A Novel DCT-based Just Noticeable Difference Model for Videos Based on Structure Complexity

Abstract

The Just Noticeable Difference (JND) model involves the minimum level of visibility where visual content can be distinguished, which plays a significant role in terms of the perceptual image/video. For the estimation of the JND threshold, the contrast masking effect evaluation is a critical task and has room for manoeuvre. Considering the important role of structural information for contrast masking evaluation, a structure complexity descriptor based on orientation selectivity characteristics of the human visual system (HVS) was introduced and a new contrast masking model based on structure complexity in the discrete cosine transformation (DCT) domain was estimated. Then, combining with the spatio-temporal CSF and luminance adaptation, a novel JND model for videos was proposed in the DCT domain. The experimental results of the subjective quality evaluation tests demonstrate that the proposed JND threshold can hide more noises under the same perceived quality, which is highly consistent with human subjective visual perception.