Digital image tampering detection has been an active area of research in recent times due to the ease with
which digital image can be modified to convey false or misleading information. To address this problem,
several studies have proposed forensics algorithms for digital image tampering detection. While these
approaches have shown remarkable improvement, most of them only focused on detecting a specific type of
image tampering. The limitation of these approaches is that new forensic method must be designed for
each new manipulation approach that is developed. Consequently, there is a need to develop methods
capable of detecting multiple tampering operations. In this paper, we proposed a novel general purpose
image tampering scheme based on CNNs and Local Optimal Oriented Pattern (LOOP) which is capable of
detecting five types of image tampering in both binary and multiclass scenarios. Unlike the existing deep
learning techniques which used constrained pre-processing layers to suppress the effect of image content
in order to capture image tampering traces, our method uses LOOP features, which can effectively subdue
the effect image content, thus, allowing the proposed CNNs to capture the needed features to distinguish
among different types of image tampering. Through a number of detailed experiments, our results
demonstrate that the proposed general purpose image tampering method can achieve high detection
accuracies in individual and multiclass image tampering detections respectively and a comparative
analysis of our results with the existing state of the arts reveals that the proposed model is more robust
than most of the exiting methods.
Novel framework for optimized digital forensic for mitigating complex image ...IJECEIAES
Digital Image Forensic is significantly becoming popular owing to the increasing usage of the images as a media of information propagation. However, owing to the presence of various image editing tools and software, there is also an increasing threat to image content security. Reviewing the existing approaches to identify the traces or artifacts states that there is a large scope of optimization to be implemented to enhance the processing further. Therefore, this paper presents a novel framework that performs cost-effective optimization of digital forensic technique with an idea of accurately localizing the area of tampering as well as offers a capability to mitigate the attacks of various forms. The study outcome shows that the proposed system offers better outcomes in contrast to the existing system to a significant scale to prove that minor novelty in design attributes could induce better improvement with respect to accuracy as well as resilience toward all potential image threats.
EXPLOITING REFERENCE IMAGES IN EXPOSING GEOMETRICAL DISTORTIONSijma
Nowadays, image alteration in the mainstream media has become common. The degree of manipulation is
facilitated by image editing software. In the past two decades the number indicating manipulation of
images rapidly grows. Hence, there are many outstanding images which have no provenance information
or certainty of authenticity. Therefore, constructing a scientific and automatic way for evaluating image
authenticity is an important task, which is the aim of this paper. In spite of having outstanding
performance, all the image forensics schemes developed so far have not provided verifiable information
about source of tampering. This paper aims to propose a different kind of scheme, by exploiting a group of
similar images, to verify the source of tampering. First, we define our definition with regard to tampered
image. The distinctive features are obtained by exploiting Scale- Invariant Feature Transform (SIFT)
technique. We then proposed clustering technique to identify the tampered region based on distinctive
keypoints. In contrast to k-means algorithm, our technique does not require the initialization of k value. The
experimental results over and beyond the dataset indicate the efficacy of our proposed scheme
A mediator person is required for communication between deaf person and a second person. But a
mediator should know the sign language used by deaf person. But this is also not possible always since there are
multiple sign languages for multiple languages. It is difficult for a deaf person to understand what a second
person speaks. And therefore deaf person should keep track of lip movements of second person in order to know
what he is speaking. But the lip movements do not give proper efficiency and accuracy since the facial
expressions and speech might not match. To overcome the above problems we have proposed a system, an
Android Application for recognizing sign language using hand gesture with the facility for user to define and
upload their own sign language into the system. The features of this system are the real time conversion of
gesture to text and speech. For two-way communication between deaf person and second person, the speech of
second person is converted into text. The processing steps include: gesture extraction, gesture matching and
conversion of text to speech and vice-versa. The system is not only useful for deaf community but can also be
used by common people who migrate to different regions and do not know local language.
Novel framework for optimized digital forensic for mitigating complex image ...IJECEIAES
Digital Image Forensic is significantly becoming popular owing to the increasing usage of the images as a media of information propagation. However, owing to the presence of various image editing tools and software, there is also an increasing threat to image content security. Reviewing the existing approaches to identify the traces or artifacts states that there is a large scope of optimization to be implemented to enhance the processing further. Therefore, this paper presents a novel framework that performs cost-effective optimization of digital forensic technique with an idea of accurately localizing the area of tampering as well as offers a capability to mitigate the attacks of various forms. The study outcome shows that the proposed system offers better outcomes in contrast to the existing system to a significant scale to prove that minor novelty in design attributes could induce better improvement with respect to accuracy as well as resilience toward all potential image threats.
EXPLOITING REFERENCE IMAGES IN EXPOSING GEOMETRICAL DISTORTIONSijma
Nowadays, image alteration in the mainstream media has become common. The degree of manipulation is
facilitated by image editing software. In the past two decades the number indicating manipulation of
images rapidly grows. Hence, there are many outstanding images which have no provenance information
or certainty of authenticity. Therefore, constructing a scientific and automatic way for evaluating image
authenticity is an important task, which is the aim of this paper. In spite of having outstanding
performance, all the image forensics schemes developed so far have not provided verifiable information
about source of tampering. This paper aims to propose a different kind of scheme, by exploiting a group of
similar images, to verify the source of tampering. First, we define our definition with regard to tampered
image. The distinctive features are obtained by exploiting Scale- Invariant Feature Transform (SIFT)
technique. We then proposed clustering technique to identify the tampered region based on distinctive
keypoints. In contrast to k-means algorithm, our technique does not require the initialization of k value. The
experimental results over and beyond the dataset indicate the efficacy of our proposed scheme
A mediator person is required for communication between deaf person and a second person. But a
mediator should know the sign language used by deaf person. But this is also not possible always since there are
multiple sign languages for multiple languages. It is difficult for a deaf person to understand what a second
person speaks. And therefore deaf person should keep track of lip movements of second person in order to know
what he is speaking. But the lip movements do not give proper efficiency and accuracy since the facial
expressions and speech might not match. To overcome the above problems we have proposed a system, an
Android Application for recognizing sign language using hand gesture with the facility for user to define and
upload their own sign language into the system. The features of this system are the real time conversion of
gesture to text and speech. For two-way communication between deaf person and second person, the speech of
second person is converted into text. The processing steps include: gesture extraction, gesture matching and
conversion of text to speech and vice-versa. The system is not only useful for deaf community but can also be
used by common people who migrate to different regions and do not know local language.
Efficient resampling features and convolution neural network model for image ...IJEECSIAES
The extended utilization of picture-enhancing or manipulating tools has led to ease of manipulating multimedia data which includes digital images. These manipulations will disturb the truthfulness and lawfulness of images, resulting in misapprehension, and might disturb social security. The image forensic approach has been employed for detecting whether or not an image has been manipulated with the usage of positive attacks which includes splicing, and copy-move. This paper provides a competent tampering detection technique using resampling features and convolution neural network (CNN). In this model range spatial filtering (RSF)-CNN, throughout preprocessing the image is divided into consistent patches. Then, within every patch, the resampling features are extracted by utilizing affine transformation and the Laplacian operator. Then, the extracted features are accumulated for creating descriptors by using CNN. A wide-ranging analysis is performed for assessing tampering detection and tampered region segmentation accuracies of proposed RSF-CNN based tampering detection procedures considering various falsifications and post-processing attacks which include joint photographic expert group (JPEG) compression, scaling, rotations, noise additions, and more than one manipulation. From the achieved results, it can be visible the RSF-CNN primarily based tampering detection with adequately higher accurateness than existing tampering detection methodologies.
Efficient resampling features and convolution neural network model for image ...nooriasukmaningtyas
The extended utilization of picture-enhancing or manipulating tools has led to ease of manipulating multimedia data which includes digital images. These manipulations will disturb the truthfulness and lawfulness of images, resulting in misapprehension, and might disturb social security. The image forensic approach has been employed for detecting whether or not an image has been manipulated with the usage of positive attacks which includes splicing, and copy-move. This paper provides a competent tampering detection technique using resampling features and convolution neural network (CNN). In this model range spatial filtering (RSF)-CNN, throughout preprocessing the image is divided into consistent patches. Then, within every patch, the resampling features are extracted by utilizing affine transformation and the Laplacian operator. Then, the extracted features are accumulated for creating descriptors by using CNN. A wide-ranging analysis is performed for assessing tampering detection and tampered region segmentation accuracies of proposed RSF-CNN based tampering detection procedures considering various falsifications and post-processing attacks which include joint photographic expert group (JPEG) compression, scaling, rotations, noise additions, and more than one manipulation. From the achieved results, it can be visible the RSF-CNN primarily based tampering detection with adequately higher accurateness than existing tampering detection methodologies.
Frequency based edge-texture feature using Otsu’s based enhanced local ternar...journalBEEI
Sharing information through images is a trend nowadays. Advancements in the technology and user-friendly image editing tool make easy to edit the image and spread fake news through different social networking platforms. Forged image has been generated through an advanced image editing tool, so it is very challenging for image forensics to detect the micro discrepancy which distorted the micro pattern. This paper proposes an image forensic detection technique, which implies multi-level discrete wavelet transform to implement digital image filtering. Canny edge detection technique is implemented to detect the edge of the image to implement Otsu’s based enhanced local ternary pattern (OELTP), which can detect forgery-related artifact. DWT is implemented over Cb and Cr components of the image and using edge texture to improve the Otsu global threshold, which is used to extract features using ELTP technique. Support vector machine (SVM) is used for classification to find the image is forged or not. The performance of the work evaluated on three different open available data sets CASIA v1, CASIA v2, and Columbia. Our proposed work gives better results with some of the previous states of the work in terms of detection accuracy.
