Article Preview
Top1. Introduction
Digital images are being edited deliberately or involuntarily to make them more informative or to hide some content in the image. The vast growth of commercial and open source digital photo editing tools leads to the increase of tampered images in day-to-day life. The trustworthiness of digital image plays a major role in many applications, viz., criminal examination, journalism, forensic analysis and surveillance systems (Mahdian & Saic, 2010). A beginner in digital forensics area can refer to its various applications in (Li, 2013). Digital Image Forgery (DIF) detection is plausible in two approaches (Hashmi & Keskar, 2015), viz., Active and Passive. Active approach involves pre-processing of a genuine image by embedding an identifier before it is used. Watermarking and signature embedding technologies are active approaches useful for detection and localization of image forgery but pre-processing of digital data limit its usage. Passive method (Al-Qershi & Khoo, 2013) explore statistics or features from an image for CMFD.
Copy-Move type of forgery is one in which some snippet of an authentic image is copied and pasted within the image with an intent to hide a specific content in the image. The pasted portion relates to authentic image, hence it affects statistical properties of the image and these variations are explored to detect the forgery. It is clear from Figure 1, that the forgery does not leave any visual clue to identify the tampering. CMF problem can be addressed in two ways, i. Localization and ii. Detection. The localization process recognizes at which locations the image is being tampered whereas the detection process classifies whether the given image is forged or not.
Figure 1. Copy-move forgery: (a) CMF image; (b) Original image
The CMF localization process is shown in Figure 2 and it focusses on extracting the features from the overlapping blocks of the suspicious image in the block-based methods. In the case of key-point based methods, it explores the key-points, i.e. high-entropy regions of the image. Feature matching is performed to identify similar blocks or key-points in the given image. These matched regions are considered as potential blocks of forgery.
The CMF detection process is illustrated in Figure 3. The detection process involves extracting features from all the images in the dataset. A suspicious image is tested against the trained set to confirm whether it is an original or forged image.
The detection process seems to be uncomplicated from Figure 3 and it is true for simple copy move forgeries. However, CMF images can be affected by various post-processing attacks viz., JPEG compression, blurring, noise addition, and color reduction, etc. Detecting the images under these attacks is critical and several methods are available in the literature which can identify simple CMF images. Authors in (Shen et al., 2017) developed an image splicing detection method using textural features from Gray Level Co-occurrence Matrix (GLCM) but it has not concentrated on CMFD.
Another work developed by (Suresh & Srinivasa Rao, 2016) based on GLCM texture features for CMFD is available. It needs 22 statistical features and these are calculated in one direction. Even though other methods are able to handle complicated forgeries, but they are computationally expensive. Moreover, the feature set of an image plays a critical role in the classification.