Invariances of JPEG-quantized DCT coefficients and their application in robust image steganography

摘要

The JPEG compression is employed in many image-related fields, and the characteristics of DCT coefficients before and after JPEG compression are always one of the research hotspots. In this paper, we discover a characteristic by analyzing the image compression and decompression process, and we name this characteristic the "nonzero invariance of JPEG-quantized DCT coefficients." Based on this, we deduce another characteristic, "position invariance of the last nonzero JPEG-quantized DCT coefficients." This characteristic is then successfully applied to robust image steganography, a research field dedicated to achieving covert communication on lossy channels. A method resisting JPEG compression and statistical detection is proposed. This method is robust to JPEG compression attacks by skillfully utilizing the "position invariance of the last nonzero JPEG-quantized DCT coefficients." Moreover, a novel distortion function is proposed. It is designed by searching for the most suitable embedding position using particular rules. The searching scope is defined between the last two nonzero coefficients to balance good antistatistical detection performance and image details retention. Particular rules are constructed with position parity of coefficients to reduce limitation on embedding positions. Experimental results on three open image databases show the high reliability of two characteristics and the robustness of the proposed method in JPEG compression and statistical detection.