Analysis and Detection of Image Forgery Methodologiesijsrd.com
"Forgery" is a subjective word. An image can become a forgery based upon the context in which it is used. An image altered for fun or someone who has taken a bad photo, but has been altered to improve its appearance cannot be considered a forgery even though it has been altered from its original capture. The other side of forgery are those who perpetuate a forgery for gain and prestige. They create an image in which to dupe the recipient into believing the image is real and from this they are able to gain payment and fame. Detecting these types of forgeries has become serious problem at present. To determine whether a digital image is original or doctored is a big challenge. To find the marks of tampering in a digital image is a challenging task. Now these marks of tampering can be done by various operations such as rotation, scaling, JPEG compression, Gaussian noise etc. called as attacks. There are various methods proposed in this field in recent years to detect above mentioned attacks. This paper provides a detailed analysis of different approaches and methodologies used to detect image forgery. It is also analysed that block-based features methods are robust to Gaussian noise and JPEG compression and the key point-based feature methods are robust to rotation and scaling.
Image forgery detection using error level analysis and deep learningTELKOMNIKA JOURNAL
Many images are spread in the virtual world of social media. With the many editing software that allows so there is no doubt that many forgery images. By forensic the image using Error Level Analysis to find out the compression ratio between the original image and the fake image, because the original image compression and fake images are different. In addition to knowing whether the image is genuine or fake can analyze the metadata of the image, but the possibility of metadata can be changed. In this case the authors apply Deep Learning to recognize images of manipulations through the dataset of a fake image and original images via Error Level Analysis on each image and supporting parameters for error rate analysis. The result of our experiment is that we get the best accuracy of training 92.2% and 88.46% validation by going through 100 epoch.
Extraction of image resampling using correlation aware convolution neural ne...IJECEIAES
Detecting hybrid tampering attacks in an image is extremely difficult; especially when copy-clone tampered segments exhibit identical illumination and contrast level about genuine objects. The existing method fails to detect tampering when the image undergoes hybrid transformation such as scaling, rotation, compression, and also fails to detect under smallsmooth tampering. The existing resampling feature extraction using the Deep learning techniques fails to obtain a good correlation among neighboring pixels in both horizontal and vertical directions. This work presents correlation aware convolution neural network (CA-CNN) for extracting resampling features for detecting hybrid tampering attacks. Here the image is resized for detecting tampering under a small-smooth region. The CA-CNN is composed of a three-layer horizontal, vertical, and correlated layer. The correlated layer is used for obtaining correlated resampling feature among horizontal sequence and vertical sequence. Then feature is aggregated and the descriptor is built. An experiment is conducted to evaluate the performance of the CA-CNN model over existing tampering detection methodologies considering the various datasets. From the result achieved it can be seen the CA-CNN is efficient considering various distortions and post-processing attacks such joint photographic expert group (JPEG) compression, and scaling. This model achieves much better accuracies, recall, precision, false positive rate (FPR), and F-measure compared existing methodologies.
Passive Image Forensic Method to Detect Resampling Forgery in Digital 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.
Top Cited Articles in Signal & Image Processing 2021-2022sipij
Signal & Image Processing : An International Journal is an Open Access peer-reviewed journal intended for researchers from academia and industry, who are active in the multidisciplinary field of signal & image processing. The scope of the journal covers all theoretical and practical aspects of the Digital Signal Processing & Image processing, from basic research to development of application.
Authors are solicited to contribute to the journal by submitting articles that illustrate research results, projects, surveying works and industrial experiences that describe significant advances in the areas of Signal & Image processing.
Facial recognition based on enhanced neural networkIAESIJAI
Accurate automatic face recognition (FR) has only become a practical goal of biometrics research in recent years. Detection and recognition are the primary steps for identifying faces in this research, and The Viola-Jones algorithm implements to discover faces in images. This paper presents a neural network solution called modify bidirectional associative memory (MBAM). The basic idea is to recognize the image of a human's face, extract the face image, enter it into the MBAM, and identify it. The output ID for the face image from the network should be similar to the ID for the image entered previously in the training phase. The tests have conducted using the suggested model using 100 images. Results show that FR accuracy is 100% for all images used, and the accuracy after adding noise is the proportions that differ between the images used according to the noise ratio. Recognition results for the mobile camera images were more satisfactory than those for the Face94 dataset.
AN OVERVIEW OF COPY MOVE FORGERY DETECTION APPROACHESCSEIJJournal
Images have greater expressive power than any other forms of documents. With the Internet, images are
widespread in several applications. But the availability of efficient open-source online photo editing tools
has made editing these images easy. The fake images look more appealing and original than the real image
itself, which makes them indistinguishable and hence difficult to detect. The authenticity of digital images
like medical reports, scan images, financial data, crime evidence, legal evidence, etc. is of high
importance. Detecting the forgery of images is therefore a major research area. Image forgery is
categorized as copy-move forgery, splicing, and retouching. In this work, a review of copy-move forgery is
discussed along with the existing research on its detection and localization using both conventional and
deep-learning mechanisms. The datasets used and challenges towards improving or developing novel
algorithms are also presented
An Overview of Copy Move Forgery Detection ApproachesCSEIJJournal
Images have greater expressive power than any other forms of documents. With the Internet, images are
widespread in several applications. But the availability of efficient open-source online photo editing tools
has made editing these images easy. The fake images look more appealing and original than the real image
itself, which makes them indistinguishable and hence difficult to detect. The authenticity of digital images
like medical reports, scan images, financial data, crime evidence, legal evidence, etc. is of high
importance. Detecting the forgery of images is therefore a major research area. Image forgery is
categorized as copy-move forgery, splicing, and retouching. In this work, a review of copy-move forgery is
discussed along with the existing research on its detection and localization using both conventional and
deep-learning mechanisms. The datasets used and challenges towards improving or developing novel
algorithms are also presented.
Efficient resampling features and convolution neural network model for image ...IJEECSIAES
The extended utilization of picture-enhancing or manipulating tools has led to ease of manipulating multimedia data which includes digital images. These manipulations will disturb the truthfulness and lawfulness of images, resulting in misapprehension, and might disturb social security. The image forensic approach has been employed for detecting whether or not an image has been manipulated with the usage of positive attacks which includes splicing, and copy-move. This paper provides a competent tampering detection technique using resampling features and convolution neural network (CNN). In this model range spatial filtering (RSF)-CNN, throughout preprocessing the image is divided into consistent patches. Then, within every patch, the resampling features are extracted by utilizing affine transformation and the Laplacian operator. Then, the extracted features are accumulated for creating descriptors by using CNN. A wide-ranging analysis is performed for assessing tampering detection and tampered region segmentation accuracies of proposed RSF-CNN based tampering detection procedures considering various falsifications and post-processing attacks which include joint photographic expert group (JPEG) compression, scaling, rotations, noise additions, and more than one manipulation. From the achieved results, it can be visible the RSF-CNN primarily based tampering detection with adequately higher accurateness than existing tampering detection methodologies.
Efficient resampling features and convolution neural network model for image ...nooriasukmaningtyas
The extended utilization of picture-enhancing or manipulating tools has led to ease of manipulating multimedia data which includes digital images. These manipulations will disturb the truthfulness and lawfulness of images, resulting in misapprehension, and might disturb social security. The image forensic approach has been employed for detecting whether or not an image has been manipulated with the usage of positive attacks which includes splicing, and copy-move. This paper provides a competent tampering detection technique using resampling features and convolution neural network (CNN). In this model range spatial filtering (RSF)-CNN, throughout preprocessing the image is divided into consistent patches. Then, within every patch, the resampling features are extracted by utilizing affine transformation and the Laplacian operator. Then, the extracted features are accumulated for creating descriptors by using CNN. A wide-ranging analysis is performed for assessing tampering detection and tampered region segmentation accuracies of proposed RSF-CNN based tampering detection procedures considering various falsifications and post-processing attacks which include joint photographic expert group (JPEG) compression, scaling, rotations, noise additions, and more than one manipulation. From the achieved results, it can be visible the RSF-CNN primarily based tampering detection with adequately higher accurateness than existing tampering detection methodologies.
Frequency based edge-texture feature using Otsu’s based enhanced local ternar...journalBEEI
Sharing information through images is a trend nowadays. Advancements in the technology and user-friendly image editing tool make easy to edit the image and spread fake news through different social networking platforms. Forged image has been generated through an advanced image editing tool, so it is very challenging for image forensics to detect the micro discrepancy which distorted the micro pattern. This paper proposes an image forensic detection technique, which implies multi-level discrete wavelet transform to implement digital image filtering. Canny edge detection technique is implemented to detect the edge of the image to implement Otsu’s based enhanced local ternary pattern (OELTP), which can detect forgery-related artifact. DWT is implemented over Cb and Cr components of the image and using edge texture to improve the Otsu global threshold, which is used to extract features using ELTP technique. Support vector machine (SVM) is used for classification to find the image is forged or not. The performance of the work evaluated on three different open available data sets CASIA v1, CASIA v2, and Columbia. Our proposed work gives better results with some of the previous states of the work in terms of detection accuracy.
Analysis and Detection of Image Forgery Methodologiesijsrd.com
"Forgery" is a subjective word. An image can become a forgery based upon the context in which it is used. An image altered for fun or someone who has taken a bad photo, but has been altered to improve its appearance cannot be considered a forgery even though it has been altered from its original capture. The other side of forgery are those who perpetuate a forgery for gain and prestige. They create an image in which to dupe the recipient into believing the image is real and from this they are able to gain payment and fame. Detecting these types of forgeries has become serious problem at present. To determine whether a digital image is original or doctored is a big challenge. To find the marks of tampering in a digital image is a challenging task. Now these marks of tampering can be done by various operations such as rotation, scaling, JPEG compression, Gaussian noise etc. called as attacks. There are various methods proposed in this field in recent years to detect above mentioned attacks. This paper provides a detailed analysis of different approaches and methodologies used to detect image forgery. It is also analysed that block-based features methods are robust to Gaussian noise and JPEG compression and the key point-based feature methods are robust to rotation and scaling.
