SW: a blind LSBR image steganalysis technique

Abstract

Blind image steganalysis is exploring body of digital images for the likely presence of hidden secret messages without knowledge of the employed steganographic technique. This paper proposes a novel image steganalysis technique to attack spatial domain LSBR stego images. The chosen steganalytic feature is the relation between length of the embedded message and the regressed proportion of intensity identical pixels and color channels. A trained SVM analyzes the pixels and the final decision is made based on union of the pixel analysis results. In SW, a number of innovative contributions are made to the field of blind image steganalysis. First, measuring pixel and cannel color correlativity as steganalytic feature. Second, defining pixel membership degree, thereby the pixels gain different level of influence on the process. Third, generating six references for statistical patterns of cover and stego pixels. And fourth, achieving 99.626% steganalyzer sensitivity on 0.25bpp stego images by only two analysis dimensions.