parameters for i-vector UBM size, i-vector dimension, x-vector max frame size, and num repeats, reporting equal error rates to evaluate performance on test segments of 3, 10 and 30 seconds.
Design and Development of a 2D-Convolution CNN model for Recognition of Handw...CSCJournals
Owing to the innumerable appearances due to different writers, their writing styles, technical environment differences and noise, the handwritten character recognition has always been one of the most challenging task in pattern recognition. The emergence of deep learning has provided a new direction to break the limits of decades old traditional methods. There exist many scripts in the world which are being used by millions of people. Handwritten character recognition studies of several of these scripts are found in the literature. Different hand-crafted feature sets have been used in these recognition studies. Feature based approaches derive important properties from the test patterns and employ them in a more sophisticated classification model. Feature extraction using Zernike moment and Polar harmonic transformation techniques was also performed and a moderate classification accuracy was also achieved. The problems faced while using these techniques led us to use CNN based recognition approach which is capable of learning the feature vector from the training character image samples in an unsupervised manner in the sense that no hand-crafting is employed to determine the feature vector. This paper presents a deep learning paradigm using a Convolution Neural Network (CNN) which is implemented for handwritten Gurumukhi and devanagari character recognition (HGDCR). In the present experiment, the training of a 34-layer CNN for a 35 class self-generated handwritten Gurumukhi and 60 class (50 alphabet and 10 digits) handwritten Devanagari character dataset was performed on a GPU (Graphic Processing Unit) machine. The experiment resulted with an average recognition accuracy of more than 92% in case of Handwritten Gurumukhi Character dataset and 97.25% in case of Handwritten Devanagari Character dataset. It was also concluded that the training and classification through our network design performed about 10 times faster than on a moderately fast CPU. The advantage of this framework is proved by the experimental results.
Handwriting character recognition (HCR) is the ability of a computer to receive and interpret handwritten input. Handwritten Character Recognition is one of the active and challenging research areas in the field of Pattern Recognition. Pattern recognition is a process that taking in raw data and making an action based on the category of the pattern. HCR is one of the well-known applications of pattern recognition. Handwriting recognition especially for Indian languages is still in infant stage because not much work has been done it. This paper discuss about an idea to recognize Kannada vowels using chain code features. Kannada is a South Indian language. For any recognition system, an important part is feature extraction. A proper feature extraction method can increase the recognition ratio. In this paper, a chain code based feature extraction method is investigated for developing HCR system. Chain code is working based on 4-neighborhood or 8–neighborhood methods. Chain code is a sequence of code directions of a character and connection to a starting point which is often used in image processing. In this paper, 8–neighborhood method has been implemented which allows generation of eight different codes for each character. These codes have been used as features of the character image, which have been later on used for training and testing for K-Nearest Neighbor (KNN) classifiers. The level of accuracy reached to 100%.
Dimension Reduction for Script Classification - Printed Indian Documentsijait
Automatic identification of a script in a given document image facilitates many important applications such as automatic archiving of multilingual documents, searching online archives of document images and for the selection of script specific OCR in a multilingual environment. This paper provides a comparison study of three dimension reduction techniques, namely partial least squares (PLS), sliced inverse regression (SIR)
and principal component analysis (PCA), and evaluates the relative performance of classification procedures incorporating those methods. For given script we extracted different features like Gray Level Co-occurrence Method (GLCM) and Scale invariant feature transform (SIFT) features. The features are
extracted globally from a given text block which does not require any complex and reliable segmentation of the document image into lines and characters. Extracted features are reduced using various dimension reduction techniques. The reduced features are fed into Nearest Neighbor classifier. Thus the proposed
scheme is efficient and can be used for many practical
pplications which require processing large volumes
of data. The scheme has been tested on 10 Indian scripts and found to be robust in the process of scanning and relatively insensitive to change in font size. This proposed system achieves good classification accuracy on a large testing data set.
Bangla handwritten character recognition using MobileNet V1 architecturejournalBEEI
Handwritten character recognition is a very tough task in case of complex shaped alphabet set like Bangla script. As optical character recognition (OCR) has a huge application in mobile devices, model needs to be suitable for mobile applications. Many researches have been performed in this arena but none of them achieved satisfactory accuracy or could not detect more than 200 characters. MobileNet is a state of art (convolutional neural network) CNN architecture which is designed for mobile devices as it requires less computing power. In this paper, we used MobileNet for handwritten character recognition. It has achieved 96.46% accuracy in recognizing 231 classes (171 compound, 50 basic and 10 numerals), 96.17% accuracy in 171 compound character classes, 98.37% accuracy in 50 basic character classes and 99.56% accuracy in 10 numeral character classes.
Character recognition of Devanagari characters using Artificial Neural Networkijceronline
This document summarizes a paper on character recognition of Devanagari script using artificial neural networks. It discusses the challenges in recognizing handwritten Devanagari characters due to variations in writing style and speed. It presents the methodology used which includes data acquisition, preprocessing, segmentation, feature extraction and classification. In feature extraction, techniques like Gabor transform are used to extract features. A probabilistic neural network is then used for classification. The methodology achieved a recognition rate of 80% for handwritten Devanagari characters.
IRJET- Sign Language Interpreter using Image Processing and Machine LearningIRJET Journal
This document describes a system to translate sign language gestures to text or audio using image processing and machine learning techniques. The system takes an image of a sign language gesture as input using a webcam. It then performs preprocessing steps like skin detection and edge detection. Features are extracted from the preprocessed image using a Histogram of Oriented Gradients algorithm. These features are fed into a Support Vector Machine classifier that has been trained on a dataset of 6000 images of English alphabet signs. The system is able to recognize the signs with 88% accuracy and translate them to text or audio output, aiding communication for deaf individuals.
International Journal of Research in Engineering and Science is an open access peer-reviewed international forum for scientists involved in research to publish quality and refereed papers. Papers reporting original research or experimentally proved review work are welcome. Papers for publication are selected through peer review to ensure originality, relevance, and readability.
Artificial Neural Network For Recognition Of Handwritten Devanagari CharacterIOSR Journals
1) The document discusses recognizing handwritten Devanagari characters using artificial neural networks and zone-based feature extraction.
2) It proposes extracting features from images by dividing them into zones and calculating average pixel distances to the image and zone centroids.
3) This zone-based feature vector is then input to a feedforward neural network for character recognition.
The presentation will describe an algorithm through which one can recognize Devanagari Characters. Devanagari is the script in which Hindi is represented. This algorithm
could automatically segment character from the image of Devenagari text and then recognize them.
For extracting the individual characters from the image of Devanagari text, algorithm segmented the image several
times using the vertical and horizontal projection.
The algorithm starts with first segmenting the lines separately from the document by taking horizontal projection and then the line
into words by taking vertical projection of the line. Another step which is particular to the separation of
Devanagari characters was required and was done by first removing the header line by finding horizontal projection
of each word. The characters can then be extracted by vertical projection of the word without the header line.
Algorithm uses a Kohonen Neural Netowrk for the recognition task. After the separation of the characters from the
image, the image matrix was then downsampled to bring it down to a fixed size so as to make the recognition
size independent. The matrix can then be fed as input neurons to the Kohonen Neural Network and the winning neuron is
found which identifies the recognized the character. This information in Kohonen Neural Network was stored
earlier during the training phase of the neural network. For this, we first assigned random weights from input neurons
to output neurons and then for each training set, the winning neuron was calculated by finding the maximum
output produced by the neurons. The wights for this winning neuron were then adjusted so that it responds to this
pattern more strongly the next time.
A survey on Script and Language identification for Handwritten document imagesiosrjce
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
PERFORMANCE EVALUATION OF STATISTICAL CLASSIFIERS USING INDIAN SIGN LANGUAGE ...IJCSEA Journal
Sign language is the key for communication between deaf people. The significance of sign language is accentuated by various research activities and the technical aspects will definitely improve the communication needs. General view based sign language recognition systems extract manual parameters by a single camera view because it seems to be user friendly and hardware complexity; however it needs a high accuracy classifier for classification and recognition purpose. The decision making of the system in this work employs Indian sign language datasets and the performance evaluation of the system is compared by deploying the K-NN, Naïve Bayes and PNN classifiers. Classification using an instance-based classifiercan be a simple matter of locating the instance space and labelling the unknown instance with the same class label as that of the located (known) neighbour. Classifier always tries to improve the classification rate by pushing classifiers into an optimised structure. In each hand posture, a measure of properties like area, mean intensity, centroid, perimeter and diameter are taken; the classifier then uses these properties to determine the sign in different angles. They estimate the probability that a sign belongs to each of the target classes that is fixed. The impact of such study may reflect the exploration for using such algorithms
in other similar applications such as text classification and the development of automated systems.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
IRJET - A Survey on Recognition of Strike-Out Texts in Handwritten DocumentsIRJET Journal
This document summarizes research on recognizing strike-out or crossed-out text in handwritten documents. It discusses 4 papers that studied this problem in different languages like English, Bengali, and Devanagari. The first paper used an LSTM model to recognize struck-out English text with 11% character error rate. The second used HMMs to recognize French text crossed out with lines or waves, achieving 46-92% accuracy. The third used graph algorithms and SVM to detect and remove Bengali strike-outs, obtaining 95% accuracy. The fourth used decision trees to classify English forensic documents, identifying 48% of crossed texts correctly. Overall the document reviews approaches to strike-out text recognition and removal.
A bidirectional text transcription of braille for odia, hindi, telugu and eng...eSAT Journals
This document describes a bidirectional text transcription system for converting Braille documents in Odia, Hindi, Telugu, and English to text and vice versa using image processing techniques on an FPGA. It discusses prior work on Braille recognition and text-to-speech systems. The proposed algorithm segments Braille cells from image inputs and uses a modified database to map dot patterns to letters/words in each language based on number of dots. Letter patterns are rearranged for Hindi, Telugu, and English. The database is tested and dumped onto an FPGA for hardware implementation and bidirectional conversion between Braille and text in multiple languages.
Haralick Texture Features based Syriac(Assyrian) and English or Arabic docume...Editor IJCATR
This document presents a method for classifying Syriac, English, and Arabic script documents using Haralick texture features and a k-nearest neighbor algorithm. 300 text blocks were extracted from documents in the three scripts and rotated between 0-135 degrees. 13 Haralick texture features were extracted from each block to form vectors for classification. Using a kNN classifier with k=3, the method achieved 100% accuracy in classifying Syriac vs. English text blocks and Syriac vs. Arabic text blocks, even when the scripts were rotated, demonstrating the potential of Haralick texture features for script identification.
