Abstract: Aiming at the problem that the insufficient extraction of feature points in existing forgery detection methods leads to low accuracy of forgery detection and poor recognition rate of feature points descriptor, a color image copy-move forgery detection algorithm based on color moment region division and quaternion Hu moment is proposed. Firstly, an adaptive morphological reconstruction algorithm is adopted to perform superpixel segmentation on the image, and then a density clustering algorithm is used to adaptively divide the image into regions. Secondly, a key point extraction method is proposed to obtain uniform SIFT feature points, and then a local Gaussian pyramid is constructed in a novel color image quaternion representation method to extract the Hu moment features. Finally, after matching features using the 2NN, the paper proposes to locate the copy-move forgery region by the Delaunay triangle algorithm. Experimental results on public datasets show that this method can effectively locate the forgery region.

Key words: copy-move, region division, SIFT, Hu moment, Delaunay triangle algorithm