The quality evaluation of image recovery attack for visible watermarking algorithms

Image quality evaluation is an important technique to judge the image fidelity of the image applications. In this study, it not only analyzes and evaluates the entire watermarked image before and after attacks, but also helps the encoder to design the superior watermarking algorithms. In order to get the best practice of transmission power on watermark intensity and perceptual efficiency, this study investigates new visible watermarking algorithms based on human vision system (HVS) model and noise visibility function to provide adequate flexibility during the watermark energy embedding. Since it is not possible for all digital images to go through detailed subjective evaluation, 12 image quality metrics are utilized here to automatically predict the subjective judgments of watermarked image since image quality assessments are influenced by the transparency and robustness of watermarked images. In addition, malicious attack like image recovery is implemented to verify the consistency of image quality metrics. From the experimental results, the algorithm of the perceptual model of contrast-sensitive function (CSF) with noise reduction of the visibility thresholds for HVS in DWT domain outperforms the CSF only based algorithm. On the other hand, the image quality metric evaluation values before and after the malicious attack demonstrate the similar trend of judgment which prove the evaluation consistency. Therefore, such study could help the algorithm design for better image fidelity preservation and secure robustness of practical visual communication.