Markov features based DTCWS algorithm for online image forgery detection using ensemble classifier in the pandemic

Abstract

The unexpected blast of the COVID has been found everywhere in the world, and everyone has been shocked unusually. In this critical situation, the number of online activities increases, whether related to online education, research, business meetings, virtual conferences, or virtual court. Under this pandemic situation, digital images are the only source of information that can be generally shared and visualized in virtual conferences and social media, and it's challenging to share the document for forgery detection. Today, it's straight forward to forge these images using image-editing software, and it's essential to detect image forgery for such images. In this paper, an efficient novel Discrete-Time Cosine Wavelet and Spatial (DTCWS) Markov feature-based algorithm has been designed for the detection of such forgery, especially for this pandemic situation. For this work, high-dimensional Markov features have been extracted in the DTCWS domain, and the dimensionality of these Markov features has been reduced with Principal Component Analysis (PCA). Furthermore, the co-occurrence matrix has increased the correlation among coefficients. For classification, an optimized ensemble classifier is used for evaluating the results instead of using a support vector machine classifier. Due to the time constraint in online activities, the proposed algorithm shows the best accuracy of 99.9% without taking too much time and fewer complexes compared to the current work.

Introduction

COVID-19 has been declared as a pandemic situation in this world in 2020 that has never been ever seen in this world in the last ten decade. Under this situation, in the starting days, nobody can come out from his or her homes, nobody can meet others, nobody can share documents with each other, and it is essential to maintain social distancing (Battiato & Messina, 2009). Due to this, people are bound to do day-to-day activities in their homes, whether it's related to school education, related to attending professional meetings for their job, or related to research, training, and virtual conferences (Bishop, 2006). Even courts are not open, and the trend of the virtual court has been seen in this pandemic situation. In virtual courts, the lawyers have to produce their evidence online from their homes. Therefore, there is the maximum chance to make forgery in the court evidence (Dong, Wang, Tan, & Shi, 2009).

In the same way, the culprit can use these forged images in research, training, and online exams very easily. The motivation behind this research is the number of growing cases of forgery due to the online activities in an unusual way in this pandemic situation. Therefore, image forgery in video conferences is the leading research problem, and the proposed method is the best solution to overcome this problem (Cox, Miller, Bloom, Fridrich, & Kalker, 2008). Thus, the brief discussion of image forgery and the proposed solution has been discussed below.

Due to the advancement in imaging tools, image forgery has been increased day by day, and it is very important to develop an efficient image forgery method, especially under this pandemic. A number of related methods have been developed previously and classified as active and passive forgery methods (Bhatia, 2021, Bhatia et al., 2019). In the active method, prior information of the image is required for the digital signature and watermark, and a watermark or signature is embedded in the image to check the originality of the image, but in passive methods, no prior information is required, and no other acquisition devices are required. In passive methods, traces of forgery are being used (Cheah et al., 2021, Khare et al., 2020).

Digital image forensics has many types, but digital image splicing forgery plays an essential role in this pandemic situation. This technique is also a very commonly known technique in image forensics (Fawcett, 2006). In this, the portion of the image is copied from the original image and pasted on another image, then resave the image for making the forged image, which is called a spliced image. The image splicing forgery can be seen in the different sectors where images are being used as a document-proof in different areas like video conferencing, criminology, insurance, education, research, law enforcement, and many more (Chen, Shi, & Su, 2007). This splicing forgery can be detected by the extraction of efficient features, which can reduce the redundancy of the information and reduce the information loss during splicing operation (Kouadri et al., 2020). For example, in September 2020, the most audacious case has been published in the Times of India under COVID-19 that adoctor got arrested for making the forged COVID reports(TOI-News, 2020). On October 10, 2020, the news has been published in timenownews.com for document forgery case in Tamilnadu, where forged documents have been produced in the virtual court by the lawyers for availing bails for their clients, and the lawyers got arrested for their forgery (Timenownews.com, 2020). One more case under sports arrived for cricket ball-tampering on April 25, 2020. In this, the international cricket council legalized ball-tampering by allowing the artificial substance to the ball for shining instead of using saliva on the ball that may cause the COVID virus to spread (TheHinduBusinessLine.com, 2020). On December 20, 2020, the wrong news had been published in India lokmat newspaper with a forged image of corona vaccine and claimed that the vaccine Pfizer is made in wrong china and found that this vaccine is discovered in the UK by NYS lab research as see in Fig. 1 (Altnews.in, 2020).

Now a days, in a paperless environment, it has become very popular to manipulate images without leaving perceptual clues due to the availability of millions of pictures on the internet for e-governance and personal benefits. These types of image forgery may also force innocent people in the direction of crime in image forensics areas (Aggarwal, Bhamrah, & Ryait, 2019). Therefore, this is the prime concern for developing automatic photo tampering techniques for the detection of such forged images. Hence, a discrete-time cosine wavelet and spatial domain (DTCWS) based algorithm has been proposed todetect such forgeriesin this COVID-19 pandemic (Han et al., 2016, Mohdiwale et al., 2021).

The novelty of this DTCWS based algorithm liesin the extraction of domain wise Markov features, which includes coefficient wise Markov features with spatial domain, block-wise Markov features with Discrete Cosine Transform (DCT) domain, and multi-resolution wise Markov features with Discrete Wavelet Transform (DWT) domain (Hsu et al., 2007, Kumar et al., 2019). The first one shows the correlation among coefficients in the whole image; the second one shows the correlation among coefficients with neighboured values in a successive manner. The third one shows the multi-resolution information of the textured coefficient in the whole image.By concatenating, all these three domain Markov features, the global feature vector has been obtained (Mehta and Agarwal, 2018, Mehta and Agarwal, 2018). With this combination, the dimensionality of features has been automatically increased. To reduce the time constraint and to increase the efficiency of the algorithm, this proposed algorithm needs to reduce the dimensionality of the features with Principal Component Analysis (PCA), and this process definitely increases the speed of the algorithm. With PCA, the maximum variance Eigenvalue and Eigenvector has been calculated, and then, correlation among coefficients has been enhanced with co-occurrence matrix (Kodovsky et al., 2012, Saba et al., 2020). Further, these highly correlated features are fed to the optimized ensemble classifier for the decision-making of authentic and spliced/forged images.