An Optimization-Based Family of Predictive, Fusion-Based Models for Full-Reference Image Quality Assessment.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Full-reference image quality assessment (FR-IQA) traditionally relies on hand-crafted metrics.
• Existing FR-IQA metrics have limitations in accurately capturing perceptual quality.
• Fusion-based approaches aim to improve FR-IQA by combining multiple metrics.

Purpose of the Study:

• To develop a novel framework for FR-IQA.
• To formulate FR-IQA as an optimization problem by fusing multiple existing metrics.
• To enhance the accuracy of perceptual quality assessment.

Main Methods:

• A novel framework for FR-IQA is proposed, formulating the task as an optimization problem.
• The framework fuses multiple hand-crafted FR-IQA metrics using a weighted product approach.
• Weights are determined via an optimization framework that maximizes correlation and minimizes RMSE with ground-truth scores.

Main Results:

• The proposed fusion-based FR-IQA metrics demonstrate superior performance on benchmark databases.
• The method outperforms existing state-of-the-art algorithms, including deep learning-based approaches.
• Optimized fusion of metrics leads to improved perceptual quality prediction.

Conclusions:

• The novel optimization-based fusion framework offers a significant advancement in FR-IQA.
• This approach effectively leverages the strengths of individual FR-IQA metrics.
• The proposed method provides a robust and accurate solution for perceptual image quality assessment.