Geometrically Invariant Image Watermarking Using Scale-Invariant Feature Transform and K-Means Clustering

Abstract

In the traditional feature-base robust image watermarking, all bits of watermark message are bound with the feature point. If a few of points are attacked badly or lost, the performance of the watermarking scheme will decline or fail. In this paper, we present a robust image watermarking scheme by the use of k-means clustering, scale-invariant feature transform (SIFT) which is invariant to rotation, scaling, translation, partial affine distortion and addition of noise. SIFT features are clustered into clusters by k-means clustering. Watermark message is embedded bit by bit in each cluster. Because one cluster contains only one watermark bit but one cluster contains many feature points, the robustness of watermarking is not lean upon individual feature point. We use twice voting strategy to keep the robustness of watermarking in watermark detecting process. Experimental results show that the scheme is robust against various geometric transformation and common image processing operations, including scaling, rotation, affine transforms, cropping, JPEG compression, image filtering, and so on.