Double JPEG compression detection based on block statistics

摘要

In this paper, a novel method is proposed to detect the aligned double JPEG compression with different quantization matrix. The proposed method is based on the theory that the correlation among adjacent coefficients of frequency spectrum in DCT blocks is enhanced after DCT transformation, and the correlation among same locations in adjacent DCT blocks is strong. Classification features are divided into two types, the intra-block frequency domain features and the inter-block frequency domain features. The intra-block frequency domain features are used to catch the strong correlation among adjacent coefficients in DCT blocks, and the inter-block frequency domain features are used to catch the correlation among same locations of adjacent DCT blocks. Then the intra-block frequency domain features with inter-block frequency domain features are combined as the classification features. Finally, the classification features are used to train classifiers to detect double JPEG compression. The results of extensive experiments demonstrate the effectiveness of the proposed method.