A DCT-domain multi-channel CM-JND model in YCbCr space

Abstract

The just noticeable distortion(JND) model has the better feature that can represent the human visual system's characteristics. To improve the accuracy of the JND model in multi-channel space, this paper proposes a DCT-domain multi-channel JND model in the YCbCr space. Firstly, the proposed JND model can directly compute the JND thresholds in the YCbCr space to be applied effectively in the video coding standards using the YCbCr format. Secondly, four effects are jointly considered by the proposed JND thresholds in chroma channels for the first time. Finally, in two chroma channels, we measure the contrast masking effect by psychophysical tests with the 8×8 DCT block size to accurately estimate the HVS thresholds in DCT domain. In tests, we also find out that, as the texture complexity is invariant, the contrast masking effects of the chroma components show a low pass property in frequency, which differs from the luma component that has a bypass property in frequency. Based on this observation, we model a continuous function of DCT frequency, the texture complexity, and the contrast masking effect in Cb and Cr channels, respectively. Experimental results show that the proposed multi-channel JND yields more tolerance for distortions with similar perceptual quality, compared to the Luo's JND model in the YCbCr space.