Image forgery detection using error level analysis and deep learningTELKOMNIKA JOURNAL
Many images are spread in the virtual world of social media. With the many editing software that allows so there is no doubt that many forgery images. By forensic the image using Error Level Analysis to find out the compression ratio between the original image and the fake image, because the original image compression and fake images are different. In addition to knowing whether the image is genuine or fake can analyze the metadata of the image, but the possibility of metadata can be changed. In this case the authors apply Deep Learning to recognize images of manipulations through the dataset of a fake image and original images via Error Level Analysis on each image and supporting parameters for error rate analysis. The result of our experiment is that we get the best accuracy of training 92.2% and 88.46% validation by going through 100 epoch.
Extraction of image resampling using correlation aware convolution neural ne...IJECEIAES
Detecting hybrid tampering attacks in an image is extremely difficult; especially when copy-clone tampered segments exhibit identical illumination and contrast level about genuine objects. The existing method fails to detect tampering when the image undergoes hybrid transformation such as scaling, rotation, compression, and also fails to detect under smallsmooth tampering. The existing resampling feature extraction using the Deep learning techniques fails to obtain a good correlation among neighboring pixels in both horizontal and vertical directions. This work presents correlation aware convolution neural network (CA-CNN) for extracting resampling features for detecting hybrid tampering attacks. Here the image is resized for detecting tampering under a small-smooth region. The CA-CNN is composed of a three-layer horizontal, vertical, and correlated layer. The correlated layer is used for obtaining correlated resampling feature among horizontal sequence and vertical sequence. Then feature is aggregated and the descriptor is built. An experiment is conducted to evaluate the performance of the CA-CNN model over existing tampering detection methodologies considering the various datasets. From the result achieved it can be seen the CA-CNN is efficient considering various distortions and post-processing attacks such joint photographic expert group (JPEG) compression, and scaling. This model achieves much better accuracies, recall, precision, false positive rate (FPR), and F-measure compared existing methodologies.
Passive Image Forensic Method to Detect Resampling Forgery in Digital 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.
Top Cited Articles in Signal & Image Processing 2021-2022sipij
Signal & Image Processing : An International Journal is an Open Access peer-reviewed journal intended for researchers from academia and industry, who are active in the multidisciplinary field of signal & image processing. The scope of the journal covers all theoretical and practical aspects of the Digital Signal Processing & Image processing, from basic research to development of application.
Authors are solicited to contribute to the journal by submitting articles that illustrate research results, projects, surveying works and industrial experiences that describe significant advances in the areas of Signal & Image processing.
Facial recognition based on enhanced neural networkIAESIJAI
Accurate automatic face recognition (FR) has only become a practical goal of biometrics research in recent years. Detection and recognition are the primary steps for identifying faces in this research, and The Viola-Jones algorithm implements to discover faces in images. This paper presents a neural network solution called modify bidirectional associative memory (MBAM). The basic idea is to recognize the image of a human's face, extract the face image, enter it into the MBAM, and identify it. The output ID for the face image from the network should be similar to the ID for the image entered previously in the training phase. The tests have conducted using the suggested model using 100 images. Results show that FR accuracy is 100% for all images used, and the accuracy after adding noise is the proportions that differ between the images used according to the noise ratio. Recognition results for the mobile camera images were more satisfactory than those for the Face94 dataset.
AN OVERVIEW OF COPY MOVE FORGERY DETECTION APPROACHESCSEIJJournal
Images have greater expressive power than any other forms of documents. With the Internet, images are
widespread in several applications. But the availability of efficient open-source online photo editing tools
has made editing these images easy. The fake images look more appealing and original than the real image
itself, which makes them indistinguishable and hence difficult to detect. The authenticity of digital images
like medical reports, scan images, financial data, crime evidence, legal evidence, etc. is of high
importance. Detecting the forgery of images is therefore a major research area. Image forgery is
categorized as copy-move forgery, splicing, and retouching. In this work, a review of copy-move forgery is
discussed along with the existing research on its detection and localization using both conventional and
deep-learning mechanisms. The datasets used and challenges towards improving or developing novel
algorithms are also presented
An Overview of Copy Move Forgery Detection ApproachesCSEIJJournal
Images have greater expressive power than any other forms of documents. With the Internet, images are
widespread in several applications. But the availability of efficient open-source online photo editing tools
has made editing these images easy. The fake images look more appealing and original than the real image
itself, which makes them indistinguishable and hence difficult to detect. The authenticity of digital images
like medical reports, scan images, financial data, crime evidence, legal evidence, etc. is of high
importance. Detecting the forgery of images is therefore a major research area. Image forgery is
categorized as copy-move forgery, splicing, and retouching. In this work, a review of copy-move forgery is
discussed along with the existing research on its detection and localization using both conventional and
deep-learning mechanisms. The datasets used and challenges towards improving or developing novel
algorithms are also presented.
We have compiled the most important slides from each speaker's presentation. This year’s compilation, available for free, captures the key insights and contributions shared during the DfMAy 2024 conference.
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdffxintegritypublishin
Advancements in technology unveil a myriad of electrical and electronic breakthroughs geared towards efficiently harnessing limited resources to meet human energy demands. The optimization of hybrid solar PV panels and pumped hydro energy supply systems plays a pivotal role in utilizing natural resources effectively. This initiative not only benefits humanity but also fosters environmental sustainability. The study investigated the design optimization of these hybrid systems, focusing on understanding solar radiation patterns, identifying geographical influences on solar radiation, formulating a mathematical model for system optimization, and determining the optimal configuration of PV panels and pumped hydro storage. Through a comparative analysis approach and eight weeks of data collection, the study addressed key research questions related to solar radiation patterns and optimal system design. The findings highlighted regions with heightened solar radiation levels, showcasing substantial potential for power generation and emphasizing the system's efficiency. Optimizing system design significantly boosted power generation, promoted renewable energy utilization, and enhanced energy storage capacity. The study underscored the benefits of optimizing hybrid solar PV panels and pumped hydro energy supply systems for sustainable energy usage. Optimizing the design of solar PV panels and pumped hydro energy supply systems as examined across diverse climatic conditions in a developing country, not only enhances power generation but also improves the integration of renewable energy sources and boosts energy storage capacities, particularly beneficial for less economically prosperous regions. Additionally, the study provides valuable insights for advancing energy research in economically viable areas. Recommendations included conducting site-specific assessments, utilizing advanced modeling tools, implementing regular maintenance protocols, and enhancing communication among system components.
NUMERICAL SIMULATIONS OF HEAT AND MASS TRANSFER IN CONDENSING HEAT EXCHANGERS...ssuser7dcef0
Power plants release a large amount of water vapor into the
atmosphere through the stack. The flue gas can be a potential
source for obtaining much needed cooling water for a power
plant. If a power plant could recover and reuse a portion of this
moisture, it could reduce its total cooling water intake
requirement. One of the most practical way to recover water
from flue gas is to use a condensing heat exchanger. The power
plant could also recover latent heat due to condensation as well
as sensible heat due to lowering the flue gas exit temperature.
Additionally, harmful acids released from the stack can be
reduced in a condensing heat exchanger by acid condensation. reduced in a condensing heat exchanger by acid condensation.
Condensation of vapors in flue gas is a complicated
phenomenon since heat and mass transfer of water vapor and
various acids simultaneously occur in the presence of noncondensable
gases such as nitrogen and oxygen. Design of a
condenser depends on the knowledge and understanding of the
heat and mass transfer processes. A computer program for
numerical simulations of water (H2O) and sulfuric acid (H2SO4)
condensation in a flue gas condensing heat exchanger was
developed using MATLAB. Governing equations based on
mass and energy balances for the system were derived to
predict variables such as flue gas exit temperature, cooling
water outlet temperature, mole fraction and condensation rates
of water and sulfuric acid vapors. The equations were solved
using an iterative solution technique with calculations of heat
and mass transfer coefficients and physical properties.
Online aptitude test management system project report.pdfKamal Acharya
The purpose of on-line aptitude test system is to take online test in an efficient manner and no time wasting for checking the paper. The main objective of on-line aptitude test system is to efficiently evaluate the candidate thoroughly through a fully automated system that not only saves lot of time but also gives fast results. For students they give papers according to their convenience and time and there is no need of using extra thing like paper, pen etc. This can be used in educational institutions as well as in corporate world. Can be used anywhere any time as it is a web based application (user Location doesn’t matter). No restriction that examiner has to be present when the candidate takes the test.
Every time when lecturers/professors need to conduct examinations they have to sit down think about the questions and then create a whole new set of questions for each and every exam. In some cases the professor may want to give an open book online exam that is the student can take the exam any time anywhere, but the student might have to answer the questions in a limited time period. The professor may want to change the sequence of questions for every student. The problem that a student has is whenever a date for the exam is declared the student has to take it and there is no way he can take it at some other time. This project will create an interface for the examiner to create and store questions in a repository. It will also create an interface for the student to take examinations at his convenience and the questions and/or exams may be timed. Thereby creating an application which can be used by examiners and examinee’s simultaneously.
Examination System is very useful for Teachers/Professors. As in the teaching profession, you are responsible for writing question papers. In the conventional method, you write the question paper on paper, keep question papers separate from answers and all this information you have to keep in a locker to avoid unauthorized access. Using the Examination System you can create a question paper and everything will be written to a single exam file in encrypted format. You can set the General and Administrator password to avoid unauthorized access to your question paper. Every time you start the examination, the program shuffles all the questions and selects them randomly from the database, which reduces the chances of memorizing the questions.
KuberTENes Birthday Bash Guadalajara - K8sGPT first impressionsVictor Morales
K8sGPT is a tool that analyzes and diagnoses Kubernetes clusters. This presentation was used to share the requirements and dependencies to deploy K8sGPT in a local environment.