BrailleOCR: An Open Source Document to Braille Converter Applicationpijush15
This presentation is actually about an Open Source application, BrailleOCR that helps to convert scanned documents to Braille and thus helps the Visually Impaired.
What is the use of this application in real life? Well, BrailleOCR is currently the only app that integrated Optical character recognition and Braille Translation together. This app will eventually help converting a lot of important documents to Braille. The project site for this project is given here
IJCA Paper: http://www.ijcaonline.org/archives/volume68/number16/11664-7254
Project site: https://code.google.com/p/brailleocr/
The app uses a four step process. Initially, we have a scanned image, which is a RGB image. The first step or the Pre-Processing step deals with conversion of a RGB image to grayscale. The 2nd step deals with Character Recognition using the Tesseract Engine. Now, the recognition step may have errors and we require post processing to correct them. The 3rd step is thus the Post-Processing step and it actually corrects errors in the previous step. The final and the most important step is the Braille Conversion step.
Wavelet Packet Based Features for Automatic Script IdentificationCSCJournals
In a multi script environment, an archive of documents having the text regions printed in different scripts is in practice. For automatic processing of such documents through Optical Character Recognition (OCR), it is necessary to identify different script regions of the document. In this paper, a novel texture-based approach is presented to identify the script type of the collection of documents printed in seven scripts, to categorize them for further processing. The South Indian documents printed in the seven scripts - Kannada, Tamil, Telugu, Malayalam, Urdu, Hindi and English are considered here. The document images are decomposed through the Wavelet Packet Decomposition using the Haar basis function up to level two. The texture features are extracted from the sub bands of the wavelet packet decomposition. The Shannon entropy value is computed for the set of sub bands and these entropy values are combined to use as the texture features. Experimentation conducted involved 2100 text images for learning and 1400 text images for testing. Script classification performance is analyzed using the K-nearest neighbor classifier. The average success rate is found to be 99.68%.
Script Identification for printed document images at text-line level using DC...IOSR Journals
This document summarizes a research paper that proposes a script identification approach for Indian scripts at the text-line level of printed documents. The approach uses visual appearance-based recognition by extracting features from text lines using Discrete Cosine Transform (DCT) and Principal Component Analysis (PCA). These features are classified using a Modified K-Nearest Neighbor algorithm. The method achieves 95% accuracy in recognizing 11 major Indian languages and discusses the importance of script identification in multilingual environments like India for applications such as optical character recognition.
A Comprehensive Study On Handwritten Character Recognition Systemiosrjce
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
This document is a project report submitted by Mohammad Saiful Islam for a CMPUT 551 course on December 21st, 2010 regarding Bengali handwritten digit recognition using support vector machines. The report discusses building a dataset of Bengali digits written by the author, preprocessing and feature extraction steps, and using a multiclass support vector machine with different kernels for classification. The author hypothesizes that SVM will perform well, RBF kernels will improve performance over linear and polynomial kernels, and using raw pixel values can achieve good accuracy, though testing on different writers may reduce performance. Experiments are planned to test these hypotheses using the collected dataset.
Pattern Recognition of Japanese Alphabet Katakana Using Airy Zeta FunctionEditor IJCATR
Character recognition is one of common pattern recognition study. There are many object used in pattern recognition, such
as Japanese alphabet character, which is a very complex character compared to common Roman character. This research focus on
pattern recognition of Japanese character handwriting, Katakana. The pattern recognition process of a letter of the alphabet uses Airy
Zeta Function, with its input file is a .bmp file. User can write directly on an input device of the system. The testing of the system
examines 460 letter characters. The first testing that examines 230 characters result in an accuracy of 55,65%, whilst the second testing
that examines 460 characters produces an accuracy of 64,56% in recognizing the letters. These accuracy are much determined by the
quantity of training. The approach of pattern recognition is a statistical approach, where more pattern of letters are trained and saved as
a reference, more intelligent the system . The implementation of Airy zeta function methods in recognizing Japanese letter pattern is
able to produce high accuracy level.
Optimal Clustering Technique for Handwritten Nandinagari Character RecognitionEditor IJCATR
In this paper, an optimalclustering technique for handwritten Nandinagari character recognition is proposed. We compare two
different corner detector mechanisms and compare and contrast various clustering approachesfor handwritten Nandinagari characters.
In this model, the key interest points on the images which are invariant to Scale, rotation, translation, illumination and occlusion are
identified by choosing robust Scale Invariant Feature Transform method(SIFT) and Speeded Up Robust Feature (SURF) transform
techniques. We then generate a dissimilarity matrix, which is in turn fed as an input for a set of clustering techniques like K Means,
PAM (Partition Around Medoids) and Hierarchical Agglomerative clustering. Various cluster validity measures are used to assess the
quality of clustering techniques with an intent to find a technique suitable for these rare characters. On a varied data set of over 1040
Handwritten Nandinagari characters, a careful analysis indicate this combinatorial approach used in a collaborative manner will aid in
achieving good recognition accuracy. We found that Hierarchical clustering technique is most suitable for SIFT and SURF features as
compared to K Means and PAM techniques.
Performance Comparison between Different Feature Extraction Techniques with S...IJERA Editor
This document summarizes research on recognizing handwritten characters in the Gurumukhi script using different feature extraction techniques and a support vector machine (SVM) classifier. It describes evaluating distance profiles, diagonal features, and background directional distribution for feature extraction. Diagonal features achieved the highest recognition accuracy of 95.39% compared to other methods. The document also reviews related work on Gurumukhi and other language recognition and discusses the proposed methodology in more detail.
IRJET- Review on Optical Character RecognitionIRJET Journal
This document provides a review of optical character recognition (OCR) technologies. It discusses research that has been conducted on OCR for English, Arabic, and Devanagari characters. For English characters, various studies that used techniques like neural networks, support vector machines, and nearest neighbor classification are summarized. For Arabic characters, research using neural networks, Haar-like feature extraction and boosting classifiers achieved recognition rates from 87% to 99%. Studies on Devanagari characters employed techniques such as curvelet transforms, neural networks, k-means clustering and support vector machines and achieved recognition rates from 90% to 98.5%. In general, the document reviews past work on OCR and the techniques researchers have used to recognize
STRUCTURAL FEATURES FOR RECOGNITION OF HAND WRITTEN KANNADA CHARACTER BASED O...ijcseit
Research in image processing involves many active areas, of these Recognition of Handwritten character holds lots of promises and is challenging one .The idea is to enable the computer to be able to recognize intelligibly hand written inputs In this paper, a new method that uses structural features and support vector Machine (SVM) classifier for recognition of Handwritten Kannada characters is presented. On an average recognition accuracy of 89.84 % and 85.14% for handwritten Kannada vowels and Consonants obtained with this proposed method, inspite of inherent variations.
S TRUCTURAL F EATURES F OR R ECOGNITION O F H AND W RITTEN K ANNADA C ...ijcsit
Research in image processing involves many active a
reas, of these Recognition of Handwritten character
holds lots of promises and is challenging one .The
idea is to enable the computer to be able to recogn
ize
intelligibly hand written inputs In this paper, a
new method that uses structural features and suppo
rt
vector Machine (SVM) classifier for recognition of
Handwritten Kannada characters is presented. On an
average recognition accuracy of 89.84 % and 85.14%
for handwritten Kannada vowels and Consonants
obtained with this proposed method, inspite of inhe
rent variations
STRUCTURAL FEATURES FOR RECOGNITION OF HAND WRITTEN KANNADA CHARACTER BASED O...ijcseit
Research in image processing involves many active areas, of these Recognition of Handwritten character
holds lots of promises and is challenging one .The idea is to enable the computer to be able to recognize
intelligibly hand written inputs In this paper, a new method that uses structural features and support
vector Machine (SVM) classifier for recognition of Handwritten Kannada characters is presented. On an
average recognition accuracy of 89.84 % and 85.14% for handwritten Kannada vowels and Consonants
obtained with this proposed method, inspite of inherent variations.
IRJET- Sign Language Interpreter using Image Processing and Machine LearningIRJET Journal
This document describes a system to translate sign language gestures to text or audio using image processing and machine learning techniques. The system takes an image of a sign language gesture as input using a webcam. It then performs preprocessing steps like skin detection and edge detection. Features are extracted from the preprocessed image using a Histogram of Oriented Gradients algorithm. These features are fed into a Support Vector Machine classifier that has been trained on a dataset of 6000 images of English alphabet signs. The system is able to recognize the signs with 88% accuracy and translate them to text or audio output, aiding communication for deaf individuals.
International Journal of Research in Engineering and Science is an open access peer-reviewed international forum for scientists involved in research to publish quality and refereed papers. Papers reporting original research or experimentally proved review work are welcome. Papers for publication are selected through peer review to ensure originality, relevance, and readability.
Artificial Neural Network For Recognition Of Handwritten Devanagari CharacterIOSR Journals
1) The document discusses recognizing handwritten Devanagari characters using artificial neural networks and zone-based feature extraction.
2) It proposes extracting features from images by dividing them into zones and calculating average pixel distances to the image and zone centroids.
3) This zone-based feature vector is then input to a feedforward neural network for character recognition.
The presentation will describe an algorithm through which one can recognize Devanagari Characters. Devanagari is the script in which Hindi is represented. This algorithm
could automatically segment character from the image of Devenagari text and then recognize them.
For extracting the individual characters from the image of Devanagari text, algorithm segmented the image several
times using the vertical and horizontal projection.
The algorithm starts with first segmenting the lines separately from the document by taking horizontal projection and then the line
into words by taking vertical projection of the line. Another step which is particular to the separation of
Devanagari characters was required and was done by first removing the header line by finding horizontal projection
of each word. The characters can then be extracted by vertical projection of the word without the header line.
Algorithm uses a Kohonen Neural Netowrk for the recognition task. After the separation of the characters from the
image, the image matrix was then downsampled to bring it down to a fixed size so as to make the recognition
size independent. The matrix can then be fed as input neurons to the Kohonen Neural Network and the winning neuron is
found which identifies the recognized the character. This information in Kohonen Neural Network was stored
earlier during the training phase of the neural network. For this, we first assigned random weights from input neurons
to output neurons and then for each training set, the winning neuron was calculated by finding the maximum
output produced by the neurons. The wights for this winning neuron were then adjusted so that it responds to this
pattern more strongly the next time.
