A new scheme for watermark extraction using combined noise-induced resonance and support vector machine with PCA based feature reduction

Abstract

This manuscript presents a new scheme for binary watermark extraction using the combined application of noise-induced resonance (NIR) and support vector machine (SVM). The principal component analysis (PCA) is incorporated to minimize the dimension of the feature set obtained from the attacked watermarked image. The scheme utilizes lifting wavelet transform to decompose the original image into three levels, and blocks of low frequency sub-band coefficients are used for embedding purpose. Reference and signature information is embedded by quantizing the maximum and minimum coefficients of the corresponding block. Whereas, to extract the watermark, NIR-based tuning operation is performed. The transformed coefficients of the attacked watermarked image are tuned using iterative procedure of NIR in such a way that the transformed coefficients change their state from low signal-to-noise ratio (SNR) to maximum SNR or enhanced state. Finally, the tuned coefficients are fed into the machine i.e. SVM to classify as binary classes (0 or 1) which result in the corresponding watermark extraction. Experimental results of the proposed algorithm demonstrate noteworthy robustness against various signal processing attacks and remarkable improvements comparing with some of the recent techniques. Also, the scheme fulfills the requirements of image integrity in case of new strategic attack (i.e. print attack).