Detection of LSB Matching Steganography in Decompressed Images

Studies show that no detectors for LSB matching have yet proven universally reliable and it is hard to predict which types of image are suitable for a specific steganalyzer. For the kind of decompressed images, this paper presents an efficient steganalyzer that exploits the fact that the noise residuals in the DCT domain are rather concentrated on zero and very sensitive to LSB matching. Experimental results show that it is almost perfect at embedding rate 0.5 bpp and that it is the accuracy of 90.9% at 0.1 bpp against the accuracy of 44.6% with the WAM steganalyzer. However, the proposed detector works only as long as the exact JPEG decompressor is known.