An efficient circular block approach For copy-move forgery detection

Abstract

Due to availability of powerful retouching or editing software tools now-a-days it is very easy to tamper any type of digital images. That's why it has been very common to add or remove anything from an original image which causes the lead of digital image forgery. Several types of digital forgery may be happened but copy-move forgery is di cult to detect by our naked-eyes. Copy-move forgery is a special type of digital forgery in which a part of the image is copied and pasted somewhere else in the image with the intent to cover an important image feature. To detect this type of forgery we need a robust detection method which ensures the correct detection even if the image is noised, compressed, scaled, rotated, ipped etc. Several methods has already been proposed but none methods are suitable for all kinds of robustness. Some methods are showing good performance to approx- imate match or noised or compressed but fails towards scaled or rotated. So still no method can ensure the detection of the forged image in any types of challenges with compatible time performance. We have gone through the existing methods and found out their limitations. We have also analyzed their comparison to nd out their comparative performance. In this thesis paper, we propose an e cient method for detecting the copy-move forgery using circular block extraction and calculating the mean, contrast of the image which is robust in rotation, ipping, JPEG compression, blurring, noise etc. The main success of the proposed method is the robustness in JPEG compression, blurred image and rotation in any an- gels. Basically for using circular block approach we are getting rotation invariant detection method. Another thing is that we are comparing the blocks up to nth consecutive blocks for which our method can show better performance for JPEG compression with low quality factor and also applicable for blurring. The perfor- mance of the proposed method has been evaluated with di erent challenges like Gaussian noise, Gaussian blurring, rotation, ipping, JPEG compression etc. as well as the existing methods. For dataset we consider the benchmark image set so that we can get the real strength of our proposed method.