Adaptive Spatial Domain Image Watermarking Based on Support Vector Machine

An adaptive spatial domain image watermarking algorithm based on support vector machine (SVM) is proposed. Since there is very close similarity between SVM and human visual system (HVS) in self-learning, generalization and non-linear approximation, the watermark embedding locations and strength can be adaptively identified by applying SVM algorithm based on the HVS. In this scheme, a kind of un-supervisory machine learning method, named fuzzy c-mean clustering algorithm, is first used to label the pixels in a cover image. Then, only those pixels whose subjection-value exceed a given threshold are selected from each label to be the training sample set of SVM. Sequentially, an SVM based multi-classification model is established. According to this model, the watermark embedding locations are further optimized. Finally, a bit of the watermark is adaptively embedded by adjusting the corresponding pixel value, according to the image local correlation. The presented watermarking scheme can extract the watermark without the help of the original image. Experimental results show that the proposed adaptive scheme has both sound perceptual quality and high robustness to various signal processing such as lossy compression, noise addition, image enhancement, filtering, cropping, mosaic, blurring, and so on. The watermarking performance notably outperforms the similar algorithm.