Robust DCT-Based Image Watermarking Using Adaptive Population Based Incremental Learning for Quantization Error Minimization

This study proposed a reliable digital image watermarking system in the frequency domain to overcome the rounding error problem in watermarking. To achieve the optimum sensitivity and robustness of the watermark, a discrete cosine transformation (DCT) was used to embed the watermark in the mid-frequency components. However, during the inverse DCT (IDCT) process to obtain the original image, quantization noise is introduced by converting real numbers into integers, which compromises the watermark quality. A population-based incremental learning algorithm (PBIL) was used to optimize the rounding process to minimize distortions. An adaptive learning rate was integrated into the above algorithm to increase the convergence rate for higher-quality solutions with different image contents. The proposed method was tested using traditional metrics and performance indicators, such as PSNR, SSSI, and NCSI. The results clearly show that the proposed methods are superior to conventional methods, such as fixed PBIL and genetic algorithms (GA), particularly in the face of conventional attacks on image processing, such as Gaussian noise, rotation, and blur.