Feature Based Affine Invariant Watermarking Robust to Geometric Distortions

摘要

This work, proposes a robust digital watermarking scheme based on feature points as a defense against illegal user attacks, such as, rotating, scaling and translation. First of all, a Gaussian filter is first adopted to discover the invariant feature points. The local maximum and the local minimum values are then calculated from those feature points. The Affine Invariant Region (AIR) is then divided by combination the values: local maximum, local minimum values and the center of mass. The AIR region is then transformed using affine transform to a block image, called the normalized image, in which the size of the block image is predefined. The watermarks are then embedded into the normalized image. Experimental results indicate that the proposed method can still extract the correct watermark, even after a geometric attack. The proposed method has a lower false positive rate and higher accuracy than Lu's [12] that making it a robust blind watermarking scheme.