FORENSIC ANALYSIS OF CLONED REGIONS OF DIGITAL IMAGES USING FEATURE BASED METHODS

Abstract

Digital technologies have gained immense attention of the researchers recently due to extensive use of digital techniques in everyday life. Our government has also started programs like Digital India, which has also enhanced the use of Digital media and mostly Digital images in everyday life and thus its use in forensics has also increased tremendously. Digital images are quite simple to tamper because advanced image processing and editing tools are easily available today. Today, adding or removing important features from an image without leaving any visible mark that the image has been tampered is very easy, which is a serious social issue as well as it is used by criminals to alter the important features of the image so that they can escape punishments easily. As information processing in digital domain is replacing analog processors and the digital image and video cameras replacing analog ones, the requirement to authenticate the digital images to legitimize their content, and detecting copy move forgery is only increasing. In this thesis, a special type of widely used image tampering method i.e. cloning, in which a single or multiple portions of an image are copied and then they are moved somewhere else in the same image and pasted there with the motive to hide any important detail of the image or to replicate the given feature many times in the image of interest. Image forgery is the technique of detecting image modification, either with the previous information about the source image (active) or without (passive). Further the images can be scaled, rotated, shifted or flipped, thus detection of the forgery in the image is further made difficult. In comparison to the earlier studies, our thesis1gives a better representation as well as comparison, challenges and future work in a well organized manner.

In this thesis, we study a new feature based technique which when combined with clustering, SWT, RANSAC will provides us better results. Following techniques are used to detect the cloned image and localization of the cloned areas of the image which describe the thesis in a nutshell. (v)

Cloning detection is a block based technique which use extensive search to detect the cloned portions of the image. There are 2 block based search algorithms that can be used in this technique, which are Exact Match and Robust Match. Exhaustive search is done after computing the autocorrelation of the given image and then it is ready for application of various feature based and non feature based techniques which are described below:

 DCT based technique: In this a sliding window of 16X16 dimension is used to compute the DCT matrix of the image whose coefficients are quantized and inserted in a row which is lexicographically sorted. Rows that match exactly will give the location of the cloned portions of the image.

 SIFT based technique: In this SWT is computed to preprocess the image then SIFT is computed on the approximate sub band produced during SWT. The 4 steps used to calculate SIFT and then key points are extracted and then matching and clustering of the extracted key points is done and at last RANSAC is used to remove false positives in this image. This technique will also provide the cloned regions but is better than DCT because it can be used for rotated rescaled clones.

 MIFT based technique: In this SWT is computed to preprocess the image same like we did in SIFT to compute the approximate sub band. The 4 steps used to calculate MIFT which generates quality vectors with the help of their neighborhood and then descriptor is formed and then matching and clustering of the extracted key points is done and at last RANSAC is used to remove false positives in this image. This technique will also provide the cloned regions but is better than DCT because it can be used for rotated rescaled clones.