Steganalysis of LSB matching based on local binary patterns

Least significant bit (LSB) matching is a well-known steganographic method, which can embed large payload into cover data with good visual and statistical imperceptibility. However, it disturbs the correlation of adjacent pixels in smooth image regions as it randomly modifies half of the payload pixels by 1. Local binary patterns (LBPs) are first proposed as texture features, and can summarize local image structures efficiently by comparing each pixel with its neighbors. In this paper, we propose to utilize LBPs to detect LSB matching steganography. In brief, multi-scaled rotation invariant LBPs are extracted from smooth pixels as distinctive features, and the features are trained and classified using linear support vector machine. Extensive experiments are conducted to compare our method with some state-of-the-art targeted steganalyzer, and the results show the superiority of our method with a higher detection accuracy.