Image forgery detection using image similarity

Abstract

Ideally, sophisticated image forgery methods leave no perceptible evidence of tampering. In response to such stringent context, researchers have proposed digital methods to detect such indiscernible tampering. In this paper, we present a blind image forgery detection method that uses a steerable pyramid decomposition technique and copulas ensemble. This method can accurately detect forgery in regions as small as 16 pixels, which is the smallest size reported in the literature with perfect accuracy. The proposed method is innovative in that: (i) it works on both grey scale images as well as colored images; (ii) the copula functions are used to calculate image similarity (or dissimilarity) which represents image forgery; (iii) the precision of the copula results on the image steerable pyramid bands motivated the idea of selecting the band with minimum number of elements to represent the block(s) in the image, which is 16 elements, in our case. The idea of using smallest number of elements to represent the blocks can significantly speed up the method as the testing is done on such small number of pixels; finally (iv) this method can be applied to more than one kind of image forgery with similar results. To verify the performance of the proposed method, we tested it on the well-known Copy Move Forgery Detection database (CoMoFoD) using 5123 image variations of the database. Also, we compared our results with five previously published algorithms and found that the proposed method outperformed those algorithms even when the forged images were subjected to postprocessing manipulations and transformations.