A divide-and-conquer fragile self-embedding watermarking with adaptive payload

摘要

This paper proposes a divide-and-conquer fragile self-embedding watermarking with adaptive payload for digital images. A graph-based visual saliency (GBVS) model is adopted to automatically classify image blocks into region of interest (ROI) and background (ROB). The divide-and-conquer mechanisms aim to protect the ROI blocks with higher priority, which is embodied in two procedures: backup information collection and payload allocation. We collect the ROI backup information without compression, and allocate payload in a water-filling order to preferentially maintain the visual quality of ROI. The collected backup information are encoded as reference bits through a measurement process, in which a flexible scaling factor adaptively modulates the size of payload. Auxiliary information, which records the ROI locations, is embedded into the host images together with the reference bits. Hash-based authentication bits are responsible for detecting tampered blocks. A legitimate recipient can sequentially restore the auxiliary information and the original image content as long as the tampering is not too severe. The qualitative and quantitative results demonstrate the effectiveness and the superiority of the proposed methods compared with the previous works.