Visual JND: A Perceptual Measurement in Video Coding

Abstract

Humans cannot perceive the minimal level of difference in the pixel variation. To overcome the problem, the concept of just-noticeable difference (JND) was proposed. JND measures the minimal amount that must be changed for the variation to be detectable by humans. However, JND characteristics were not considered in the traditional perceptual measurements. In this paper, we provide a comprehensive survey of the latest JND-related studies. First, we provide an extensive overview of JND models. JND models comprise human visual system characteristics and masking effects. Next, we introduce the applications of JND models in the perceptual quality evaluation and video compression coding, especially in applying machine-learning techniques to JND prediction. In addition to a thorough summary of the recent progress and applications of JND, we summarize some unsolved technical challenges. We believe that our overview and findings can provide some insights into the related issues and future research directions in video coding.

Keywords

• Just noticeable difference (JND)
• human visual system (HVS)
• machine learning
• perceptual measurement
• video coding