Image steganalysis based on space-frequency domain interaction and collaborative enhancement.

Abstract

Recent advances in image steganalysis have achieved remarkable progress in spatial-domain modeling, but limitations remain in the simultaneous utilization of both spatial and frequency-domain characteristics. This, to some extent, restricts the model's sensitivity and discriminative capability to subtle steganographic perturbations. Additionally, steganographic information often exists in a weakened and cross-domain distributed form, with its small scale and non-locality making it difficult for single-domain modeling to effectively extract key discriminative clues, thereby limiting the model's performance in complex scenarios. To address these issues, this paper proposes a cross-domain collaborative enhancement steganalysis method that jointly models spatial-domain texture perturbations and frequency-domain structural distributions to enhance the perception of steganographic features. Specifically, we design a multi-path enhanced residual structure to reveal multi-dimensional difference features, interact and fuse information between space and frequency, thereby effectively improving the signal-to-noise ratio. In addition, we design a dedicated collaborative enhancement network based on the residual features to strengthen the recognition and extraction of steganographic patterns, thus enabling more efficient capture of cross-domain correlations. The experimental results indicate that, compared with existing methods, the proposed model demonstrates a clear advantage in detection accuracy. In addition, comprehensive ablation studies further validate the effectiveness of multi-domain steganographic information fusion.