A survey on Script and Language identification for Handwritten document imagesiosrjce
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
PERFORMANCE EVALUATION OF STATISTICAL CLASSIFIERS USING INDIAN SIGN LANGUAGE ...IJCSEA Journal
Sign language is the key for communication between deaf people. The significance of sign language is accentuated by various research activities and the technical aspects will definitely improve the communication needs. General view based sign language recognition systems extract manual parameters by a single camera view because it seems to be user friendly and hardware complexity; however it needs a high accuracy classifier for classification and recognition purpose. The decision making of the system in this work employs Indian sign language datasets and the performance evaluation of the system is compared by deploying the K-NN, Naïve Bayes and PNN classifiers. Classification using an instance-based classifiercan be a simple matter of locating the instance space and labelling the unknown instance with the same class label as that of the located (known) neighbour. Classifier always tries to improve the classification rate by pushing classifiers into an optimised structure. In each hand posture, a measure of properties like area, mean intensity, centroid, perimeter and diameter are taken; the classifier then uses these properties to determine the sign in different angles. They estimate the probability that a sign belongs to each of the target classes that is fixed. The impact of such study may reflect the exploration for using such algorithms
in other similar applications such as text classification and the development of automated systems.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
IRJET - A Survey on Recognition of Strike-Out Texts in Handwritten DocumentsIRJET Journal
This document summarizes research on recognizing strike-out or crossed-out text in handwritten documents. It discusses 4 papers that studied this problem in different languages like English, Bengali, and Devanagari. The first paper used an LSTM model to recognize struck-out English text with 11% character error rate. The second used HMMs to recognize French text crossed out with lines or waves, achieving 46-92% accuracy. The third used graph algorithms and SVM to detect and remove Bengali strike-outs, obtaining 95% accuracy. The fourth used decision trees to classify English forensic documents, identifying 48% of crossed texts correctly. Overall the document reviews approaches to strike-out text recognition and removal.
A bidirectional text transcription of braille for odia, hindi, telugu and eng...eSAT Journals
This document describes a bidirectional text transcription system for converting Braille documents in Odia, Hindi, Telugu, and English to text and vice versa using image processing techniques on an FPGA. It discusses prior work on Braille recognition and text-to-speech systems. The proposed algorithm segments Braille cells from image inputs and uses a modified database to map dot patterns to letters/words in each language based on number of dots. Letter patterns are rearranged for Hindi, Telugu, and English. The database is tested and dumped onto an FPGA for hardware implementation and bidirectional conversion between Braille and text in multiple languages.
Haralick Texture Features based Syriac(Assyrian) and English or Arabic docume...Editor IJCATR
This document presents a method for classifying Syriac, English, and Arabic script documents using Haralick texture features and a k-nearest neighbor algorithm. 300 text blocks were extracted from documents in the three scripts and rotated between 0-135 degrees. 13 Haralick texture features were extracted from each block to form vectors for classification. Using a kNN classifier with k=3, the method achieved 100% accuracy in classifying Syriac vs. English text blocks and Syriac vs. Arabic text blocks, even when the scripts were rotated, demonstrating the potential of Haralick texture features for script identification.
BrailleOCR: An Open Source Document to Braille Converter Applicationpijush15
This presentation is actually about an Open Source application, BrailleOCR that helps to convert scanned documents to Braille and thus helps the Visually Impaired.
What is the use of this application in real life? Well, BrailleOCR is currently the only app that integrated Optical character recognition and Braille Translation together. This app will eventually help converting a lot of important documents to Braille. The project site for this project is given here
IJCA Paper: http://www.ijcaonline.org/archives/volume68/number16/11664-7254
Project site: https://code.google.com/p/brailleocr/
The app uses a four step process. Initially, we have a scanned image, which is a RGB image. The first step or the Pre-Processing step deals with conversion of a RGB image to grayscale. The 2nd step deals with Character Recognition using the Tesseract Engine. Now, the recognition step may have errors and we require post processing to correct them. The 3rd step is thus the Post-Processing step and it actually corrects errors in the previous step. The final and the most important step is the Braille Conversion step.
Wavelet Packet Based Features for Automatic Script IdentificationCSCJournals
In a multi script environment, an archive of documents having the text regions printed in different scripts is in practice. For automatic processing of such documents through Optical Character Recognition (OCR), it is necessary to identify different script regions of the document. In this paper, a novel texture-based approach is presented to identify the script type of the collection of documents printed in seven scripts, to categorize them for further processing. The South Indian documents printed in the seven scripts - Kannada, Tamil, Telugu, Malayalam, Urdu, Hindi and English are considered here. The document images are decomposed through the Wavelet Packet Decomposition using the Haar basis function up to level two. The texture features are extracted from the sub bands of the wavelet packet decomposition. The Shannon entropy value is computed for the set of sub bands and these entropy values are combined to use as the texture features. Experimentation conducted involved 2100 text images for learning and 1400 text images for testing. Script classification performance is analyzed using the K-nearest neighbor classifier. The average success rate is found to be 99.68%.
Script Identification for printed document images at text-line level using DC...IOSR Journals
This document summarizes a research paper that proposes a script identification approach for Indian scripts at the text-line level of printed documents. The approach uses visual appearance-based recognition by extracting features from text lines using Discrete Cosine Transform (DCT) and Principal Component Analysis (PCA). These features are classified using a Modified K-Nearest Neighbor algorithm. The method achieves 95% accuracy in recognizing 11 major Indian languages and discusses the importance of script identification in multilingual environments like India for applications such as optical character recognition.
A Comprehensive Study On Handwritten Character Recognition Systemiosrjce
IOSR Journal of Computer Engineering (IOSR-JCE) is a double blind peer reviewed International Journal that provides rapid publication (within a month) of articles in all areas of computer engineering and its applications. The journal welcomes publications of high quality papers on theoretical developments and practical applications in computer technology. Original research papers, state-of-the-art reviews, and high quality technical notes are invited for publications.
This document is a project report submitted by Mohammad Saiful Islam for a CMPUT 551 course on December 21st, 2010 regarding Bengali handwritten digit recognition using support vector machines. The report discusses building a dataset of Bengali digits written by the author, preprocessing and feature extraction steps, and using a multiclass support vector machine with different kernels for classification. The author hypothesizes that SVM will perform well, RBF kernels will improve performance over linear and polynomial kernels, and using raw pixel values can achieve good accuracy, though testing on different writers may reduce performance. Experiments are planned to test these hypotheses using the collected dataset.
Pattern Recognition of Japanese Alphabet Katakana Using Airy Zeta FunctionEditor IJCATR
Character recognition is one of common pattern recognition study. There are many object used in pattern recognition, such
as Japanese alphabet character, which is a very complex character compared to common Roman character. This research focus on
pattern recognition of Japanese character handwriting, Katakana. The pattern recognition process of a letter of the alphabet uses Airy
Zeta Function, with its input file is a .bmp file. User can write directly on an input device of the system. The testing of the system
examines 460 letter characters. The first testing that examines 230 characters result in an accuracy of 55,65%, whilst the second testing
that examines 460 characters produces an accuracy of 64,56% in recognizing the letters. These accuracy are much determined by the
quantity of training. The approach of pattern recognition is a statistical approach, where more pattern of letters are trained and saved as
a reference, more intelligent the system . The implementation of Airy zeta function methods in recognizing Japanese letter pattern is
able to produce high accuracy level.
Optimal Clustering Technique for Handwritten Nandinagari Character RecognitionEditor IJCATR
In this paper, an optimalclustering technique for handwritten Nandinagari character recognition is proposed. We compare two
different corner detector mechanisms and compare and contrast various clustering approachesfor handwritten Nandinagari characters.
In this model, the key interest points on the images which are invariant to Scale, rotation, translation, illumination and occlusion are
identified by choosing robust Scale Invariant Feature Transform method(SIFT) and Speeded Up Robust Feature (SURF) transform
techniques. We then generate a dissimilarity matrix, which is in turn fed as an input for a set of clustering techniques like K Means,
PAM (Partition Around Medoids) and Hierarchical Agglomerative clustering. Various cluster validity measures are used to assess the
quality of clustering techniques with an intent to find a technique suitable for these rare characters. On a varied data set of over 1040
Handwritten Nandinagari characters, a careful analysis indicate this combinatorial approach used in a collaborative manner will aid in
achieving good recognition accuracy. We found that Hierarchical clustering technique is most suitable for SIFT and SURF features as
compared to K Means and PAM techniques.
Performance Comparison between Different Feature Extraction Techniques with S...IJERA Editor
This document summarizes research on recognizing handwritten characters in the Gurumukhi script using different feature extraction techniques and a support vector machine (SVM) classifier. It describes evaluating distance profiles, diagonal features, and background directional distribution for feature extraction. Diagonal features achieved the highest recognition accuracy of 95.39% compared to other methods. The document also reviews related work on Gurumukhi and other language recognition and discusses the proposed methodology in more detail.
IRJET- Review on Optical Character RecognitionIRJET Journal
This document provides a review of optical character recognition (OCR) technologies. It discusses research that has been conducted on OCR for English, Arabic, and Devanagari characters. For English characters, various studies that used techniques like neural networks, support vector machines, and nearest neighbor classification are summarized. For Arabic characters, research using neural networks, Haar-like feature extraction and boosting classifiers achieved recognition rates from 87% to 99%. Studies on Devanagari characters employed techniques such as curvelet transforms, neural networks, k-means clustering and support vector machines and achieved recognition rates from 90% to 98.5%. In general, the document reviews past work on OCR and the techniques researchers have used to recognize
STRUCTURAL FEATURES FOR RECOGNITION OF HAND WRITTEN KANNADA CHARACTER BASED O...ijcseit
Research in image processing involves many active areas, of these Recognition of Handwritten character holds lots of promises and is challenging one .The idea is to enable the computer to be able to recognize intelligibly hand written inputs In this paper, a new method that uses structural features and support vector Machine (SVM) classifier for recognition of Handwritten Kannada characters is presented. On an average recognition accuracy of 89.84 % and 85.14% for handwritten Kannada vowels and Consonants obtained with this proposed method, inspite of inherent variations.
