An Improved Sample Pairs Method for Detection of LSB Embedding

Abstract

This paper presents a fast algorithm to detect the least significant bit (LSB) steganography. Our approach is inspired by the work of Dumitrescu et al. [1] who detected LSB steganography via sample pair analysis. Our new algorithm combines with the statistical measures developed in [1] and a new least square estimation. The motivation comes from the high accuracy and robustness of the least square method used in parameter estimation. Plentiful experimental results show that our novel method has much lower false alarm rate of 5% than that of 13.79% in [1]. Meanwhile, the estimating precision of our algorithm is about 9% higher than that of the algorithm [1] if the embedding ratio is less than 10%, and the speed of our algorithm is also about 15% faster than the algorithm [1]. Some theoretical derivations are also included. © Springer-Verlag 2004.