Image Deeper Passive Forgery Detection Based on SIFT and Higher-Order Statistics Features

摘要

In this paper, an image deeper forgery detection approach based on combined SIFT(scale invariant features transform) feature and Higher order Statistics, bi-coherence amplitude and phase features, is presented. For the foundation of image forgery detection based on SIFT feature, a deeper analysis based on bi-coherence features, distinguishing between the original source block of an image with its forgery counterpart, is performed. Firstly, SIFT keypoints features are extracted and matched by Euclidean distance, and all matched keypoints are clustered to two blocks which constitute to the two sub-images, respectively. Subsequently, the bi-coherence amplitude and phase features of the two sub-images are estimated. Our experiments are launched on three datasets corresponding to three image formats: BMP, PNG and JPEG. On Cal Photos dataset, the quantitative distribution of bi-coherence features has been estimated and on the other two datasets, the performance analysis of forgery detection and distinguishing has been provided. Experimental results demonstrate the efficiency of the method.