Just Noticeable Difference for Natural Images using RMS Contrast and Feed-back Mechanism

Abstract

Contrast Sensitivity (CS), Luminance Adaptation (LA) and Contrast Masking (CM) are important contributing factors for Just Noticeable Difference (JND) in images. Most of the existing pixel domain JND algorithms are based only on LA and CM. Research shows that the human vision depends significantly on CS, and an underlying assumption in the existing algorithms is that CS cannot be estimated in the pixel domain JND algorithms. However, in the case of natural images, this assumption is not true. Studies on human vision suggest that CS can be estimated using the Root Mean Square (RMS) contrast in the pixel domain. With this in perspective, we propose the first pixel-based JND algorithm that includes a very important component of the human vision, namely CS by measuring RMS contrast. This RMS contrast is combined with LA and CM to form a comprehensive pixel-domain model to efficiently estimate JND in the low frequency regions. For high frequency regions (edge and texture), a feedback mechanism is proposed to efficiently alleviate the over- and under-estimation of CM. To facilitate this, a prediction based algorithm is used to classify an image into low (smooth) and high frequency regions. This feed-back mechanism is based on the relationship between the CS and RMS contrast. Experiments validate that the proposed JND algorithm efficiently matches with human perception and produces significantly better results when compared to existing pixel domain JND algorithms.