Gray-scale image watermarking based on DE-KELM in DCT domain

Abstract

A semi-blind discrete cosine transform (DCT) domain watermarking technique for gray-scale images using differential evolution (DE) and kernel extreme learning machine (KELM) has been presented in this paper. The selection of m×m sized blocks (non-overlapping) is done with the help of entropy. It gives the degree of randomness of the sub blocks. DCT for the selected blocks is carried out, following which the low frequency coefficients are selected in zig-zag manner from each block. The coefficients from each block form the dataset of input vectors and target vector for KELM which uses the non-linear regression model to predict the coefficient where the watermark bit is to be embedded for each block. DE is used for obtaining the optimal scaling factors and controls the strength of the watermark. Use of multiple scaling factors helps in better optimization as it uses the local information like tolerance and noise sensitivity to determine the strength of watermark bit in each block. The watermarking scheme shows robustness against numerous attacks like JPEG compression, Weiner filtering, median filtering, average filtering, Gaussian filtering, histogram equalization, etc. as well as the watermarked image retains its quality and is not compromised. Hence, the proposed technique is good candidate for watermarking.