PSNR over JND: A JND-Based Watermark Imperceptibility Metric for Color Image
Na Guo, Ying Huang, Hu Guan, Baoning Niu, Zhi Zeng, Yang Zheng · 2023
The most common image watermark imperceptibility metric is PSNR. However, human vision has different perception thresholds in different regions of an image, but PSNR is simply defined by the MSE of the image pixels. PSNR don’t consider distinctions between different regions of the image, resulting in the metric result is not completely in line with the subjective perception of the human visual system. In this paper, the space and frequency domain information of the original image are extracted by the wavelet transform, and analyzed in conjunction with the characteristics of the human visual system such as luminance adaptation, color masking and texture masking. The just noticeable difference(JND) is calculated by integrating these information with the appropriate JND model. Using the JND as weights to calculate the weighted error of the original image and the watermarked image, and giving the metric results in the form of peak signal-to-noise ratio(PSNR), an optimized watermark imperceptibility metric for digital color images can be obtained. Experimental results on public image datasets demonstrate that the metric in this paper can improve the consistency between the objective measurement results of invisibility and the subjective perception of human vision.