Reducing coding loss with irregular syndrome trellis codes

Abstract

We propose an extension to the Syndrome Trellis Code (STC) algorithm, aiming to reduce distortion by realizing embedding change probabilities closer to the optimal than the existing framework. A proxy for detectability, the minimization of distortion plays a critical role in producing good stego objects. STCs have become the tool of choice for many steganographers, because they approach the theoretical bound for embedding performance in quasi-linear time, for arbitrary length covers and payloads. However, until recently little attention has been paid to how closely STCs realize optimal change probabilities, particularly near the start and end of the embedding path. Recent work by Köhler et al, aimed to modify the parity-check matrix used by STCs to produce change vectors with change probabilities closer to the optimal probabilities. However, there is a cost of reduced capacity, or increased distortion. This paper demonstrates a modification to the block-structured STC parity-check matrix that both achieves changes closer to the optimal probabilities, and can be used for arbitrary length covers, and payloads.