Saudi Arabia stands as a titan in the global energy landscape, renowned for its abundant oil and gas resources. It's the largest exporter of petroleum and holds some of the world's most significant reserves. Let's delve into the top 10 oil and gas projects shaping Saudi Arabia's energy future in 2024.
Hierarchical Digital Twin of a Naval Power SystemKerry Sado
A hierarchical digital twin of a Naval DC power system has been developed and experimentally verified. Similar to other state-of-the-art digital twins, this technology creates a digital replica of the physical system executed in real-time or faster, which can modify hardware controls. However, its advantage stems from distributing computational efforts by utilizing a hierarchical structure composed of lower-level digital twin blocks and a higher-level system digital twin. Each digital twin block is associated with a physical subsystem of the hardware and communicates with a singular system digital twin, which creates a system-level response. By extracting information from each level of the hierarchy, power system controls of the hardware were reconfigured autonomously. This hierarchical digital twin development offers several advantages over other digital twins, particularly in the field of naval power systems. The hierarchical structure allows for greater computational efficiency and scalability while the ability to autonomously reconfigure hardware controls offers increased flexibility and responsiveness. The hierarchical decomposition and models utilized were well aligned with the physical twin, as indicated by the maximum deviations between the developed digital twin hierarchy and the hardware.
Cosmetic shop management system project report.pdfKamal Acharya
Buying new cosmetic products is difficult. It can even be scary for those who have sensitive skin and are prone to skin trouble. The information needed to alleviate this problem is on the back of each product, but it's thought to interpret those ingredient lists unless you have a background in chemistry.
Instead of buying and hoping for the best, we can use data science to help us predict which products may be good fits for us. It includes various function programs to do the above mentioned tasks.
Data file handling has been effectively used in the program.
The automated cosmetic shop management system should deal with the automation of general workflow and administration process of the shop. The main processes of the system focus on customer's request where the system is able to search the most appropriate products and deliver it to the customers. It should help the employees to quickly identify the list of cosmetic product that have reached the minimum quantity and also keep a track of expired date for each cosmetic product. It should help the employees to find the rack number in which the product is placed.It is also Faster and more efficient way.
Final project report on grocery store management system..pdfKamal Acharya
In today’s fast-changing business environment, it’s extremely important to be able to respond to client needs in the most effective and timely manner. If your customers wish to see your business online and have instant access to your products or services.
Online Grocery Store is an e-commerce website, which retails various grocery products. This project allows viewing various products available enables registered users to purchase desired products instantly using Paytm, UPI payment processor (Instant Pay) and also can place order by using Cash on Delivery (Pay Later) option. This project provides an easy access to Administrators and Managers to view orders placed using Pay Later and Instant Pay options.
In order to develop an e-commerce website, a number of Technologies must be studied and understood. These include multi-tiered architecture, server and client-side scripting techniques, implementation technologies, programming language (such as PHP, HTML, CSS, JavaScript) and MySQL relational databases. This is a project with the objective to develop a basic website where a consumer is provided with a shopping cart website and also to know about the technologies used to develop such a website.
This document will discuss each of the underlying technologies to create and implement an e- commerce website.
Final project report on grocery store management system..pdf
General Purpose Image Tampering Detection using Convolutional Neural Network and Local Optimal Oriented Pattern (LOOP)
1. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
DOI: 10.5122/sipij.2021.12202 13
GENERAL PURPOSE IMAGE TAMPERING
DETECTION USING CONVOLUTIONAL NEURAL
NETWORK AND LOCAL OPTIMAL ORIENTED
PATTERN (LOOP)
Ali Ahmad Aminu1, 2
and Nwojo Nnanna Agwu1
1
Department of Computer Science, Nile University of Nigeria, Abuja, Nigeria
2
Department of Computer Science, Gombe State University, Gombe, Gombe, Nigeria
ABSTRACT
Digital image tampering detection has been an active area of research in recent times due to the ease with
which digital image can be modified to convey false or misleading information. To address this problem,
several studies have proposed forensics algorithms for digital image tampering detection. While these
approaches have shown remarkable improvement, most of them only focused on detecting a specific type of
image tampering. The limitation of these approaches is that new forensic method must be designed for
each new manipulation approach that is developed. Consequently, there is a need to develop methods
capable of detecting multiple tampering operations. In this paper, we proposed a novel general purpose
image tampering scheme based on CNNs and Local Optimal Oriented Pattern (LOOP) which is capable of
detecting five types of image tampering in both binary and multiclass scenarios. Unlike the existing deep
learning techniques which used constrained pre-processing layers to suppress the effect of image content
in order to capture image tampering traces, our method uses LOOP features, which can effectively subdue
the effect image content, thus, allowing the proposed CNNs to capture the needed features to distinguish
among different types of image tampering. Through a number of detailed experiments, our results
demonstrate that the proposed general purpose image tampering method can achieve high detection
accuracies in individual and multiclass image tampering detections respectively and a comparative
analysis of our results with the existing state of the arts reveals that the proposed model is more robust
than most of the exiting methods.
KEYWORDS
Image Tampering, General purpose Tampering Detection, Convolutional Neural Network, Local Optimal
Oriented Pattern
1. INTRODUCTION
Today, digital media including images have become one of the main medium of communication
due to their expressive abilities, ease of acquisition, distribution, and storage [1]. The
effectiveness of digital images in conveying information have made them more preferable than
text information as a means of communication. Consequently, it is becoming common more than
ever to see an image representing a prime source of evidence in the court of law, a prime source
of information in crime investigation, a source of news by mass media and news publishing
agencies. At the same time, the nature of the digital image has raised a lot of questions in most of
these positive aspects where they are employed. A digital image can be easily modified to convey
false or misleading information. The development of sophisticated computers and image editing
software has made the manipulation of digital images very easy [2]. Thus, the manipulations of
images for malicious purposes are now rampant in our society leading to many ethical and moral
2. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
14
consequences, such as the spread of fake news, wrong verdict, and damage of reputation among
others [3]. This difference between the importance of digital image on one hand and the
uncertainties regarding their vulnerability to manipulations on the other hand calls for a reliable
and efficient method of assessing their authenticity before basing important societal decisions on
them. Hence, the study and research on detecting digital image tampering becomes vital in
today’s digital era.
In line with the above, a handful of image tampering detection and localization approaches have
been proposed in recent times, aiming at improving the state of the art image tampering detection
and localization methods. The earliest methods such as [4], [5], and [6] perform image tampering
detection by exploiting frequency domain features, Color Filter Array features, and local binary
descriptors. Although, these approaches have been successfully used to detect different type of
image tampering, they could not be used to detect more than one type of image tampering
operations because they operate based on the fact that each image tampering operation usually
leave traces behind. To identify a particular image tampering operation, researcher designed
algorithms that will extract these traces and use them to detect targeted image manipulation. The
limitations of these approaches is that they are designed to detect only one image tampering type.
Thus, several test must be carried out to detect whether a given image is tampered or authentic
[3]. Consequently, there is a need to develop methods capable of detecting multiple tampering
operations.
To address these problems, recent studies in image tampering have focused on designing general
purposed or universal approaches capable of detecting more than one tampering type. Inspired by
the performance of Spatial Rich Model (SRM) in image steganalysis, many general purpose
image tampering detection approaches [7] [8] [9] [10] utilizing SRM features have been
proposed, which yielded excellent results. With the success of deep learning methods,
specifically, CNN in many visual recognition tasks, recent studies [11] [12] [13] [14] in image
forensics also seek to leverage the strength of deep learning methods to solve the problem of
detecting and localizing digital image forgery. These methods can automatically learn
manipulations traces directly from data without the need for handcrafted features or human
analysis by using a set of convolution kernels whose weights are learned via neural network
training technique known as back-propagation. Hence, they provide better performance than the
earliest methods. While these methods have improve on the performance of image tampering
detection systems, the performances of these systems still requires significant improvement, so
that they can meet the need of real-world forensic task.
Therefore, in this paper, we proposed a novel general purpose CNNs and LOOP based image
tampering detection method capable of detecting different image tampering operations in both
binary and multiclass classification scenarios. Unlike existing approaches that uses hand designed
features or constrained pre-processing layers, the proposed method can directly extract image
tampering traces directly from the LOOP images, which has the effect of suppressing the effect
of image content allowing the proposed CNN to capture the different traces needed to detect
different types of image tampering concurrently. To assess the performance of the proposed
model, we have tested it through a number of experiments and the results of these experiments
demonstrate the effectiveness of the proposed model in detecting individual, multiple, as well as
manipulation chains image tampering in both un-compressed and compressed image formats.
The rest of this paper is organized as follows: Section 2 presents the related work. Section 3
discusses the proposed method. Section 4 presents the experiment and results. Finally, section 5
concludes the paper and highlights future research direction.
3. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
15
2. RELATED WORK
A handful of techniques have been proposed for detecting image tampering such as copy-move,
image splicing, object removal, and content preserving manipulations such as median filtering
and JPEG compression. In this section, we will briefly discuss some of the existing methods used
for detecting multiple digital image tampering.
Inspired by the success of Spatial Rich Model (SRM) in many image steganalysis task, Qiu et al.
[15], suggested the use of steganalytic features such as SPAM (Subtractive Pixel Adjacency
Matrix) [16] and SRM [17] to detect six different types of image processing operations. The work
of Fan et al. [18] used Gaussian mixture models to model the statistics of image process by
different image operation. Bayar and Stamm, [3] presented a universal image manipulation
detection technique that utilizes deep learning approach. They proposed a new convolution layer
called constrained convolution layer capable of suppressing image content to learn image
manipulation operation directly from data. Their method could effectively detect specific and
multiple image manipulation operation and it showed superiority over Spatial Rich Model (SRM)
based general purpose image manipulation approaches.