S TRUCTURAL F EATURES F OR R ECOGNITION O F H AND W RITTEN K ANNADA C ...ijcsit
Research in image processing involves many active a
reas, of these Recognition of Handwritten character
holds lots of promises and is challenging one .The
idea is to enable the computer to be able to recogn
ize
intelligibly hand written inputs In this paper, a
new method that uses structural features and suppo
rt
vector Machine (SVM) classifier for recognition of
Handwritten Kannada characters is presented. On an
average recognition accuracy of 89.84 % and 85.14%
for handwritten Kannada vowels and Consonants
obtained with this proposed method, inspite of inhe
rent variations
STRUCTURAL FEATURES FOR RECOGNITION OF HAND WRITTEN KANNADA CHARACTER BASED O...ijcseit
Research in image processing involves many active areas, of these Recognition of Handwritten character
holds lots of promises and is challenging one .The idea is to enable the computer to be able to recognize
intelligibly hand written inputs In this paper, a new method that uses structural features and support
vector Machine (SVM) classifier for recognition of Handwritten Kannada characters is presented. On an
average recognition accuracy of 89.84 % and 85.14% for handwritten Kannada vowels and Consonants
obtained with this proposed method, inspite of inherent variations.
STRUCTURAL FEATURES FOR RECOGNITION OF HAND WRITTEN KANNADA CHARACTER BASED O...ijcseit
This document presents a method for recognizing handwritten Kannada characters using structural features and a support vector machine (SVM) classifier. The method extracts structural features like perimeter, area, eccentricity from preprocessed character images. These features are used to train an SVM classifier. On average, the method achieved 89.84% recognition accuracy for handwritten Kannada vowels and 85.14% for consonants. The method works in two phases - a training phase where the SVM is trained on extracted structural features, and a testing phase where unknown characters are classified based on their structural features and the trained SVM model.
A New Method for Identification of Partially Similar Indian ScriptsCSCJournals
In this paper, the texture symmetry/non symmetry factor has been exploited to get the script texture by using the Bi Wavelants which give the factor of symmetry/non symmetry in terms of the third cumulant and the Bi-spectra gives the quadratically coupled frequencies. The envelope of Bi-spectra (Bi-Wavelant) provides an accurate behavior of the symmetry/non symmetry factor of the script texture. Classification has been better performed by SVM with training set of roots of the envelope found using the Newton-Raphson technique. The method could successfully identify 8 Indian scripts like Devanagari, Urdu, Gujrati, Telugu, Assamese, Gurmukhi, Kannada, and Bangla. The method can segment any kind of document with very good results. The identification results are excellent.
A Survey of Modern Character Recognition Techniquesijsrd.com
This document summarizes several modern techniques for handwritten character recognition. It discusses common feature extraction methods like statistical, structural and global transformation features. It then summarizes several papers that have proposed different techniques for handwritten character recognition, including using associative memory nets, moment invariants with support vector machines, neural networks, hidden markov models, gradient features, and multi-scale neural networks. The document concludes that neural networks are commonly used for training, and that feature extraction methods continue to be improved, but handwritten character recognition remains an active area of research.
Hand-written Hindi Word Recognition - A Comprehensive SurveyIRJET Journal
This document provides a comprehensive survey of techniques for handwritten Hindi word recognition. It discusses the key stages in a handwritten word recognition system, including preprocessing, segmentation, feature extraction, classification, and post-processing. It then reviews several past studies on handwritten Hindi and Devanagari character recognition, comparing their different approaches to preprocessing, feature extraction, classification, and performance results. Finally, it provides a parametric evaluation of the techniques discussed and concludes that handwritten Hindi word recognition remains an active area of research with varying approaches.
The document describes an optical character recognition (OCR) system for recognizing characters in historical records written in the Kannada language. The system uses Gabor and zonal feature extraction along with an artificial neural network for classification. It was tested on nearly 150 samples of ancient Kannada epigraphs from the Ashoka and Hoysala periods, achieving average recognition accuracies of 80.2% and 75.6% respectively. The key steps of the OCR system include preprocessing images, segmenting characters, extracting Gabor and zonal features, training an ANN classifier, and mapping recognized characters to a modern form.
DIMENSION REDUCTION FOR SCRIPT CLASSIFICATION- PRINTED INDIAN DOCUMENTSijait
Automatic identification of a script in a given document image facilitates many important applications such
as automatic archiving of multilingual documents, searching online archives of document images and for
the selection of script specific OCR in a multilingual environment. This paper provides a comparison study
of three dimension reduction techniques, namely partial least squares (PLS), sliced inverse regression (SIR)
and principal component analysis (PCA), and evaluates the relative performance of classification
procedures incorporating those methods
Survey On Broken and Joint Devanagari Handwritten Characters Recognition Usin...IRJET Journal
This document provides a survey of deep learning approaches for recognizing broken and joint handwritten Devanagari characters. It reviews architectures like CNNs, RNNs, and hybrid models applied in various studies. Datasets like DHCD and ILHCD used for training models are discussed. The performance of different approaches is compared based on metrics like accuracy. Studies using techniques like wavelet transform, transfer learning, and layer-wise training are summarized along with their strengths and limitations. The survey serves as a resource for researchers on handwritten Devanagari character recognition using deep learning. It analyzes 10 relevant publications, providing details of techniques used, datasets, accuracy achieved, and research gaps identified in each work.
1) The document discusses recognizing handwritten Devanagari characters using an artificial neural network approach.
2) It proposes extracting features from images by dividing them into zones and calculating average pixel distances to the image and zone centroids.
3) This zone-based feature extraction method is used to create a feature vector for each character image that is then classified by a feedforward neural network.
Literature Review on Indian Sign Language Recognition SystemIRJET Journal
This document discusses literature on Indian sign language recognition systems. It reviews the common approaches used in sign language recognition research, which typically involve data capture, preprocessing, segmentation, feature extraction, and classification. Several prior studies on sign language recognition systems for different languages are summarized that utilize techniques like Hidden Markov models, neural networks, Fourier transforms, and more. The key stages of a sign language recognition system are then outlined, including preprocessing, feature extraction from images/videos, and classification. Popular classification methods for sign language recognition like CNNs, KNN, SVMs are also described. In conclusion, the goal of sign language recognition systems is to identify the language of deaf individuals but current methods need to improve robustness and speed.
Writer Identification via CNN Features and SVMIRJET Journal
This document summarizes a research paper that proposes a model for identifying the author of Arabic handwritten documents based on their handwriting style. The model uses a convolutional neural network to extract features from images of handwritten text that have been augmented through data preprocessing techniques. A support vector machine is then used to classify the extracted features and identify the writer. The proposed method was tested on a dataset of 202 Arabic writers, achieving a classification accuracy of 97.20%.
This document summarizes a research paper that proposes a technique for Gujarati handwritten character recognition using radial histogram feature extraction and Euclidean distance classification. The technique extracts 72 feature vectors from a character image by counting black pixels in radial directions at 5 degree intervals to create a radial histogram. Characters are classified by calculating the Euclidean distance between their feature vectors and pre-defined vectors for each character. The method achieves 26.86% accuracy on a database of Gujarati characters. While easy to implement, the radial histogram approach provides low accuracy due to similarities between characters and variability in handwriting styles.
Optical character recognition is one of the emerging research topics in the field of image processing, and it has extensive area of application in pattern recognition. Odia handwritten script is the most research concern area because it has eldest and most likable language in the state of odisha, India. Odia character is a usually handwritten, which was generally occupied by scanner into machine readable form. In this regard several recognition technique have been evolved for variance kind of languages but writing pattern of odia character is just like as curve appearance; Hence it is more difficult for recognition. In this article we have presented the novel approach for Odia character recognition based on the different angle based symmetric axis feature extraction technique which gives high accuracy of recognition pattern. This empirical model generates a unique angle based boundary points on every skeletonised character images. These points are interconnected with each other in order to extract row and column symmetry axis. We extracted feature matrix having mean distance of row, mean angle of row, mean distance of column and mean angle of column from centre of the image to midpoint of the symmetric axis respectively. The system uses a 10 fold validation to the random forest (RF) classifier and SVM for feature matrix. We have considered the standard database on 200 images having each of 47 Odia character and 10 Odia numeric for simulation. As we have noted outcome of simulation of SVM and RF yields 96.3% and 98.2% accuracy rate on NIT Rourkela Odia character database and 88.9% and 93.6% from ISI Kolkata Odia numerical database.
PERFORMANCE EVALUATION OF FUZZY LOGIC AND BACK PROPAGATION NEURAL NETWORK FOR...ijesajournal
ABSTRACT
Fuzzy c-mean is one of the efficient tools used in character recognition. Back propagation neural network is another powerful that may be used in such field. A comparison between fuzzy c-mean and BP neural network classifiers are presented in this research to obtain the performance of both classifiers. The comparison was based on recognition efficiency; this efficiency was evaluated as the ratio of the number of assigned characters with unknown one to the number of character set related to that character. The fuzzy C-mean and BP neural network algorithms were tested on a set of hand written and machine printed dataset named Chars74K dataset using Matlab (2016 b) programming language and the result was that neural network classifier gave 82% recognition efficiency while fuzzy c –mean gave 78%. Neural network classifier is more superior than fuzzy C-mean in recognition due to the limitations of processing time of fuzzy C-mean that requires smaller image size and eventually this will cause less efficiency.
A Novel Framework For Numerical Character Recognition With Zoning Distance Fe...IJERD Editor
Advancements of Computer technology has made every organization to implement the automatic processing systems for its activities. One of the examples is the recognition of handwritten characters, which has always been a challenging task in image processing and pattern recognition. In this paper we propose Zone based features for recognition of the handwritten characters. In this zoning approach a digit image is divided into 8x8 zones and centre pixel is computed for each zone. This procedure is sequentially repeated for entire zone. Finally features are extracted for classification and recognition.
Angular Symmetric Axis Constellation Model for Off-line Odia Handwritten Char...IJAAS Team
This document presents a novel approach for recognizing offline handwritten Odia characters based on angular symmetric axis feature extraction. The approach generates unique boundary points for each skeletonized character image based on angles from the image center. It then extracts row and column symmetry axes by connecting these points. Features are extracted including mean distance and angle of the row and column symmetry axes. The approach was tested on 200 Odia character images using random forest and SVM classifiers, achieving recognition accuracy of 96.3% and 98.2% respectively.
