Noise reduction approach for LSB matching revisited

摘要

This paper proposes a new approach to LSB Matching Revisited (LSB-MR), a steganographic method for embedding message bits into a still image. Our method uses binary images as messages. Instead of embedding a full of message bits, the modified method just uses a half of message bits for embedding. In extracting phase, after reconstructing the embedding part, the remain bits of message (without embedding) are generated with an small error ratio. Most of error bits (missing bits) are restored by Median denoise filter. Therefore, the proposed method allows embedding the same payload as LSB-MR but with fewer changes to cover image. The experimental results, evaluated on 1338 uncompressed images, show that the new method has better performance, compared to LSB-MR, in terms of cover image quality and resistance against existing steganalysis.