Mitigating Steganalysis Collapse Under Re-Compression via Pre-Transmission Guidance

Current steganalysis primarily focuses on analyzing clean pre-transmission images, which we term Transmission-Prior Steganalysis (TPS), neglecting the performance degradation caused by lossy transmission channels. This creates a critical mismatch in real-world scenarios where modern JPEG-resistant steganography preserves message integrity despite aggressive recompression, whereas transmission-induced distortions largely compromise detection performance. We formally identify this problem as Transmission-Disturbed Steganalysis (TDS) and propose PGD-Net (TPS Guides TDS Network), a teacher-student framework that bridges transmission-prior knowledge and distorted-image analysis through dual alignment mechanisms. The framework simultaneously ensures prediction consistency through output distribution alignment and preserves discriminative features via structured relation alignment. Experimental results demonstrate great improvements in detection performance for existing steganalyzers when applied to distorted images. By establishing the first benchmark for quality-loss scenarios, this work addresses a new practical deployment challenge, further advancing the field toward robust real-world applications.