This document summarizes a research paper that examines pricing strategy in a two-stage supply chain consisting of a supplier and retailer. The supplier offers a credit period to the retailer, who then offers credit to customers. A mathematical model is formulated to maximize total profit for the integrated supply chain system. The model considers three cases based on the relative lengths of the credit periods offered at each stage. Equations are developed to represent the profit functions for the supplier, retailer and overall system in each case. The goal is to determine the optimal selling price that maximizes total integrated profit.
The document discusses melanoma skin cancer detection using a computer-aided diagnosis system based on dermoscopic images. It begins with an introduction to skin cancer and melanoma. It then reviews existing literature on automated melanoma detection systems that use techniques like image preprocessing, segmentation, feature extraction and classification. Features extracted in other studies include asymmetry, border irregularity, color, diameter and texture-based features. The proposed system collects dermoscopic images and performs preprocessing, segmentation, extracts 9 features based on the ABCD rule, and classifies images using a neural network classifier to detect melanoma. It aims to develop an automated diagnosis system to eliminate invasive biopsy procedures.
This document summarizes various techniques for image segmentation that have been studied and proposed in previous research. It discusses edge-based, threshold-based, region-based, clustering-based, and other common segmentation methods. It also reviews applications of segmentation in medical imaging, plant disease detection, and other fields. While no single technique can segment all images perfectly, hybrid and adaptive methods combining multiple approaches may provide better results. Overall, image segmentation remains an important but challenging task in digital image processing and computer vision.
This document presents a test for detecting a single upper outlier in a sample from a Johnson SB distribution when the parameters of the distribution are unknown. The test statistic proposed is based on maximum likelihood estimates of the four parameters (location, scale, and two shape) of the Johnson SB distribution. Critical values of the test statistic are obtained through simulation for different sample sizes. The performance of the test is investigated through simulation, showing it performs well at detecting outliers when the contaminant observation represents a large shift from the original distribution parameters. An example application to census data is also provided.
This document summarizes a research paper that proposes a portable device called the "Disha Device" to improve women's safety. The device has features like live location tracking, audio/video recording, automatic messaging to emergency contacts, a buzzer, flashlight, and pepper spray. It is designed using an Arduino microcontroller connected to GPS and GSM modules. When the button is pressed, it sends an alert message with the woman's location, sets off an alarm, activates the flashlight and pepper spray for self-defense. The goal is to provide women a compact, one-click safety system to help them escape dangerous situations or call for help with just a single press of a button.
- The document describes a study that constructed physical fitness norms for female students attending social welfare schools in Andhra Pradesh, India.
- Researchers tested 339 students in classes 6-10 on speed, strength, agility and flexibility tests. Tests included 50m run, bend and reach, medicine ball throw, broad jump, shuttle run, and vertical jump.
- The results showed that 9th class students had the best average time for the 50m run. 10th class students had the highest flexibility on average. Strength and performance generally improved with increased class level.
This document summarizes research on downdraft gasification of biomass. It discusses how downdraft gasifiers effectively convert solid biomass into a combustible producer gas. The gasification process involves pyrolysis and reactions between hot char and gases that produce CO, H2, and CH4. Downdraft gasifiers are well-suited for biomass gasification due to their simple design and ability to manage the gasification process with low tar production. The document also reviews previous studies on gasifier configuration upgrades and their impact on performance, and the principles of downdraft gasifier operation.
This document summarizes the design and manufacturing of a twin spindle drilling attachment. Key points:
- The attachment allows a drilling machine to simultaneously drill two holes in a single setting, improving productivity over a single spindle setup.
- It uses a sun and planet gear arrangement to transmit power from the main spindle to two drilling spindles.
- Components like gears, shafts, and housing were designed using Creo software and manufactured. Drill chucks, bearings, and bits were purchased.
- The attachment was assembled and installed on a vertical drilling machine. It is aimed at improving productivity in mass production applications by combining two drilling operations into one setup.
The document presents a comparative study of different gantry girder profiles for various crane capacities and gantry spans. Bending moments, shear forces, and section properties are calculated and tabulated for 'I'-section with top and bottom plates, symmetrical plate girder, 'I'-section with 'C'-section top flange, plate girder with rolled 'C'-section top flange, and unsymmetrical plate girder sections. Graphs of steel weight required per meter length are presented. The 'I'-section with 'C'-section top flange profile is found to be optimized for biaxial bending but rolled sections may not be available for all spans.
This document summarizes research on analyzing the first ply failure of laminated composite skew plates under concentrated load using finite element analysis. It first describes how a finite element model was developed using shell elements to analyze skew plates of varying skew angles, laminations, and boundary conditions. Three failure criteria (maximum stress, maximum strain, Tsai-Wu) were used to evaluate first ply failure loads. The minimum load from the criteria was taken as the governing failure load. The research aims to determine the effects of various parameters on first ply failure loads and validate the numerical approach through benchmark problems.
This document summarizes a study that investigated the larvicidal effects of Aegle marmelos (bael tree) leaf extracts on Aedes aegypti mosquitoes. Specifically, it assessed the efficacy of methanol extracts from A. marmelos leaves in killing A. aegypti larvae (at the third instar stage) and altering their midgut proteins. The study found that the leaf extract achieved 50% larval mortality (LC50) at a concentration of 49 ppm. Proteomic analysis of larval midguts revealed changes in protein expression levels after exposure to the extract, suggesting its bioactive compounds can disrupt the midgut. The aim is to identify specific inhibitor proteins in the midg
This document presents a system for classifying electrocardiogram (ECG) signals using a convolutional neural network (CNN). The system first preprocesses raw ECG data by removing noise and segmenting the signals. It then uses a CNN to extract features directly from the ECG data and classify arrhythmias without requiring complex feature engineering. The CNN architecture contains 11 convolutional layers and is optimized using techniques like batch normalization and dropout. The system was tested on ECG datasets and achieved classification accuracy of over 93%, demonstrating its effectiveness at automated ECG classification.
This document presents a new algorithm for extracting and summarizing news from online newspapers. The algorithm first extracts news related to the topic using keyword matching. It then distinguishes different types of news about the same topic. A term frequency-based summarization method is used to generate summaries. Sentences are scored based on term frequency and the highest scoring sentences are selected for the summary. The algorithm was evaluated on news datasets from various newspapers and showed good performance in intrinsic evaluation metrics like precision, recall and F-score. Thus, the proposed method can effectively extract and summarize online news for a given keyword or topic.
Introduction- e - waste – definition - sources of e-waste– hazardous substances in e-waste - effects of e-waste on environment and human health- need for e-waste management– e-waste handling rules - waste minimization techniques for managing e-waste – recycling of e-waste - disposal treatment methods of e- waste – mechanism of extraction of precious metal from leaching solution-global Scenario of E-waste – E-waste in India- case studies.
A review on techniques and modelling methodologies used for checking electrom...nooriasukmaningtyas
The proper function of the integrated circuit (IC) in an inhibiting electromagnetic environment has always been a serious concern throughout the decades of revolution in the world of electronics, from disjunct devices to today’s integrated circuit technology, where billions of transistors are combined on a single chip. The automotive industry and smart vehicles in particular, are confronting design issues such as being prone to electromagnetic interference (EMI). Electronic control devices calculate incorrect outputs because of EMI and sensors give misleading values which can prove fatal in case of automotives. In this paper, the authors have non exhaustively tried to review research work concerned with the investigation of EMI in ICs and prediction of this EMI using various modelling methodologies and measurement setups.
Advanced control scheme of doubly fed induction generator for wind turbine us...IJECEIAES
This paper describes a speed control device for generating electrical energy on an electricity network based on the doubly fed induction generator (DFIG) used for wind power conversion systems. At first, a double-fed induction generator model was constructed. A control law is formulated to govern the flow of energy between the stator of a DFIG and the energy network using three types of controllers: proportional integral (PI), sliding mode controller (SMC) and second order sliding mode controller (SOSMC). Their different results in terms of power reference tracking, reaction to unexpected speed fluctuations, sensitivity to perturbations, and resilience against machine parameter alterations are compared. MATLAB/Simulink was used to conduct the simulations for the preceding study. Multiple simulations have shown very satisfying results, and the investigations demonstrate the efficacy and power-enhancing capabilities of the suggested control system.
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...University of Maribor
Slides from talk presenting:
Aleš Zamuda: Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapter and Networking.
Presentation at IcETRAN 2024 session:
"Inter-Society Networking Panel GRSS/MTT-S/CIS
Panel Session: Promoting Connection and Cooperation"
IEEE Slovenia GRSS
IEEE Serbia and Montenegro MTT-S
IEEE Slovenia CIS
11TH INTERNATIONAL CONFERENCE ON ELECTRICAL, ELECTRONIC AND COMPUTING ENGINEERING
3-6 June 2024, Niš, Serbia
Comparative analysis between traditional aquaponics and reconstructed aquapon...bijceesjournal
The aquaponic system of planting is a method that does not require soil usage. It is a method that only needs water, fish, lava rocks (a substitute for soil), and plants. Aquaponic systems are sustainable and environmentally friendly. Its use not only helps to plant in small spaces but also helps reduce artificial chemical use and minimizes excess water use, as aquaponics consumes 90% less water than soil-based gardening. The study applied a descriptive and experimental design to assess and compare conventional and reconstructed aquaponic methods for reproducing tomatoes. The researchers created an observation checklist to determine the significant factors of the study. The study aims to determine the significant difference between traditional aquaponics and reconstructed aquaponics systems propagating tomatoes in terms of height, weight, girth, and number of fruits. The reconstructed aquaponics system’s higher growth yield results in a much more nourished crop than the traditional aquaponics system. It is superior in its number of fruits, height, weight, and girth measurement. Moreover, the reconstructed aquaponics system is proven to eliminate all the hindrances present in the traditional aquaponics system, which are overcrowding of fish, algae growth, pest problems, contaminated water, and dead fish.
Using recycled concrete aggregates (RCA) for pavements is crucial to achieving sustainability. Implementing RCA for new pavement can minimize carbon footprint, conserve natural resources, reduce harmful emissions, and lower life cycle costs. Compared to natural aggregate (NA), RCA pavement has fewer comprehensive studies and sustainability assessments.