Moreover, Sundus et al [19] also investigate and demonstrate the performance of SRM and Local
Binary Pattern (LBP) in detecting multiple image tampering. They embedded LBP in SRM sub-
models to capture detailed statistics of the quantized version of image noise residuals. The
resulting features were used for classification using an ensemble classifier. In [20], chen et al.
proposed a new CNN based method to adaptively learn discriminative features for identifying
image processing operations. They carefully designed the high pass filter bank to get the image
residuals of the input image, the channel expansion layer to mix up the resulting residuals. Their
method achieved state of the art result. The authors of [21], suggested a dual stream CNN model
for detecting resampling of recompressed image. They used low-order high pass filter for
computing image residuals used for classification. Zhan et al. [22] proposed a transfer learning
based technique which enable them to train their model using small amount of training data. In
[23], authors proposed a universal image forensics method based on Siamese network. The
proposed method takes as input a pair of image patches and decides whether they are identically
or differently processed.
Recently, the work of [24], Fridrich and Boroumond presented a model for detecting the
processing history of image based on CNN with an IP layer accepting statistical moments of
feature maps. The proposed model could correctly classify images of different size and is robust
to JPEG compression. In [25], authors proposed a general purpose forgery detection and
localization using anomalous features. Their approach works on image of arbitrary size and could
detect and localize many known and unknown types of image forgery. In [26], chen et al.
proposed a multipurpose image forensics tool using densely connected CNN that could detect 11
different types of manipulations. Their method works efficiently on image of different sizes and it
was robust against JPEG compression. Zhang and Ni [27] proposed a dense Unit with a cross-
layer intersection for detection and localization of image forgeries. They initialized the weights of
their network with high pass filters used in SRM and used a multi-stage training approach to
speed up convergence.
3. PROPOSED METHODOLOGY
In this section, we present an overview of the proposed approach for general purpose digital
image tampering detection, followed by a detail description of the key aspects involved in the
subsequent subsections. The proposed method utilizes the discriminative strengths of Local
4. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
16
Optimal Oriented Pattern (LOOP) [28] and the feature extraction capabilities of Convolutional
Neural Networks (CNNs) to build a general framework for digital image tampering detection.
Developing a general purpose image tampering detection approach requires the use of low-level
features that can reveal fingerprints of different types of image tampering. LOOP is capable of
suppressing the semantic information in an image allowing the proposed CNN to extract and
learn the low level features required to distinguish different types of image tampering. Thus,
unlike existing techniques that utilize actual image as input to their CNN model, we employed
LOOP representation of images as the input of the proposed CNN design due to their
discrimination strengths and the fact that image manipulation traces are more pronounced in
LOOP image than the original image. Figure 3.1, gives an overview of the proposed method.
First an RGB image is converted to grayscale image, from which the LOOP image is computed.
The LOOP image having the same size as the original image is then fed to the designed CNN
architecture which then extract and learn deep features that will be used for classification by the
fully connected layers. To further assess the feature extraction capabilities of the proposed CNN,
deep features extracted from the second dense layer of the proposed CNN are used to train and
evaluate two tree based classifiers, XGBOOST and Extra tree classifier, on the same task
performed by the proposed CNN.
3.1. Local Optimal Oriented Pattern (LOOP)
LOOP is a binary descriptor recently proposed by Chakraborti et al [28], mostly used in the
description of local image features and the classification of texture. Texture in this context refers
to gray scale or color pixel intensities of image. LOOP is an enhancement of local binary
descriptors such as Local Binary pattern (LBP) [29], and Local Directional Pattern (LDP) [30]
and their variants which have recorded many success in image classification tasks including
many image forensics tasks [19][31][32]. LOOP improved on existing local descriptors by adding
rotation invariance into the main formulation of local descriptors, decreasing computational time
and increasing overall accuracy [28].The texture of an image provides details about the grayscale
or colour pixel intensities variation. Tampering operations such as, median filtering, copy move,
image splicing etc. inevitably introduce textural variations in images. LOOP are powerful texture
descriptors which can capture such variations caused by the tampering operations. To leverage on
the feature extraction capabilities of CNNs and the discriminative strengths of LOOP, we
proposed a novel image tampering approach that combines the strength of CNNs and LOOP
features.
3.2. LOOP Computation
Given an image I, LOOP representation can be obtained using equation (1) and (2) below.
Suppose 𝑖𝑐 represent the intensity of the image I at pixel (𝑥𝑐, 𝑦𝑐) and 𝑖𝑛 (n = 0, 1… 7) is the
intensity of a pixel in the 3×3 neighborhood of (𝑥𝑐, 𝑦𝑐) ignoring the center pixel 𝑖𝑐 . The 8 Kirsch
masks are oriented in the direction of the 8 neighboring pixels 𝑖𝑛 (n = 0, 1, 2… 7), giving a
measure of the strength of intensity variation in those directions respectively.
Suppose 𝑚𝑛denotes the responses of the 8 Kirsch masks corresponding to pixels with intensity
𝑖𝑛 (n = 0, 1, 2… 7). An exponential weight 𝑤𝑛(a digit between 0 and 7) is assigned to each of
these pixels according to the rank of the magnitude of 𝑚𝑛among the 8 Kirsch masks activations.
Then, the LOOP value for the pixel(𝑥𝑐, 𝑦𝑐)) can be computed as:
7
0
, (1)
.2 n
c n c
c
n
w
Loop f
y
x i i
5. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
17
Where
The LOOP image is formed by accumulating the occurrence of the LOOP values over the entire
image, i.e. each image is represented as a collection of LOOP values obtained from equation (1).
Each of 𝑖𝑛neighbouring pixel oriented in the direction of the response of the 8 kirsch masks is
compare with the central pixel 𝑖𝑐which evaluates to either 0 or 1 according to equation (2)
forming an 8 digit binary number. Weights are then assigned to each of these binary numbers
according to the rank of the magnitude of 𝑚𝑛 among the output of the 8 Kirsch masks. The
resulting binary number is then used as a label for that central pixel. For convenience, the binary
number is converted to decimal. Repeating this procedure for the entire pixels of an image, the
LOOP image is computed having the same size as the original image.
Figure 1: An overview of the proposed framework for general purpose image tampering detection. First,
an input image is converted to LOOP representation which serves as input to the CNN network. Then the
convolution layers of the network extracts and learns generic low level features from the LOOP images.
The learned features are forwarded to the fully connected layers (dense units) which classifies an image as
either tampered or not. To further validate the feature extraction capabilities of the proposed network, deep
features from the second dense layer of the proposed network are extracted and used for training and
testing of XGBOOST and extra tree classifiers on the same task performed by the proposed network.
The resulting LOOP images are used as input to our proposed model. Figure 3.2 illustrates the
LOOP images of the original image and that of the four corresponding tampered image. Applying
LOOP have successfully suppressed the effect of the image content, thus, allowing us to easily
extract tampering traces introduced by different image tampering operation. The five images
differs in their textural properties which shows that the different tampering operation affects the
pixel intensity of an image differently. The LOOP images for the altered images have introduced
noises of varying intensities and forms when compared to that of the original image. These
1, 0
( ) (2)
0,
if x
f x otherwise
6. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
18
noticeable differences in appearance among the five LOOP images signifies that the LOOP image
has successfully capture the different texture variation induced by the different tampering
operations.
(a) (b) (c)
(d) (e) (f)
Figure 2: illustration of LOOP images of Original and Tampered images. (a) Original Image, LOOP image
of (b) Original Image, (c) Gaussian Blurred Image, (d) gamma Corrected image (e) Median Filtered Image
and (f) JPEG Compressed image.
3.3. The Proposed CNN Architecture for General Purpose Image Tampering
Detection
Figure 3.3 illustrates the overall architecture of the proposed CNN for general purpose image
tampering detection. The proposed CNN contains 13 layers, the first 10 are convolution layers
and the last 3 are fully connected layers. Batch Normalization operation and activation function
are disable in the first convolution layer to prevent the loss of information, since they are
information losing operations. Similarly, layer 1 to 4 is directed connected without pooling
operations. Pooling operations were only introduced after layers 5, 7, 9 and 10 respectively to
reduce the size of the feature maps. The output of the second dense unit (FC 2) is then fed to the
Softmax, XGBOOST and Extra Tree classifiers which respectively classify an image as either
original or tampered. In the proposed CNN design, the input layer is a 256×256 and 128
grayscale LOOP images, and network parameters such as the number of filters, their size, and
initial values are arrived at experimentally. Below, we describe some of the features of the new
network architecture.
3.3.1. Weight Initialization
CNN with many layers are usually difficult to train due to their huge amount of parameters and
the fact that their loss function is non-convex [73], however, their performance can be optimized
by proper network initialization. In the proposed CNN design, we experimented with different
weight initialization schemes namely Xavier initialization with normal bias and zero bias, He-
normal with normal bias and zero bias, and lastly truncated normal with normal and zero bias.
Experimenting with all methods of initialization leads to the realization that our model performs
best when the weights are initialized using Xavier initialization and biases are initialized with
7. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
19
zeros in all convolutional layers. Unlike other related studies such as [73], we did not constrain
the parameters of our kernels to fixed high pass filters which may affect the generalization ability
of models.
3.3.2. Convolution layers
The convolution operation uses one or several kernels to filter the input image, generating an
array of feature maps for subsequent processing. The convolution operation is typically expressed
by:
𝑋𝑗
𝑙+1
= ∑ 𝑋𝑖
𝑙
∗ 𝑊𝑖 𝑗
𝑙
+ 𝑏𝑗
𝑖
(3)
Where 𝑋𝑖
𝑙
denote the j-th
feature map of the i-th
layer, 𝑊𝑖 𝑗
𝑙
is the trainable convolution kernels
connecting the j-th
output map and the i-th
input map, 𝑏𝑗
𝑖
is the trainable bias parameter for the j-th
output map. As depicted in Figure 3.3, the proposed model uses 10 convolution layers. The first
convolution layer “conv1” filters the 256×256×1 LOOP input image with 5 filters of size 7×7
using a stride of 1. The second convolution layer “conv2” filters the feature maps (250×250×5)
produced by the first convolution layer with 144 kernels of size 5×5 using a stride of 2.
