[Paper Review] No-Reference Quality Assessment of Contrast-Distorted Images using Contrast Enhancement

Jia Yan, Jie Li+1 more|arXiv (Cornell University)|Apr 18, 2019

Image and Video Quality Assessment34 references71 citations

TL;DR

A no-reference image quality assessment metric for contrast-distorted images based on contrast enhancement. It uses histogram equalization to create an enhanced image, then combines SSIM, entropy, and cross-entropy features through regression to predict quality scores.

ABSTRACT

No-reference image quality assessment (NR-IQA) aims to measure the image quality without reference image. However, contrast distortion has been overlooked in the current research of NR-IQA. In this paper, we propose a very simple but effective metric for predicting quality of contrast-altered images based on the fact that a high-contrast image is often more similar to its contrast enhanced image. Specifically, we first generate an enhanced image through histogram equalization. We then calculate the similarity of the original image and the enhanced one by using structural-similarity index (SSIM) as the first feature. Further, we calculate the histogram based entropy and cross entropy between the original image and the enhanced one respectively, to gain a sum of 4 features. Finally, we learn a regression module to fuse the aforementioned 5 features for inferring the quality score. Experiments on four publicly available databases validate the superiority and efficiency of the proposed technique.

Motivation & Objective

• Address the lack of NR-IQA attention to contrast distortion.
• Propose a simple yet effective metric for contrast-altered images.
• Leverage a contrast enhancement step to derive informative features.
• Fuse multiple features into a regression model to predict perceptual quality.

Proposed method

• Generate an enhanced image from the original using histogram equalization.
• Compute SSIM between the original and enhanced images as a quality feature.
• Compute histogram-based entropy between the original and enhanced images.
• Compute cross-entropy between the original and enhanced images.
• Learn a regression module to fuse the five features into a quality score.

Experimental results

Research questions

• RQ1 Can NR-IQA effectively assess quality degradation due to contrast distortion?
• RQ2 Does histogram-equalization-based enhancement provide informative features for quality prediction?
• RQ3 How well do SSIM, entropy, and cross-entropy features correlate with human quality judgments when contrast is distorted?
• RQ4 Does a learned regression fusion of these features yield accurate quality scores across datasets?

Key findings

• The proposed method achieves superior and efficient quality prediction for contrast-distorted images across experiments.
• A simple feature set (SSIM, histogram entropy, cross-entropy) combined with enhancement-derived features is effective.
• Experiments on four publicly available databases validate the approach.
• The approach demonstrates both performance and computational efficiency advantages.