An improved first quantization matrix estimation for nonaligned double compressed JPEG images

Abstract

Estimation of the first quantization matrix of a nonaligned double compressed JPEG (NA-DJPG) image remains a challenging problem. To solve this problem, a novel method for estimating the original quantization steps of NA-DJPG images is proposed in this paper. Our method is separated into two phases: estimation of the misalignment parameters and estimation of the quantization steps. In the first phase, since the texture of the image has great influence on the traces of the first compression grid, an appropriate block selection strategy is proposed that takes the interference of texture into consideration, and this is followed by a differential boundary analysis to estimate the misalignment parameters between the first and second compression grids. In the second phase, a DCT coefficient histogram rearrangement strategy is first proposed to make the periodicity of the histogram more apparent. This is followed by a histogram frequency spectrum analysis to estimate the period of the histogram, which is close to the first quantization step. To further suppress the interference of the second compression, several quantization step candidates around the rounded period value are selected, and finally, the quantization step for each band is estimated via a comparison strategy of periodic bin accumulation. The experimental results and a comparison with existing methods demonstrate the effectiveness and superiority of the proposed method.