ACEP Magazine edition 4th launched on 05.06.2024Rahul
This document provides information about the third edition of the magazine "Sthapatya" published by the Association of Civil Engineers (Practicing) Aurangabad. It includes messages from current and past presidents of ACEP, memories and photos from past ACEP events, information on life time achievement awards given by ACEP, and a technical article on concrete maintenance, repairs and strengthening. The document highlights activities of ACEP and provides a technical educational article for members.
Literature Review Basics and Understanding Reference Management.pptxDr Ramhari Poudyal
Three-day training on academic research focuses on analytical tools at United Technical College, supported by the University Grant Commission, Nepal. 24-26 May 2024
Batteries -Introduction – Types of Batteries – discharging and charging of battery - characteristics of battery –battery rating- various tests on battery- – Primary battery: silver button cell- Secondary battery :Ni-Cd battery-modern battery: lithium ion battery-maintenance of batteries-choices of batteries for electric vehicle applications.
Fuel Cells: Introduction- importance and classification of fuel cells - description, principle, components, applications of fuel cells: H2-O2 fuel cell, alkaline fuel cell, molten carbonate fuel cell and direct methanol fuel cells.
DEEP LEARNING FOR SMART GRID INTRUSION DETECTION: A HYBRID CNN-LSTM-BASED MODELgerogepatton
As digital technology becomes more deeply embedded in power systems, protecting the communication
networks of Smart Grids (SG) has emerged as a critical concern. Distributed Network Protocol 3 (DNP3)
represents a multi-tiered application layer protocol extensively utilized in Supervisory Control and Data
Acquisition (SCADA)-based smart grids to facilitate real-time data gathering and control functionalities.
Robust Intrusion Detection Systems (IDS) are necessary for early threat detection and mitigation because
of the interconnection of these networks, which makes them vulnerable to a variety of cyberattacks. To
solve this issue, this paper develops a hybrid Deep Learning (DL) model specifically designed for intrusion
detection in smart grids. The proposed approach is a combination of the Convolutional Neural Network
(CNN) and the Long-Short-Term Memory algorithms (LSTM). We employed a recent intrusion detection
dataset (DNP3), which focuses on unauthorized commands and Denial of Service (DoS) cyberattacks, to
train and test our model. The results of our experiments show that our CNN-LSTM method is much better
at finding smart grid intrusions than other deep learning algorithms used for classification. In addition,
our proposed approach improves accuracy, precision, recall, and F1 score, achieving a high detection
accuracy rate of 99.50%.
A SYSTEMATIC RISK ASSESSMENT APPROACH FOR SECURING THE SMART IRRIGATION SYSTEMSIJNSA Journal
The smart irrigation system represents an innovative approach to optimize water usage in agricultural and landscaping practices. The integration of cutting-edge technologies, including sensors, actuators, and data analysis, empowers this system to provide accurate monitoring and control of irrigation processes by leveraging real-time environmental conditions. The main objective of a smart irrigation system is to optimize water efficiency, minimize expenses, and foster the adoption of sustainable water management methods. This paper conducts a systematic risk assessment by exploring the key components/assets and their functionalities in the smart irrigation system. The crucial role of sensors in gathering data on soil moisture, weather patterns, and plant well-being is emphasized in this system. These sensors enable intelligent decision-making in irrigation scheduling and water distribution, leading to enhanced water efficiency and sustainable water management practices. Actuators enable automated control of irrigation devices, ensuring precise and targeted water delivery to plants. Additionally, the paper addresses the potential threat and vulnerabilities associated with smart irrigation systems. It discusses limitations of the system, such as power constraints and computational capabilities, and calculates the potential security risks. The paper suggests possible risk treatment methods for effective secure system operation. In conclusion, the paper emphasizes the significant benefits of implementing smart irrigation systems, including improved water conservation, increased crop yield, and reduced environmental impact. Additionally, based on the security analysis conducted, the paper recommends the implementation of countermeasures and security approaches to address vulnerabilities and ensure the integrity and reliability of the system. By incorporating these measures, smart irrigation technology can revolutionize water management practices in agriculture, promoting sustainability, resource efficiency, and safeguarding against potential security threats.
A SYSTEMATIC RISK ASSESSMENT APPROACH FOR SECURING THE SMART IRRIGATION SYSTEMS
762019128
1. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
184
Doi: 10.32622/ijrat.762019128
Online Handwritten Gujarati Character Recognition:
Two-Layer Classification Approach
Vishal A. Naik and Apurva A. Desai
Abstract- The Gujarati language has large and
complex character set and many characters have
similar strokes, which makes OCR more challenging.
Here we suggest a two-layer classification technique
with SVM (RBF) and k-NN classifiers in order to
propose a robust online handwritten character
recognition for Gujarati language. In the first layer of
classification, SVM classifier with the RBF kernel is
used and in the second layer, k-NN classifier is used.
The training data of second layer classifier is decided
based on the outcome of first layer classifier. Training
data of a group of characters which are similar to a
character returned by first layer classifier, is supplied
to k-NN classifier. A hybrid feature set consisting first
and second order derivative of pixel values, zoning,
and normalized chain code feature. The data set of
around 12000 samples was generated from different
writers. Around 2000 samples of data set is used for
training and rest of the samples are used to test the
system. The proposed system has obtained an average
accuracy of 94.65% and an average processing time of
0.095 seconds per stroke.
Index Terms- Online Handwritten Character
Recognition (OHCR), Handwritten Character
Recognition (HCR), Optical Character Recognition
(OCR), Support Vector Machine (SVM), Gujarati
character Recognition
I. INTRODUCTION
The handwritten character recognition is fast-growing
and composite area of machine learning and pattern
recognition. The world is moving away from
traditional pen and paper kind of communication tool
and moving towards a digital world. As the world is
becoming digital, more and more people are adopting
digital technologies which result in increasing use of
handheld devices. Digital world opens many
opportunity and areas for researchers. There is a need
of online handwritten character recognition system to
provide an easy and efficient tool to communicate with
digital gadgets in a traditional language and traditional
way.
Manuscript revised June 16, 2019 and published on July 10, 2019
Vishal A. Naik, Department of Computer Science, Veer Narmad
South Gujarat University, Surat, India
Apurva A. Desai, Department of Computer Science, Veer Narmad
South Gujarat University, Surat, India
Indian languages have large and complex character sets
which makes a communication with the digital devices
more difficult using a simple keyboard. User
interaction in handheld devices can be made easy and
efficient using online character recognition.
There are 14 major official languages of India besides
Hindi and it is required to take serious measures for the
development of these languages [1]. Gujarati is a native
and official language of Indian state Gujarat. The
Gujarati language has a large and complex character
set. There are some strokes which are used in multiple
characters. For example, Fig. 1 shows character ‘D’,
’k’ and ‘h’. Stroke of character ‘D’ is used in other two
characters. There are some characters which has very
high similarities with some other stroke. For example,
Fig. 2 shows character ‘G’ and ‘F’. Fig. 3 shows
character ‘5’ and ‘y’. Fig. 4 shows a group of
confusing characters which includes ‘n’, ’t’, ’m’, ‘b’,
‘B’, and ‘K’.
Fig. 1 Character ‘D’, ’k’ and ‘h’
Fig. 2 Character ‘G’ and ‘F’
Fig. 3 Character ‘5’ and ‘y’
Fig. 4. Character ‘n’, ’t’, ’m’, ‘b’, ‘B’, and ‘K’
2. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
185
Doi: 10.32622/ijrat.762019128
The process of online handwritten character
recognition is very simple. The system tracks the pen
movement and an array of active pixels is recorded.
These raw pixel coordinate values must be processed
to fetch variety of features. The extracted features are
passed to the classifier for classification.
Handwritten character recognition can be categorized
into offline and online. Notable work is done by many
researchers in offline category.
In [2], the segmentation of text line and words for
Gujarati handwritten text was presented. Segmentation
was performed using projection profile-based
algorithm. The system obtained an accuracy of
89.24%. In [3], the Handwritten Gujarati numeral
recognition system was presented. The bilinear
interpolation method was used for normalization and
erosion method was used for thinning. The feature set
included four profile vectors. The FFBPNN was used
for classification. The system obtained an accuracy of
82%. In [4], the Handwritten Gujarati numeral
recognition system was presented. Global threshold,
erosion, dilation, and skew correction pre-processing
methods were used. The hybrid feature set includes
local and aspect ratio features. The k-Nearest Neighbor
classifier was used for classification. The system
obtained an accuracy of 96.99%. In [5], the zone
identification for Gujarati handwritten word was
presented. Zone identification was performed using
Euclidean transform method. The results found that
this method is simple and fast. The result showed an
accuracy of 75.2% for the upper zone, 75.2% for the
middle zone and 83.6% for the lower zone. In [6],
character segmentation method is presented for old
typewritten documents. Digitization of the document
was performed using a global threshold method. Noise
removal was done using a median filter method. Skew
correction and line segmentation were performed using
Radon transform method. Character segmentation was
performed using a vertical profile method. Line
segmentation of both languages obtained 100%
accuracy. Gujarati words segmentation obtained
94.79% accuracy and Gujarati characters segmentation
obtained 65.75% accuracy. Hindi words segmentation
obtained 96.18% accuracy. In [7], the handwritten
character recognition for Gujarati was presented. The
feature set included primary and secondary features.
The tree structure was used to compute a subset of
characters. Classification was performed using the k-
NN classifier. The system obtained an accuracy of
63.1%. In [8], the handwritten Gujarati alphabets
identification system was presented. The data set of
forty handwritten characters was collected from 199
different writers. The extracted features were aspect
ratio, the extent of the alphabet, and image subdivision
approach. Classification was performed using the SVM
classifier with a polynomial kernel. The result showed
an accuracy of 86.66%. Similar work is proposed in
[9]. The classification was performed using the Self-
Organizing Map (SOM) with k-NN. Training data was
directly provided to the SOM without implementing
any feature extraction method. The authors claimed
that their algorithm to be faster than other feature-based
algorithms. Real-time applications can be built using
this algorithm. The system obtained an accuracy of
98.13%. In [10] the authors have compared and
experiment with K-NN, SVM and Back Propagation
ANN classifiers with different possible options. The
feature set included spatial and transform domain
features like freeman chain code, Fourier descriptors,
discrete cosine transform coefficient. The k-NN, SVM,
and BPNN classifiers were compared using10-
foldcross-validation and obtained an accuracy of
85.67%, 93.60%, and 93.00% respectively.
