Deep Learning and Error Correction Code-Based Robust Watermark Extraction from DWT–Chaotic Embedding

Abstract

A blind digital watermarking framework based on Discrete Wavelet Transform (DWT) is proposed for secure image authentication and copyright protection. A binary watermark is embedded within the LH subband, and a chaotic map is used to select embedding positions, ensuring high randomness and security. During the extraction, a convolutional neural network (CNN) is applied to the embedded subband and combined with an error correction code to improve recovery performance under attacks. Experimental results on 200 COCO test images demonstrate that, with a magnitude factor of 0.35, the proposed method achieves an average Peak Signal-to-Noise Ratio (PSNR) of 29.46 dB and an SSIM of 0.9189 on 200 COCO test images. The results indicate that the framework achieves stable watermark recovery under blur, Gaussian noise, JPEG compression, and scaling, with high perceptual image quality.