Lagrangian twin SVR based grayscale image watermarking using LWT-QR decomposition

Abstract

A novel approach of image watermarking using Lagrangian twin support vector regression (LTSVR) and combination of a variant of wavelet transform and QR decomposition for copyright protection application is proposed in this chapter. Firstly, host image decomposed into low-frequency subband (LL) and detail subbands by applying lifting wavelet transform (LWT). Secondly, the blocks of LL depend on the fuzzy entropy, and the selected regions of LL are transformed using QR factorization. Then, image dataset is formed using the elements of matrix R (called feature vector) of each selected block. This image dataset acts input to LTSVR to find the function approximation which defines the relationship between the input and target. The scrambled bits of the binary watermark are inserted into the predicted values obtained through the trained LTSVR upon comparing with the target value. The scrambled bits are obtained by applying Arnold transformation, which provide security to the proposed approach. Experimental results using various kinds of images and comparison of existing methods prove that the proposed approach is highly imperceptible and robust.