Robust image watermarking using combined DSR-SVM with PCA based feature reduction

Abstract

This paper proposes a novel approach for image watermarking using combined dynamic stochastic resonance (DSR) and support vector machine (SVM). The algorithm incorporates lifting wavelet transform (LWT) to decompose the host image into three level frequency sub-bands, and a low-pass frequency sub-band is opted for watermark embedding. Watermark bits are embedded into small blocks of low-pass frequency sub-band using quantization of minimum and maximum coefficients of the corresponding blocks. And, to extract the watermark, DSR based coefficient enhancement process is incorporated. A features set of enhanced block coefficients is generated by employing different statistical parameters, and principal component analysis (PCA) is employed to reduce the dimensions of the features set, which are used for training and testing the learning machine. Training and testing patterns are generated using concatenation of reduced features with enhanced coefficients of the corresponding blocks. Finally, a binary watermark can be extracted using well trained machine (SVM) using binary classification approach. The experimental results show remarkable performance in terms of robustness against various geometrical and signal processing attacks.