123×123×144 feature maps output by the second convolution layer is filtered by the third
convolution layer “conv3” with 64 kernels of size 3×3 with a stride of 2 to yield a feature map of
61×61×64. BN and activation are disable in the first convolution layer and the first four
convolution layers are connected directly without pooling layers to prevent information loss.
These layers are responsible for extracting and learning image tampering fingerprints from the
input images. The next seven convolution layers namely “Conv4” with 64 filters of size 3×3 and
stride of 1, “Conv5” with 64 filters of size 3×3 and stride of 1, “Conv6” with 64 filters of size
3×3 and stride of 1, “Conv6” with 64 filters of size 3×3 and stride of 1, “Conv7” with 64 filters of
size 3×3 and stride of 1, “Conv8” with 64 filters of size 3×3 and stride of 1, “Conv9” with 64
filters of size 3×3 and stride of 1, “Conv10” with 128 filters of size 3×3 and stride of 1, which are
responsible for learning further discriminative features yielded the following output dimensions,
61×61×64, 30×30×64, 30×30×64, 15×15×64, 15×15×64, 7×7×64 and, 3×3×128 respectively.
3.3.3. Batch Normalization (BN)
The distribution of input data to internal layers changes as data flows through deep neural
networks during training due to changes in the network parameters. This affects the learning
capacity and accuracy of the network. To partially overcome this problem, we applied BN after
each regular convolution layer of the proposed CNN design with the exception of the first
convolution layer. The BN after each convolution layer normalizes elements in each feature map
to zero (0) mean and unit (1) variance before feeding it to the next layer while training, hence
accelerating training and increasing overall accuracy.
3.3.4. Activation Function
To introduce nonlinearity to neural networks, which greatly increases their capabilities of feature
representation, a convolution layer is usually followed by a non-linear mapping known as an
activation function. There is a variety of choice for an activation function such as tanh, sigmoid,
and relu, etc. however, we tested the performance of the proposed model using various activation
functions, combing more than one type of activation in some test cases, our investigation shows
that the proposed model works best while using elu activation which can be represented
mathematically by:
8. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
20
𝑋𝑗
𝑙+1
= 𝒇(𝑋𝑖
𝑙
) (4)
Where f(.) corresponds to the elu non-linear operation.
Similar to [73] and due to the fragile nature of feature maps extracted in early layers, our model
only introduce activation functions after the second convolution layer. Layers 2-12 uses the elu
activation and the softmax activation is applied after the last fully-connected layer (layer 13),
which is responsible for mapping features learned by the last FC layer to a set of probability
values, each corresponding to one class in the n+1 classes under consideration.
3.3.5. Pooling
Pooling operation is often applied after regular convolution layer in CNN to reduce the
dimension of the feature maps and to obtain a more compact representation of the input data.
Thus, reducing computational burden and chances of over-fitting. Max pooling and average
pooling are the most widely used pooling operations in CNN design. Max pooling does not take
into account all the activations within the pooling region, instead, it returns only the strongest
activation within the region. Such kind of activation works best for sparse feature representation
[33]. However, applying max-pooling in applications such as image forensics or steganalysis
where the weak signals are the signal of interest may lead to the loss of the desired information.
Therefore, throughout our proposed model, we used average pooling which takes into account all
the activations within the pooling region. It is also worth pointing out that since pooling
operations like non-linear activations are information losing processes [73], our CNN design
delayed the introduction of pooling layers until after the fifth convolution layer as depicted in
figure 3. Pooling operations suppresses the noise structures introduced by image manipulation
operations, which are the signal of interest in image tampering detection, leading to poor
performance. The Pooling operation used can be computed as:
𝑋𝑗
𝑙+1
= 𝒑𝒐𝒐𝒍(𝑋𝑖
𝑙
) (5)
Where 𝑋𝑖
𝑙
denote the j-th
feature map of the i-th
layer and pool (.) denote mean pooling operation
3.3.6. Implementation details of the Proposed General Purpose Image Tampering
Detection Method
All experiments were carried out using Open CV, Matlab and Tensorflow 2 on a machine with
NVidia GeForce GTX 1050 GPU with 8GB of memory. For training, we used Adam optimizer
with a learning rate of 0.0001 and the default values for the moments (β1 = 0.9, β2 = 0.999, and ε
= 10-7
). Xavier initialization was used to initialize the weights of layer 1 through 10 of the
proposed model with their biases set to 0. The three fully connected layers are initialized with
random numbers from a zero-mean Gaussian distribution with 0.01 standard deviation and their
biases were set to 0.05. We trained the proposed model in each experiment for 100 epochs
without shuffling of training data between epochs with a batch of 16 images for each training
iteration to minimize the categorical cross-entropy loss function to obtain best set of parameters
for the network.
Suppose 𝜃 is the parameter vector representing the weight vector corresponding to the image
tampering detection task, the categorical cross-entropy loss can be computed as:
9. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
21
𝐿(𝜃) =
−1
𝑀
∑ ∑ 1(𝑦𝑚
= 𝑛)log(𝑦𝑚
= 𝑛|𝑥𝑚
; 𝜃)
𝑁
𝑛=1
𝑀
𝑚=1
(6)
Where M and N denote the total number of image samples and the number of classes, 1(.)
represents an indicator function which equals 1 if m = n, otherwise 0. 𝑦𝑚
and𝑥𝑚
correspond to
the image label and the feature of the sample m. We minimizing the categorical cross-entropy
loss using the Adam optimizer with all the training samples to learn the network’s optimal set of
parameters. Using these learned parameters, the network can predict whether a given image is
tampered or not from the test samples.
Figure 3: Architecture of the proposed CNN for General Purpose Image Tampering Detection
10. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
22
4. EXPERIMENTS AND RESULTS
A series of experiments was carried out to assess the performance of the proposed method in
detecting image tampering operations. The first experiment evaluates the model’s performance
for detecting individual tampering operation, discriminating a real image from a tampered image.
Here a binary classification was carried out for each of the tampering operations showed in table
3.1 against the original image. In the other experiments, the models ability to perform multi-class
classification is evaluated with all the tampering operations listed in table 3.1. Next, we tested the
model’s performance on detecting multiple image tampering in images that have been subjected
to JPEG compression. The impact of change in input image sizes and activation functions on the
performance of the proposed model is also assessed. Finally, we compared the performance of
the proposed model with some of the existing state of the arts image tampering detection methods
from the literature. In the following subsections, we presents the proposed datasets as well as the
results of the proposed model for the various image tampering scenarios considered in this
section of the research.
4.1. Datasets
We consider five image tampering operations, Median Filtering (MF), Gaussian Blurring (GB),
JPEG Compression (JC), Gamma Correction (GC) and Contrast Enhancement (EC) as shown in
Table 4.1. We compare the performance of the proposed model with state of the art methods
from the literature for both binary and multiclass classification on image datasets consisting of
9605 images obtained from the union of Microsoft Common Object in Context (MS COCO)
[34], Bossbase datasets [35] and, IEEE image forensics [36] images databases. MS COCO is an
image database widely used for object detection and image segmentation. IEEE images is an
image database specifically created for image forensics challenges. The image database was
formed by randomly choosing 3555 mages from MS COCO, 1050 image from IEEE and the
remaining 5000 from Bossbase.
Table 1: Parameters used for creating experimental database
Tampering Operations Parameters
Gaussian Blurring Ksize= 3×3, 5×5
Gamma Correction ɣ = 0.5, 0.7, 0.8, 0.9
Median Filtering Ksize= 3×3, 5×5
Jpeg Compression QF = 70, 80, 90
Contrast Enhancement N/A
4.2. Experimental Results
4.2.1. Binaryclassification
In the binary classification, the proposed model is used to detect whether a given image is
original or tampered with one of the tampering operations listed in Table 1. Here the output layer
of the proposed model contains two neurons, each corresponding to the original or tampered
image respectively. Five experiments, each corresponding to that of one image tampering
operation in Table 1 was conducted. To perform these experiments, for each of the 9605 original
images, we generated 5 tampered versions using the tampering operations and the parameters
listed in Table 1. Thus, we have five different database of images, each containing 9605 images
and corresponding to one tampering operation in Table 1. We trained and evaluated the proposed
binary classifier using 14000 and 2000 subset of the dataset respectively for 100 epochs using the
implementation details of section 3.3.6 and kept the remaining 3000 for testing purpose. To
11. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
23
further assess the richness and discriminative abilities of features extracted by the proposed CNN,
we extracted deep features from the second dense layer of the proposed CNN and used it to train
Extra Tree and XGBOOST classifiers for the same binary classification task. The three trained
models were then used to predict the class labels for the testing images and the obtained
accuracies are shown in Table 2. From these results, it can be observed that the proposed CNN
can attain an accuracy of not less than 99.10 in detecting the different types of image tampering.
The proposed CNN obtained the highest accuracies of 100% in the detection of Gaussian
Blurring (GB) and JPEG compression (JPG). It can also be noticed that, with the Extra Tree
based CNN, there was a slight increase or equal performance in the detection rates of each
tampering when compared to that of the proposed CNN with the exception of GB detection in
which the proposed CNN was better by 0.6%. Similarly, the XGBOOST based CNN classifier
has improve the performance of the proposed CNN in GC and MF except in CE where the
proposed CNN was better off by 0.07. Overall, The ET based CNN performed best in detecting
individual image tampering operations among the three classifiers. Figure 4 shows the loss
against epochs and accuracy against epochs plots of the proposed CNN for individual image
tampering detection which further illustrate the performance of the proposed CNN model. These
results indicate that the proposed CNN can effectively extract the needed image tampering traces
for detecting different types of image tampering and converges just after few epochs.
