A Reduced Reference Image Quality Measure Using Bessel K Forms Model for Tetrolet Coefficients

TL;DR

This paper proposes a reduced reference image quality assessment method using Tetrolet transform and Bessel K Forms modeling, which effectively predicts image quality with minimal side information and aligns well with human judgments.

Contribution

It introduces a novel RRIQA measure based on Tetrolet transform and Bessel K Forms modeling, providing an efficient way to assess image quality with limited reference data.

Findings

The proposed measure shows good consistency with human quality judgments.

The method effectively summarizes reference images with minimal side information.

Experimental results validate the accuracy of the quality prediction.

Abstract

In this paper, we introduce a Reduced Reference Image Quality Assessment (RRIQA) measure based on the natural image statistic approach. A new adaptive transform called "Tetrolet" is applied to both reference and distorted images. To model the marginal distribution of tetrolet coefficients Bessel K Forms (BKF) density is proposed. Estimating the parameters of this distribution allows to summarize the reference image with a small amount of side information. Five distortion measures based on the BKF parameters of the original and processed image are used to predict quality scores. A comparison between these measures is presented showing a good consistency with human judgment.