JPEG 量化步长估计综述

As JPEG (Joint Photographic Experts Group) format images are widely used in network storage and transmission, malicious operations such as tampering and stitching based on the JPEG format have emerged, and the research on exploiting the characteristics of JPEG images has received more and more attention. Among them, JPEG image characteristics are closely related to quantization steps. If the JPEG compressed image is saved in the bitmap form, the researcher cannot get the quantization step directly and thus cannot analyze the JPEG image characteristics effectively. Therefore, the quantization step estimation is attracting more attention in the field of digital image forensics. This paper first introduces the background of quantization step estimation and divides the problem into two issues: quantization step estimation for single compressed images and first quantization step estimation for recompressed images. Secondly, based on the classical JPEG compression model, the quantization step estimation models for these two issues are given. Various methods for each type of problem are introduced and sorted out in detail. Then, the classical algorithm experiments in the same experimental environment, and the classical algorithm is analyzed and evaluated. The experimental results show that in the quantization step estimation task of single compressed images. The existing techniques are mature, and the estimation accuracy of some algorithms is higher than 90%; however, in the first quantization step estimation task of aligned recompressed images, the current methods can only estimate the quantization steps of low and medium frequencies, and the estimation accuracy is lower when the first and second compression quality factors are similar; in the first quantization step estimation task of nonaligned recompressed images, the grid offset increases the difficulty of estimating the quantization step, resulting in poorer performance of the algorithm on nonaligned recompression. Finally, the problems to be further solved and the future development trends in the quantization step estimation task are pointed out.