Table 2: Proposed CNN, Extra tree and XGBOOST based CNNs detection accuracies (%) for individual
image tampering detection
Classifiers CE GB GC MF JPG
Softmax 99.10 100 99.90 99.46 100
Extra Tree Classifier 99.13 99.94 100 99.90 100
XGBOOST 99.03 100 99.92 99.94 100
(a)
12. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
24
(b)
(c)
(d)
13. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
25
(e)
Figure 4: Loss vs. epochs and accuracy vs. epochs of the proposed CNN for individual image tampering
detection of (a) Contrast Enhancement (EC), (b) Gaussian Blurring (GB), (c) Gamma Correction (GC), (d)
Median Filtering (MF) and (e) JPEG Compression (JC).
4.2.2. Comparison of the Proposed Binary Classifier against the State of the Arts
To further validate the performance of the proposed model on detecting individual image
tampering, we compare the results of the proposed model on binary classification with the work
of [2], [37], and, [38] from the literature on the same experiments. Table 3 present the accuracies
of the proposed model on individual image tampering in comparison with previous approach
from the literature. The numbers highlighted in bold show the best result obtained for that
tampering operation. The “-- “sign indicates the given method did not consider that tampering
operations in evaluating their model.
Table 3: Comparison of the proposed model accuracies (%) for individual image tampering detection
against the state of the arts
Methods CE GB GC MF JPG
Proposed 99.13 100 100 99.94 100
Bayer [2] -- 99.95 -- 99.71 99.66
Li [37] 99.95 100 96.76 99.99 99.94
Cozolina [38] -- 99.95 -- 99.60 99.86
As shown in Table 3, it can be noticed that the proposed method improves the
performance of [2] and [38] in GB, MF and JPG detections. Moreover, the result of the
proposed model on GC showed that it has improve the performance of [90] by 3.24%.
The only exception is in CE and MF where the method of [37] outperformed the
proposed method by 0.08% and 0.05% respectively. These results demonstrate that the
proposed method has better feature extraction and discrimination abilities, hence, it’s
more robust than the state of the arts in the majority of binary classification tasks
performed.
4.2.3. Multiclass classification
In the previous subsection, the proposed model was used to construct a binary classifier which
could effectively distinguish an original image from a tampered image. In this subsection, the
14. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
26
proposed method is used to construct a multiclass classifier that could identify more than one
image tampering type (all the tampering operations in Table 1). In this experiment, the output
layer of the proposed model will have six neurons, one corresponding to the original image and
the remaining five corresponding to the five tampering operations in Table 1 since it involves six
classes. Thus, the main difference between the binary classifier and the multiclass classifier used
in this work is the number of neurons each has in the output layer. To train the multiclass
classifier, we randomly divided our image datasets into three disjoint subsets, with 7000, 1000
and 1500 images, respectively to create the training, validation and testing images. Next, for each
image in the training set, we applied the five manipulations in Table 1 with their specified
parameters and centrally crop each image to 256x256. Consequently, for each image in the
training set, we now have 6 different types of images, the original image and its 5 tampered
versions. After processing all the images in the training set, we ended up with 42000 (6 x7000)
images in the training set. The LOOPs images of the training set were then computed and used as
input for the proposed network. The same processing steps applied to the training set were
replicated on the validation and test sets which generates 6000 validation and 9600 test sets used
for evaluating and testing the model. The proposed model was then trained and evaluated for 100
epochs using the implementation details described in section 3.3.6.
Similar to the binary classification of section in the previous subsection, we extracted deep
features from the second dense layer of the proposed CNN and used it to train and evaluated the
performance of two Tree based models (Extra Tree and XGBOOST) for the same multiple image
tampering detection. The three trained models were then used to predict the class labels for the
testing image and the obtained results in terms of CM (Confusion Matrix) and average detection
accuracies are reported in Figure 5 and Table 4 respectively. From these results, it can be noticed
that the proposed CNN can achieve an average detection rate of 99.81% in multiple image
tampering detection and can detect each tampering type with an accuracy of not less 98.27%.
Table 4 also shows the average detection accuracy of extra tree based CNN and XGBOOST
based CNN. Looking closely at the results, we can observe that each of the three classifiers could
detect multiple image tampering with an average accuracy of at least 99.43%. However, unlike
the binary classification, it can be noticed that the proposed CNN outperformed the extra tree
based CNN and XGBOOST based CNN by 0.38% and 0.21% respectively. These results
demonstrates the richness and discriminative strengths of the deep features extracted from the
images by the proposed CNN model. Moreover, the results also indicate that the proposed CNN
and LOOP model can effectively suppress the effect of image contents allowing it to extract the
different image tampering traces needed for identification of various image tampering. Figure 6
presents the loss vs. epochs and accuracy vs. epochs of the proposed CNN for multiple image
tampering detection, which further illustrate the performance of the proposed model.
Table 4: Average classification accuracies (%) of the proposed CNN, extra tree based CNN and,
XGBOOST based CNN for Multiple image tampering detection
Proposed Method Proposed CNN Proposed CNN
+ extra tree
Classifier
Proposed CNN
+ xgboost Classifier
Average Accuracies 99.81 99.43 99.60
15. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
27
Figure 5: Confusion matrixes of the proposed CNN model, extra tree based CNN and, XGBOOST based
CNN for Multiple image tampering detection
Figure 6: Loss vs. epochs and accuracy vs. epochs of the proposed CNN model for multiple image
tampering detection
4.2.4. Robustness against JPEG Compression
As another example of the robustness of the proposed method, we assess its effectiveness against
JPEG compression which is often used to conceal image tampering traces. To conduct this
experiment, the processed training, validation and test sets of the images used in section 4.2.3
where JPEG compressed with two quality factors 70 and 90 respectively and the training,
validation and testing of the models proceeds in a similar fashion to that of section 4.2.3 for the
70 and 90 JPEG compressed images respectively. Figures7 and 8 report the CMs of the three
classifiers when they were applied for multiple image tampering detection in 70 and 90 JPEG
compressed images respectively. The average detection accuracies of the three models for 70 and
90 JPEG compressed images in comparison to that of uncompressed images is provided in Table
5. It can be noticed from the results that the performance of the three classifiers decreased slightly
in the JPEG compressed images. However, the proposed model could still detect all the image
tampering with an accuracy of at least 97.13% for 70 and 97.20% for the 90 compression
qualities respectively, indicating the robustness of the proposed model against JPEG
compression. Although the proposed models performed excellently even in the JPEG compressed
images, our results also indicate that JPEG compression has the effect of suppressing the different
traces introduced by image tampering operations, thus, adversely affecting the overall
performance of the proposed models.
16. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
28
Table 5: Average classification accuracies (%) of the proposed CNN, extra tree based CNN and,
XGBOOST based CNN for multiple image tampering detection in uncompressed and compressed image
formats.
Proposed Method Proposed CNN Proposed CNN
+ Extra Tree
Classifier
Proposed CNN +
XGBOOST
Uncompressed 99.81 99.43 99.60
70 compressed 99.01 99.08 99.21
90 compressed 98.93 98.72 99.24
Figure 7: confusion matrixes of the proposed CNN model, extra tree based CNN and, XGBOOST based
CNN for Multiple image tampering detection in compressed image (Q = 70)
4.2.5. Impact of Different input Image Size and Activation Functions
In the previous experiments, we have shown the robustness of the proposed model in detecting
multiple image tampering in uncompressed and compressed images. In this subsection, we
investigate the impact of different input image size and activation functions on the performance
of the proposed models. To assess the impact of different input image size, we have trained the
proposed models with two input image sizes i.e. 256 and 128 using the same experimental
settings of section 4.2.3 for multiple image tampering detection. The average detection
accuracies obtained are reported in Table 6.
Figure 8: confusion matrixes of the proposed CNN model, extra tree based CNN and, XGBOOST based
CNN for Multiple image tampering detection in compressed image (Q = 90)
17. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
29
From the results, we can notice that the average detection accuracies of the three models
decreases with a decrease in the image size due to reduction in the statistical samples that could
be extracted by the proposed CNN. However, the tree based CNN still performs excellently with
an image size of 128. These results implies that, the choice of input image size may affect the
performance of the proposed model. Larger input image size may improve performance at the
expense of computational time, whereas smaller input image size may reduce computational time
but at the expense of performance.
In another similar experiment, we investigate the impact of different activation functions on the
performance of the proposed model for multiple image tampering detection. The proposed model
was trained with two other activation functions (i.e. relu and tanh) using the same datasets and
experimental settings of section 4.2.3. The results obtained by the relu and tanh based network in
comparison with the proposed elu based network are provided in Table 7. From the results, we
observed that the proposed elu network outperformed the relu and tanh based network by 0.25%
and 0.47% respectively. Figure 9 shows the loss and accuracy plots obtained by the tanh, elu and
relu based networks respectively.
Table 6: Average classification accuracies (%) of the proposed CNN, extra tree based CNN and,
XGBOOST based CNN for different input image sizes
Image size 256 128
Proposed CNN 99.81 97.54
CNN + Extra Tree 99.43 98.67
CNN +XGBOOST 99.60 99.28
Table 7: Average classification accuracies (%) of the proposed CNN, extra tree based CNN and,
XGBOOST based CNN for different activation functions
Activation Tanh Relu elu
Proposed CNN 99.34 99.56 99.81
CNN + Extra Tree 99.43 99.31 99.43
CNN +XGBOOST 99.42 99.48 99.60
(a)
(b)
18. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
30
Figure 9: Loss vs. epochs and accuracy vs. epochs of the proposed CNN model for multiple image
tampering detection using (a) relu, (b) elu and (c) tanh activation functions.
