A color image watermarking in the frequency domain using a teaching-learning optimization algorithm

Abstract

This paper presents a color image watermark technique that employs teaching learning-based optimization algorithm (TLBO) and lagrangian twin support vector regression (LTSVR) in the frequency domain. By analyzing the statistical property of the selected wavelet band (LL sub-band) after single-level decomposition, LTSVR is used for extraction of watermark and embedding processes. TLBO is used to find the optimal value of watermark strength for different selected blocks of the image in the wavelet domain. Various kinds of images are considered to test the imperceptibility and robustness of the watermark in experimental results. The metric Peak Signal to Noise Ratio (PSNR) has been used for watermark images to evaluate the: (i) imperceptibility, (ii) quality. Bit error rate (BER) and normalized correlation (NC) value is computed to determine the effectiveness and standards of the extracted watermark. JPEG compression attack with different quality factors (QF) ranging from 10 to 90 is evaluated using robustness to determine the proficiency of the proposed work. Experimental results show that the proposed method is robust to JPEG compression as compared to state of art method.