The efficiency of applying DWT and feature extraction into copy-move images detection

The paper presents the efficiency of applying DWT into copy-move image forgery detection by developing an algorithm which combines DWT and feature extraction to improve the computational time compared to the algorithm without DWT. With the characteristic of invariant to rotation, Zernike Moments are used to extract the features of image blocks. The novelty of this article is not only combination of multiscale and features extraction but also the modification of parameters of Zernike moments in the proposed algorithm. The tested image is reduced dimension by DWT before looking for the similar regions as traces of copy-move forgery manipulation. Upon the principle that most of forged information concentrate at the low frequencies, the approximation sub-band (LL) is considered for detection. This band is then split into 16×16 overlapping blocks from which the modified Zernike moments are extracted to be block feature vectors with higher exactness than the traditional Zernike moments. These vectors are arranged into matrix and sorted lexicographically to find the similar vectors from group of consecutive vectors having correlation coefficients of 0.95. The fact that neighbor blocks may be similar and the copied regions can be comprised by many blocks and requires a distance to make sure that they are really similar, not neighbors. Simulation results running in Matlab R2013a proves the feasibility and efficiency of the proposed algorithm.