An image watermarking technique based on support vector regression

A new image watermarking technique based on support vector regression (SVR) is proposed in this paper. It uses support vector machine (SVM) to embed the watermark and gains satisfying results. Due to the good learning and generalization capability in the processing of small-sample learning problems, SVR can well memorize the relationship between the random selected pixel and its neighboring pixels in an image. Then, a bit of the watermark is embedded or extracted by adjusting or comparing the relation between the embedded pixel and the output of the trained SVR. Extensive experimental results show that the SVM is much more suitable than the neural network to model the relationship among the neighboring pixels, and the proposed technique possesses good visual perception and high robustness to common image processing and the JPEG compression, and also has high security and practicability.