SIFT Feature-Based Watermarking Method Aimed at Achieving IHC Ver.5

Abstract

We propose a watermarking method using scale-invariant feature transform (SIFT) features that have both scale and rotation invariance, and evaluate our method in accordance with the information hiding criteria (IHC) ver. 5. It is defined as evaluation criteria against several possible attacks; these attacks are JPEG compression and geometric attacks, e.g., scaling, rotation, and clipping. In our method, we use local feature regions located around the SIFT features that are robust against scaling and rotation. The regions are normalized in size and selected as marked regions. Watermarks are embedded in the marked regions. We also introduce two error-correction techniques: weighted majority voting (WMV) and low-density parity-check (LDPC) code. When a stego-image is attacked by scaling or rotation, the image is spatially distorted. WMV and LDPC code can correct errors of extracted watermarks in the distorted stego-image. On the other hand, it is not easy to detect rotated marked regions. Therefore, the correct orientation is searched for by brute force. We evaluated the proposed method in accordance with IHC ver. 5. Our method can achieve robustness against scaling and rotation attacks in the highest tolerance category.