Abstract: The existing image watermarking algorithm is of good robustness when against some common image processing attacks such as Gaussian noise, median filtering and lossy compression, etc. However, the robustness of watermark is poor when against some geometric attacks such as cropping and scaling, etc. Therefore, in order to improve the robustness of watermark, an embedded watermarking method is proposed based on invariant feature by combining the scale invariant feature transform and visual model. In the detection scheme, the probability density function of wavelet coefficients which in mid-frequency sub-band is modeled by the generalized Gaussian distribution model and the watermark decision threshold is determined by using the binary hypothesis theory. Furthermore, the receiver operating characteristic (ROC) curve between the probability of false alarm and the probability of detection is derived. Experimental results demonstrate that the proposed method is of good robustness when against some attacks such as JPEG compression, additive white Gaussian noise, scaling, cropping and rotation etc.

Key words: digital watermarking, invariant feature, wavelet transform, scale invariant feature transform