SRIS-Net: a robust image steganography algorithm based on feature score maps

• SRIS-Net integrates spatial and frequency-domain features via Laplacian pyramid decomposition, enabling robust and invisible image steganography. • A global–local embedding module (GLEM) and a dual multi-scale aggregation sub-network (DMSubNet) jointly improve carrier image quality and embedding effectiveness. • A dual-task discriminator produces feature score maps of the cover image's ROI, guiding embedding toward higher imperceptibility and undetectability. • On BOSSBase, SRIS-Net achieves over 9.2 dB and 3.4 dB visual quality improvements over mainstream methods, with capacity increased up to approximately 72–96 bits per pixel.