4.2.6. Limitation Analysis
Although the proposed image tampering detection method is robust against JPEG compression,
our experiments revealed that JPEG compression may slightly degrades the performance the
proposed model as it has the effect of suppressing the different traces introduced by image
tampering operations. Similarly, from our experiments, it was also observed the performance of
the proposed model decreases with a decrease in the image size due to reduction in the statistical
samples that could be extracted by the proposed CNN. Larger input image size may improve
performance at the expense of computational time, whereas smaller input image size may reduce
computational time but at the expense of performance.
5. CONCLUSIONS
In this paper, we have proposed a novel general purpose CNN and LOOP based image tampering
detection method. Unlike existing approaches that use hand designed features or constrained pre-
processing layer, the proposed method can directly extract image tampering traces directly from
the LOOP images, which has the effect of suppressing the effect of image content allowing the
proposed CNN to capture the different traces needed to detect different type of image tampering.
To assess the performance of the proposed model, we have tested it through a number of
experiments and the results of these experiments demonstrate the effectiveness of the proposed
model in detecting individual as well as multiple image tampering. To further assess the
performance of the proposed model, we have compared its results with that of the state of the arts
from the literature and these results also show that the proposed model could achieved a
competitive performance. In future, we plan to extend the proposed method to detect anti-
forensics image tampering operations as well as more image tampering operations.
ACKNOWLEDGEMENTS
This research was supported by National Information Technology Development Agency
(NITDA) of Nigeria.
REFERENCES
[1] Redi, J. A., Taktak, W., and Dugelay, J. L., (2011) “Digital image forensics: a booklet for
beginners”,Multimedia Tools and Applications, 133-62.
(c)
19. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
31
[2] Bayar, B., and Stamm, M.C., (2018) "Constrained convolutional neural networks: A new approach
towards general purpose image manipulation detection", IEEE Transactions on Information Forensics
and Security 13, no. 11, 2691-2706.
[3] Bayar, B., and Stamm, M.C., (2016) “A deep learning approach to universal image manipulation
detection using a new convolutional layer”, In Proceedings of the 4th ACM Workshop on
Information Hiding and Multimedia Security, 5 -10.
[4] Baby, L., and Jose, A., (2014) “Detection of Splicing in Digital Images Based on Illuminant
Features”, International Journal of Innovation and Scientific Research, 11(2351-8014), p.4
[5] Li L. L., Xue, J. Wang, X., and Tian, L., (2013) “A robust approach to detect digital forgeries by
exploring correlation patterns”, Pattern Anal Appl 1–15.
[6] Carvalho, T., Faria, F. A., Pedrini, H., Da R., Torres, S., and Rocha, A., (2016) ‘‘Illuminant-based
transformed spaces for image forensics’’, IEEE Trans. Inf. Forensics Security, vol. 11, no. 4, pp.
720–733.
[7] Qiu, X., Li, H., Luo, W., and Huang J., (2014) “A universal image forensics strategy based on
steganalytic model”, In Proceedings of the 2nd ACM workshop on Information hiding and
multimedia security, (pp. 165-170), ACM (2014 Jun 11).
[8] Sundus, F., Yousaf, M. H., and Hussain, F., (2017) "A generic passive image forgery detection
scheme using local binary pattern with rich models", Computers & Electrical Engineering 62 (2017):
459-472.
[9] Rao, Y., and Ni, J., (2016) “A deep learning approach to detection of splicing and copy-move
forgeries in images”, In IEEE International Workshop on Information Forensics and Security (WIFS)
(pp. 1-6), IEEE.
[10] Yue, W., AbdAlmageed, W., and P. Natarajan, P., (2019) "ManTra-Net: Manipulation tracing
network for detection and localization of image forgeries with anomalous features", In Proceedings of
the IEEE Conference on Computer Vision and Pattern Recognition, pp. 9543-9552.
[11] Zhou, P., Han, X., Morariu, V. I., and Davis, L. S., (2018) "Learning rich features for image
manipulation detection," In Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, pp. 1053-1061.
[12] Boroumand, M., and Fridrich, J., (2018) “Deep learning for detecting processing history of images,”
Society for Imaging Science and Technology.
[13] Chen, Y., Xiangui, K., Yun, Q. S., and Jane Wang, Z., (2019) "A multi-purpose image forensic
method using densely connected convolutional neural networks," Journal of Real-Time Image
Processing 16, no. 3 pp. 725-740.
[14] Zhan, Y., Chen, Y., Zhang, Q., and Kang, X., (2017) “Image forensics based on transfer learning and
convolutional neural network,” In Proceedings of the 5th ACM Workshop on Information Hiding and
Multimedia Security (pp. 165-170).
[15] Qiu, X., Li, H., Luo, W., and Huang, J., (2014) “A universal image forensic strategy based on
steganalytic model,” in Workshop on Information hiding and multimedia security, ACM, pp. 165–
170.
[16] Pevny, T., Bas, P., andFridrich, F., (2010) “Steganalysis by subtractive pixel adjacency matrix,” IEEE
Transactions on Information Forensics and Security, vol. 5, no. 2, pp. 215–224.
[17] Fridrich J., and J. Kodovsky, J., (2012) “Rich models for steganalysis of digital ` images,” IEEE
Transactions on Information Forensics and Security, vol. 7, no. 3, pp. 868–882.
[18] Fan, W., Wang, K., and Cayre, F., (2015) “General-purpose image forensics using patch likelihood
under image statistical models,” in IEEE International Workshop on Information Forensics and
Security, pp. 1–6.
[19] Sundus, F., Yousaf, M. H., and Hussain. F. (2017) "A generic passive image forgery detection
scheme using local binary pattern with rich models", Computers & Electrical Engineering 62, 459-
472.
[20] Chen, B., Li, H., and Luo, W. (2017) “Image processing operations identification via convolutional
neural network”, arXiv preprint arXiv: 1709.02908.
[21] Cao, G. Zhou, G. A., Huang, X. Song, G. et al, (2019) “Resampling detection of recompressed
images via dual-stream convolutional neural network,” arXiv preprint arXiv: 1901.04637.
[22] Zhan, Y., Chen, Y., Zhang, Q., and Kang, X., (2017) “Image forensics based on transfer learning and
convolutional neural network”, In Proceedings of the 5th ACM Workshop on Information Hiding and
Multimedia Security (pp. 165-170).
20. Signal & Image Processing: An International Journal (SIPIJ) Vol.12, No.2, April 2021
32
[23] Mazumdar, A., Singh, J., Tomar, Y.S., and Bora, P. K., (2018) “Universal Image Manipulation
Detection using Deep Siamese Convolutional Neural Network”,arXiv preprint arXiv: 1808.06323.
[24] Boroumand, M. and Fridrich, J., (2018) “Deep learning for detecting processing history of images”,
Society for Imaging Science and Technology.
[25] Wu, Y., Abdalmageed, W., and Natarajan, P., (2019) “ManTra-Net: Manipulation Tracing Network
for Detection and Localization of Image Forgeries With Anomalous Features”, In Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition (pp. 9543-9552).
[26] Chen, Y., Xiangui, K., Yun, Q. S., and Jane Wang, Z., (2019) "A multi-purpose image forensic
method using densely connected convolutional neural networks", Journal of Real-Time Image
Processing 16, no. 3 pp. 725-740.
[27] Zhang, R., and Ni, J., (2020) "A Dense U-Net with Cross-Layer Intersection for Detection and
Localization of Image Forgery", In ICASSP 2020-2020 IEEE International Conference on Acoustics,
Speech and Signal Processing (ICASSP), pp. 2982-2986.
[28] Chakraborti, T., McCane, B., Mills, S., and Pal, U., (2018) “Loop descriptor: Local optimal-
oriented pattern”,IEEE Signal Processing Letters, 25(5), pp.635-639.
[29] Ojala, T., Pietikinen, M. and Harwood, D., (1994) “Performance evaluation of texture measures with
classification based on Kullbackdiscrimination of distributions”, In Proc. ICPR.
[30] Jabid, T., Kabir, M. H. and Chae, O.S., (2010) “Gender Classification using Local Directional Pattern
(LDP)”,In Proc. ICPR.
[31] RemyaRevi K., and Wilscy, W., (2020) “Image forgery detection using deep textural features from
local binary pattern map”, Journal of Intelligent & Fuzzy Systems, (Preprint), pp.1-11, 2020.
[32] Surbhi, S. and Ghanekar, U., (2019) "Spliced Image Classification and Tampered Region
Localization Using Local Directional Pattern", International Journal of Image, Graphics & Signal
Processing 11, no. 3.
[33] Qian, Y., Dong, J., Wang, W. and Tan, T., (2015) “Deep learning for steganalysis via convolutional
neural networks”, In Media Watermarking, Security, and Forensics 2015International Society for
Optics and Photonics, Vol. 9409, p. 94090J.
[34] Lin, T. Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., et al... (2014) “Microsoft
COCO: Common objects in context”, InECCV, 2- 5.
[35] Bas, P., Filler, T. and Pevny T., (2011) “Break our steganographicsystem: The ins and outs of
organizing boss,” In Information Hiding, pp. 59-70.
[36] IEEE IFS-TC image forensics challenge. IEEE Trans Inf Forensics Secur 2013.
[37] Li, H., Luo, W., Qiu, X. and Huang J., (2016) “Identification of various image operations using
residual-based features”, IEEE Transactions on Circuits and Systems for Video Technology, 28(1),
pp.31-45.
[38] Cozzolino, D., Giovanni, P. and Luisa V., (2017) "Recasting residual-based local descriptors as
convolutional neural networks: an application to image forgery detection", In Proceedings of the 5th
ACM Workshop on Information Hiding and Multimedia Security, pp. 159-164.