This paper presents a digital watermarking technique based on Support Vector Machines (SVMs). Use the nice characteristic of the SVM, which can result an optimal hyperplane for the given training samples, the imperceptibility and robustness requirements of watermarks are fulfilled and optimized. In the proposed scheme, to improve imperceptibility, the watermark is embedded by asymmetrically tuning blue channels of the central and surrounding pixels at the same time. Furthermore, to promote robustness, the embedded watermark bits will be re-modified if necessary according to classifying result of the trained SVM. Our scheme uses only 128 bits in training SVM, thus it is time efficient. Watermarks are embedded in spatial domain and extracted directly from a watermarked image without the requirement of original image. Experiments show that the proposed scheme provides high PSNR of a watermarked images and low extraction error rate.
淡江理工學刊=Tamkang journal of science and engineering 9(2), pp.141-150