A semi-fragile image watermarking algorithm with two-stage detection

The ability against the collage attack of semi-fragile watermarking is improved by embedding the watermark of a block in other blocks, but the tamper detection performance is impaired under general tampering. A two-stage detection method is proposed to improve the tamper detection performance of semi-fragile watermarking. For each 8 x 8 block, six-bit watermark data generated by the significant DCT (Discrete Cosine Transformation) coefficients are divided into two parts with the same length: general tampering watermark (GTW) and collage attack watermark (CAW). The GTW and CAW data of a block are embedded in the quantized DCT coefficients of itself and other blocks, respectively. In the first-stage detection, the general tampered regions are localized by the GTW data. To identify whether the collage attack exists in the received image, the identification parameter is defined by both GTW and CAW data. The selection of the predefined threshold of the identification parameter is derived and verified by the statistical experiments. If the identification parameter is larger than the given threshold, the second stage detection is performed to detect the collaged regions. Experimental results demonstrate that the proposed two-stage detection method is able to identify tampering with high probability under general tampering, collage attack and hybrid attack.