In [11], the SVM was used for classification with
experiments and comparison between different kernels
with control variables for Gujarati numerals. The
hybrid features were implemented which includes
zoning, directional chain code features. The data set of
2000 samples were generated from different type of
writers. The authors have compared kernel’s accuracy
and processing time. They have obtained an average
accuracy of 95% using polynomial kernel and average
processing time per stroke of 0.13 seconds using linear
kernel. In [12] the SVM, MLP and k-NN classifiers are
tested and compared for Gujarati characters. They have
used aspect ratio, zoning and directional chain code
features for all classifiers. The system obtained the
highest accuracy of 91.63% using SVM with RBF
kernel. In [13] the authors have used multi-layer
classification using support vector machine classifier.
They have used different kernels in both layers.
Polynomial kernel is used at first layer and linear
kernel is used at second layer. Second classifier is used
only if first classifier classifies any confusing character
with training data of that confusing character’s group.
They have used zoning and DP based normalized chain
code features. The system obtained an average
accuracy of 94.13% and an average processing time per
stroke of 0.103 seconds.
In [14], the Online Handwriting Recognition for Tamil
was presented. Normalization, resampling using a
Gaussian low-pass filter for removing the noise, and an
equidistant resampling to remove variations in writing
speed methods were used. The feature set includes
local and directional features. The classification was
performed using the SVM with a DDAG and
discriminative classifier. The Viterbi decoding
algorithm was used for post-processing. The result
showed an accuracy of 95.78% for Malayalam and
95.12% for Telugu characters.
In [15], the recognition of Online Handwriting for
Malayalam and Telugu was presented. They have used
normalization, resampling using Gaussian low-pass
filter, and an equidistant resampling. They have used
following features, x & y co-ordinates, moments, area,
direction, curvature, length, and aspect ratio. They
3. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
186
Doi: 10.32622/ijrat.762019128
have used SVM using a DDAG and discriminative
classifier. They have used Viterbi decoding algorithm
for post processing. They have achieved accuracy of
95.78% accuracy.
In [16], the handwritten character segmentation and
recognition was presented for multiple languages. The
proposed character segmentation method uses vertical
projection, smoothing, thresholding, and over segment
reduction. The k-NN and majority voting principle was
used for classification. The geometric properties are
used as features. The system obtained an accuracy of
98.79% for segmentation and 99.6% for classification.
In [17], the authors have presented work for
handwritten Devanagari character recognition. The
combination of quadratic and SVM is used for
classification with gradient based directional features.
The complete features set has 392-dimensional
features. The system has obtained an accuracy of
95.81%.
In [18]–[20], the online Bangla character recognition
was presented. The database of 10000 data samples
was created which represents 50 different types of
strokes. Classification was performed using SVM,
SMO, MLP, Random Forest, Bayes net and Naïve
Bayes classifier. The authors compared three types of
features sets with different classifiers.
The first feature set had 136 feature values which
included center of gravity-based global and local
features. The result showed an accuracy of 98.26%
using SVM, 94.23% using SMO, 91.50% using MLP,
90% using simple logistic, 89.59% using Naïve Bayes,
and 88.24% using Bayes net classifier.
The second feature set included a Hausdorff Distance
(HD) and Directed HD features. The result showed an
accuracy of 95.57% using MLP, 94.85%using simple
logistic, 94.43% using SVM, and 88.61% using Bayes
net classifier.
The third feature set had 192 feature values which
included area features, mass distribution, chord length
features. The Krill-Herd algorithm was used to select
optimum features from all features. The result showed
an accuracy of 98.57% using SMO, 98.10% using
MLP, 90.48% using Naïve Bayes, 96.67% using Bayes
net, and 97.43% using Random forest classifier. The
highest accuracy of 98.57% was obtained using SMO
classifier with all features and computation time of
12.05 seconds.
In [21], the authors have presented work on online
Bangla character recognition. The database of 15000
data samples was used here. The feature set included
transition counts, a center of gravity-based features,
and topological features. This features set is compared
with different classifiers. The result showed an
accuracy of 94.56% using SVM, 93.07% using MLP,
91.70% using simple logistic, and 86.45% using Bayes
net classifier. After parameter tuning of SVM, an
accuracy of 95.49% was obtained.
In [22], the authors have presented work on online
handwritten English character recognition. The Hidden
Markov Model was used for classification. The
comparison between Gaussian mixture model-HMM
and hybrid deep neural network-HMM was performed.
The feature set included stroke velocity and raw points-
based features. The velocity profile was created for
every stroke which included sinusoidal based
horizontal, vertical and zero-crossing velocities. The
system was tested on the UNIPEN and the IRONOFF
databases. For the UNIPEN database, the highest
accuracy obtained was 99.04% for digits using GMM-
HMM, 97.43% for upper case characters using DNN-
HMM, and 95.42% for lower case characters using
DNN-HMM. For the IRONOFF database, the highest
accuracy obtained was 98.53% for digits using DNN-
HMM, 95.59% for upper case characters using DNN-
HMM, and 93.65% for lower case characters using
DNN-HMM. For the IRONOFF word database, the
highest accuracy obtained was 91.07% using DNN-
HMM.
II. PRE-PROCESSING
Different pre-processing methods are used to make
stroke level corrections before further processing. The
simple pre-processing methods are used here. The
normalization of stroke size and smoothing method are
used in pre-processing. Each stroke should have same
size so that similar features can be extracted. Bilinear
interpolation method is used to make a stroke size
normalized. Interpolation will be performed in X & Y
directions. Strokes should be smoothened before
feature extraction to remove additional noise.
Nonlinear median filter is used as a smoothing method.
III. FEATURE EXTRACTION
Raw pixel coordinate represents variety of information
which needs to be fetched out as meaningful unique
information. The feature set includes various features
to represent variety of information about the stroke.
Various feature extraction methods are used to fetch
variety of important values from raw coordinates.
These variety features values are the input for the
classification algorithms. Features is categorized into
global and local features and different types of
structural, statistical and hybrid methods to extract
such features.
The hybrid feature set is used here which has the
derivative of x & y, 16 zoning values, and normalized
chain code features. The first and second order
derivative of x & y at each pixel is calculated
independently of each other. The first and second order
derivative provides information about the change in the
trajectory at the current pixel. The first and second
order derivatives are calculated using the formulas 1 to
4 [23].
4. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
187
Doi: 10.32622/ijrat.762019128
The second order derivative values for character “B”
are, d2
x is 3,3,3,2,2,2,2,2,2,2,1,1,1,1,1,1,1,1,1,1,1,1,
1,1,1,1,1,2,2,2,2,2,2,3,3,3,3,4,4,4,3, 3 and d2
y is -1,-1,
0,0,0,1,1,2,2,2,3,3,3,3,3,3,3,3,3,3,3,3,3,3,2,2,2,2,2,2,2,
2,2,1,1,1,0,0,-1,-2,-2,-3.
Key points are selected from an input stroke after every
10% of pixels which are ten key points in total. These
key points are used to measure the curve direction
between each pair of key points. Different curve
direction values are shown in Fig. 5 based on which
chain code can be generated[24]. Fig. 6 shows freeman
chain code values for character “K” are 5, 12, 13, 3, 0,
1, 5, 11, and 13. Starting value of the chain code for the
same character can be different based on writing style.
The chain code can be starting point invariant, the
starting point should be redefined, starting with the
minimum value of chain code in the circular order.
Fig. 5 Constant directional values
The chain code values for character “K” are 0, 1, 5, 11,
13, 5, 12, 13, and 3. If the user writes strokes with some
skew or the stroke rotates then its chain code values
will be different. The difference between the pair of
chain code values are computed by counting the
number of code values between pair of code values
counter-clockwise. The chain code values for character
“K” is 1, 4, 6, 2, 8, 7, 1, 6.
In zoning, each stroke is divided into 16 equal sized
zones. The number of active pixels in each zone is
computed and its percentage is considered as a feature.
Fig. 7 shows stroke of character “K” which is divided
into 16 equal sized zones.
Fig. 6 Key points and chain code values of “K”
Fig. 7. Pixel distribution of “K” into 16 equal zones
Fig. 8. Steps of the proposed system
IV. CLASSIFICATION
The two-layer classification approach is proposed
using support Vector Machine classifier with Radial
Basis Function kernel at the first layer and k- Nearest
Neighbor at the second layer. The SVM classifier is
selected based on the comparison between SVM, MLP,
and k-NN[12]. The k-NN classifier is selected because
it requires low resources. Fig. 8 shows different steps
required in the proposed system.
5. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
188
Doi: 10.32622/ijrat.762019128
SVM algorithm is using a supervised learning
approach which implicitly transforms data using
different mathematics-based kernels. By using
different kernels, SVM can select appropriate
threshold. SVM tries to find global optimum and
unique solution. A constant value of control variables
is 𝛾 = 0.002 and C =1 which is selected after trial and
error method.
k-NN is a distance-based classifier. It uses Euclidean
distance or Hamming distance to calculate a distance
between testing and training data. K nearest data
samples define the class label. Different values of k
results in different accuracy and processing time. The
value of k is set to 3 after experimenting with different
values of k.
In the two-layer classification approach, the first layer
classifier uses a complete training data set. The second
layer classifier uses a limited or selected training data
based on the character returned by the first classifier.
For second layer, training data is divided into eight
groups based on the similarities between characters.
Each group is consisting of characters which is similar
to each other. Figure 4.8 shows a group of character
consisting of six half characters and six full characters.
If first layer classifier returns a character “K” which
belongs to the group shown in figure 9 then the training
data of these 12 characters are used for training of
second layer classifier.
Fig. 9 Training data group
V. RESULTS AND DISCUSSION
Training of the system will be performed in two layers.
In the first layer, all training data is used to train SVM
and limited training data of a single group is used to
train k-NN. There are total 08 groups are created for
second layer training. Selection of a group for second
layer training is based on the result of first layer. 10000
samples are used to test the proposed system. The
training and testing are performed on a touch screen
personal computer using developed Graphical User
Interface (GUI) system.
Table I describes performance comparison between
different system approaches. The system has obtained
an average accuracy of 95.65% using proposed two-
layer classification approach and it took an average
processing time of 0.095 seconds per stroke. The
performance of the proposed system is increased
compared to the multilayer classification
approach[13]. An average accuracy of 94.13% is
obtained using a multilayer system and it took an
average processing time of 0.103 seconds per stroke.
