1. INTRODUCTION
Handwriting is one of the oldest biometrics. There is a major
division between types of verification systems. On-line (dynamic)
systems involving real-time digital capture of a signature can
record multiple features of the signature as a function of time,
such as pen pressure and azimuth. Off-line (static) systems rely
purely on the handwritings’ image but require less hardware. It is
on the latter that this project focuses.
IMPLEMENTATION AND RESULTS
The best outputs for different sets of images are tabulated
below:
Number of
Images
Number of Iterations Peak Value for
Accuracy
10 1000 80.2712%
20 300 94.20811%
50 1000 83.5142%
70 500 88.7540%
100 300 90.0216%
FLOW DIAGRAM
Preprocessing (conversion to
grayscale and binary)
Feature Extraction
Mean Vector Extraction
Into ACO
Accuracy = 92.41%
Values of ACO parameters rho = 0.6, beta = 1.0, alpha = 1.0, ∆τ= 0.0, no. of ants = 5, cutoff= 0.10.
FURTHER WORK
The following points highlight the work that we are planning to
implement in future:
Calculations of individual handwriting features of a writer by
calculating moments, end junctions etc.
Feature Extraction Module can be further enhanced in order to
obtain more appropriate results.
Large data sets can be used for example- 500, 1000 images or
more.
Comparison of the proposed algorithm with the other existing
algorithms and analysis.
For large data sets Grid Framework can be used in order to make
the calculations easy.
DESIGN
Input –Scanned Handwritten Text Images in any format.
Following are the steps carried out during this project:
During Preprocessing the images are fed into a Matlab
code where the noise removal process is carried out by
applying some filters like Gaussian, Canny etc. and are
converted to grayscale format.
After noise removal the images are converted to binary
format and are provided as input to the feature extraction
code.
Once the features are extracted, mean vector calculation is
done using gcc compiler in Linux environment.
The mean feature vector is then given as input to the ACO
code.
Output – Similar Images are grouped together to form clusters
and
accuracy of the output is displayed.
Handwritten Script matching using natural computation:
An efficient approach
Submitted By: Anik Biswas (05103442)
Under the supervision of: Mr.
Yamuna Prasad Shukla
PROBLEM STATEMENT
Various approaches have been developed for analysis of
handwritten image data with some improvements but still there
are lot of research is to be carried out for it.
Based upon those, following tasks will be conducted during our
project work:
Handwritten image preprocessing in Matlab.
Feature extraction of images in Matlab
Prediction and analysis of image matching via
classification through Ant Colony Optimization over
Linux environment with gcc libraries.
Group 3
Group 2
Group 1
CONCLUSION
The Accuracy obtained after running ACO code was
remarkable and images were properly grouped into clusters.
REFERENCES
[1] Marius Bulacu, and Lambert Schomaker “Text-Independent
Writer Identification and Verification Using Textural and Allographic
Features” IEEE Transactions on pattern analysis and machine
intelligence, vol. 29, no. 4, April 2007.
[2] R. Plamondon and S.N. Srihari, “Online and Offline Handwriting
Recognition: A Comprehensive Survey,” IEEE Trans. Pattern Analysis
and Machine Intelligence, vol. 22, no. 1, Jan. 2000.