A self-referencing, key-free watermarking framework for JPEG-domain image integrity verification

چکیده مقاله

Digital image authentication is essential for verifying the integrity of visual content in modern multimedia systems. We present a key-free, self-referencing framework for image-integrity verification. The method embeds a learned watermark in a transform-domain carrier and derives the reference information from non-embedded image components. A residual neural encoder–decoder and a robust transform-domain descriptor support reference-free verification without a shared secret key, provided that the JPEG quantization table is available. The image is divided into non-overlapping blocks. Selected transform-domain components construct a compact reference descriptor, whereas another component carries the watermark. During verification, the extracted watermark is compared with the regenerated reference descriptor to identify unauthorized modifications. To improve robustness to coefficient perturbations and JPEG re-encoding, the framework uses a robust DCT descriptor before cryptographic binding. Experiments were conducted on a dataset of 1014 images using an image-level separation protocol. The proposed method achieves reliable watermark recovery under JPEG compression at high quality factors and provides effective localization of various tampering operations. Additional experiments analyze compression robustness, carrier-region attacks, adversarial perturbations, and localization limitations. The results demonstrate a practical balance between image fidelity and post-embedding tamper detection. The remaining limitations, including geometric desynchronization and adaptive watermark regeneration, are explicitly discussed.