Table I. Performance Comparison
Method
Accuracy
(%)
Avg.
Processing
time (Seconds)
Multilayer
system
94.13 0.103
Proposed system 95.65 0.095
Fig. 10. Accuracy of Gujarati Numerals
Fig. 11. Accuracy of Gujarati characters
Fig. 12. Accuracy of Gujarati characters
Fig. 10 shows an average accuracy of numerals. The
two-layer method obtained 95.7% and multilayer
method obtained 95% accuracy for numerals.
88
90
92
94
96
98
100
0 1 2 3 4 5 6 7 8 9
AVG.ACCURACY
NUMERALS
Proposed Method Multilayer Method
86
88
90
92
94
96
98
W n Y R 5 I J
AVG.ACCURACY
CHARACTERS
Proposed Method Multilayer Method
80
85
90
95
100
U G T D E A B
AVG.ACCURACY
CHARACTERS
Proposed Method Multilayer Method
6. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
189
Doi: 10.32622/ijrat.762019128
Fig. 11 shows an average accuracy of a group 1 of
similar characters. The two-layer method obtained
93.71% and multilayer method obtained 89.83%
accuracy for a group of similar characters. Fig. 12
shows an average accuracy of a group 2 of similar
characters. The two-layer method obtained 93.83% and
multilayer method obtained 82.14% accuracy for
another group of similar characters.
Fig. 13 shows an average accuracy of a group 1 of
multi-stroke characters. The two-layer method obtained
96.75% and multilayer method obtained 95% accuracy
for a group of multi-stroke characters.
Fig. 13. Accuracy of multi-stroke characters
Fig. 14. Accuracy of multi-stroke characters
Fig. 15. Accuracy of half characters
Fig. 14 shows an average accuracy of a group 2 of
multi-stroke characters. The two-layer method
obtained 94.4% and multilayer method obtained 91.8%
accuracy for another group of multi-stroke characters.
Fig. 15 shows an average accuracy of a group 1 of half
characters. The two-layer method obtained 95%
accuracy for a group of half characters. Fig. 16 shows
an average accuracy of a group 2 of half characters. The
two-layer method obtained 90.33% accuracy for
another group of half characters.
Fig. 16. Accuracy of half characters
VI. CONCLUSION
The two-layer classification system has used SVM
with RBF kernel and k-NN classifiers with a hybrid
feature set. The two-layer classification system has
achieved an average accuracy of 95.64% which is
better than multilayer system and 0.095 seconds of
average execution time per stroke which is lesser than
multilayer system. The two-layer system’s
performance is directly based on the result of first layer
classifier. If the first layer classifier fails then second
layer classifier leads to the wrong classification.
REFERENCES
[1] “Official languages of India.” [Online]. Available:
http://rajbhasha.nic.in/en/official-language-resolution-1968.
[Accessed: 27-June-2019].
[2] C. Patel and A. Desai, “Segmentation of text lines into words
for Gujarati handwritten text,” Proc. 2010 Int. Conf. Signal
Image Process. ICSIP 2010, pp. 130–134, 2010.
[3] A. A. Desai, “Gujarati handwritten numeral optical character
reorganization through neural network,” J. Pattern Recognit.,
vol. 43, no. 7, pp. 2582–2589, 2010.
[4] A. A. Desai, “Handwritten Gujarati Numeral Optical
Character Recognition using Hybrid Feature Extraction
Technique,” Int. Conf. Image Process. Comput. Vision,
Pattern Recognition, IPCV, 2010.
[5] C. Patel and A. Desai, “Zone identification for Gujarati
handwritten word,” Proc. - 2nd Int. Conf. Emerg. Appl. Inf.
Technol. EAIT 2011, pp. 194–197, 2011.
[6] A. Desai, “Segmentation of Characters from old Typewritten
Documents using Radon Transform,” Int. J. Comput. Appl.,
vol. 37, no. 9, pp. 10–15, 2012.
[7] C. Patel and A. Desai, “Gujarati Handwritten Character
Recognition Using Hybrid Method Based on Binary Tree-
Classifier And K-Nearest Neighbour,” Int. J. Eng. Res.
Technol., vol. 2, no. 6, pp. 2337–2345, 2013.
[8] A. A. Desai, “Support vector machine for identification of
handwritten Gujarati alphabets using hybrid feature space,”
CSI Trans. ICT, vol. 2, no. January, pp. 235–241, 2015.
[9] M. M. Goswami and S. K. Mitra, “Classification of Printed
Gujarati Characters Using Low-Level,” ACM Trans. Asian
Low-Resour. Lang. Inf. Process. 15, 4, Artic. 25 (April 2016),
90
92
94
96
98
100
S h ; , O C Ø
AVG.ACCURACY
MULTI-STROKE CHARACTERS
Proposed Method Multilayer Method
84
86
88
90
92
94
96
98
~ ·i Èi Éi ä ú
AVG.ACCURACY
MULTI-STROKE CHARACTERS
Proposed Method Multilayer Method
86
88
90
92
94
96
98
³ º R À ¿ ¾ ² O
AVG.ACCURACY
HALF CHARACTERS
Proposed Method
82
87
92
µ Á ¼ ¹ ¸ Ç A #
AVG.ACCURACY
HALF CHARACTERS
Proposed Method
7. International Journal of Research in Advent Technology, Vol.7, No.6, June 2019
E-ISSN: 2321-9637
Available online at www.ijrat.org
190
Doi: 10.32622/ijrat.762019128
26 pages. DOI http//dx.doi.org/10.1145/2856105, vol. 15, no.
4, 2016.
[10] A. N. Vyas and M. M. Goswami, “Classification of
handwritten Gujarati numerals,” 2015 Int. Conf. Adv. Comput.
Commun. Informatics, ICACCI 2015, pp. 1231–1237, 2015.
[11] V. A. Naik and A. A. Desai, “Online Handwritten Gujarati
Numeral Recognition Using Support Vector Machine,” Int. J.
Comput. Sci. Eng., vol. 6, no. 9, pp. 416–421, 2018.
[12] V. A. Naik and A. A. Desai, “Online Handwritten Gujarati
Character Recognition Using SVM , MLP , And k-NN,” in 8th
International Conference on Computing, Communication and
Networking Technologies (ICCCNT), Delhi, 2017, pp. 1–6.
[13] V. A. Naik and A. A. Desai, “Multi-Layer Classification
Approach for Online Handwritten Gujarati Character
Recognition,” Comput. Intell. Theor. Appl. Futur. Dir. - Vol.
II. Adv. Intell. Syst. Comput., vol. 799, pp. 595–606, 2018.
[14] K. H. Aparna, V. Subramanian, M. Kasirajan, G. V. Prakash,
and V. S. Chakravarthy, “Online Handwriting Recognition for
Tamil,” Ninth Int. Work. Front. Handwrit. Recognit., pp. 438–
443, 2004.
[15] A. Arora and A. M. Namboodiri, “A hybrid model for
recognition of online handwriting in Indian scripts,” in
International Conference on Frontiers in Handwriting
Recognition, ICFHR 2010, 2010, pp. 433–438.
[16] P. Sahare and S. B. Dhok, “Robust Character Segmentation
and Recognition Schemes for Multilingual Indian Document
Images,” IETE Tech. Rev. (Institution Electron. Telecommun.
Eng. India), vol. 4602, pp. 1–14, 2018.
[17] M. Bhalerao, S. Bonde, A. Nandedkar, and S. Pilawan,
“Combined Classifier Approach for Offline Handwritten
Devanagari Character Recognition Using Multiple Features,”
Comput. Vis. Bio Inspired Comput., pp. 45–54, 2018.
[18] S. Sen, A. Bhattacharyya, A. Das, R. Sarkar, and K. Roy,
“Design of Novel Feature Vector for Recognition of Online
Handwritten Bangla Basic Characters,” in the First
International Conference on Intelligent Computing and
Communication. Advances in Intelligent Systems and
Computing, 2016, vol. 458, pp. 485–494.
[19] S. Sen, R. Sarkar, K. Roy, and N. Hori, “Recognize Online
Handwritten Bangla Characters Using Hausdorff Distance-
Based Feature,” in the 5th International Conference on
Frontiers in Intelligent Computing: Theory and Applications.
Advances in Intelligent Systems and Computing, 2017, vol.
515, pp. 541–549.
[20] S. Sen, M. Mitra, A. Bhattacharyya, R. Sarkar, F. Schwenker,
and K. Roy, “Feature Selection for Recognition of Online
Handwritten Bangla Characters,” Neural Process. Lett., 2019.
[21] S. Sen, A. Bhattacharyya, P. K. Singh, R. A. M. Sarkar, K.
Roy, and D. Doermann, “Application of Structural and
Topological Features to Recognize Online Handwritten
Bangla Characters,” ACM Trans. Asian Low-Resour. Lang.
Inf. Process., vol. 17, no. 3, 2018.
[22] H. Choudhury and S. R. Mahadeva Prasanna, “Representation
of Online Handwriting using Multi-component Sinusoidal
Model,” Pattern Recognit., 2019.
[23] G. S. Reddy, B. Sarma, R. K. Naik, S. R. M. Prasanna, and C.
Mahanta, “Assamese online handwritten digit recognition
system using hidden Markov models,” Proceeding Work.
Doc. Anal. Recognit. - DAR ’12, p. 108, 2012.
[24] X. Q. Bo YU, Lei Guo, Tianyun Zhao, “A curve matching
algorithm based on Freeman chain code,” Int. Conf. Intell.
Comput. Intell. Syst., pp. 669–672, 2010.
AUTHORS PROFILE
Mr. V. A. Naik received Master of Science
(I.T.) from Veer Narmad South Gujarat
University, Surat, Gujarat, India in 2010. He is
currently pursuing Ph.D. from Veer Narmad
South Gujarat University, Surat, Gujarat,
India. His main research work focuses on
handwritten character recognition, pattern
matching, and image processing. He has 8
years of teaching experience.
Mr. A. A. Desai received a Ph.D. degree from
the South Gujarat University, Gujarat, India in
1997. He is a professor with the Department of
Computer Science, Veer Narmad South
Gujarat University (VNSGU) India since
2004. He is also Dean of Faculty of Computer
Science and Information Technology since
2012. He is involved in many projects during
his long academic and research experience of
over 25 years. His research activities are in the
area of Digital Image Processing, Pattern
Recognition, Natural Language Processing,
and Data Mining.