Structural-Equation-Modeling-Based Indicator Systems for Image Quality Assessment

Area of Science:

• Computer Vision
• Image Processing
• Perception Science

Background:

• Advanced image denoising algorithms enhance visual quality but challenge existing image quality assessment (IQA) metrics.
• Current single-indicator IQA methods struggle to align with human visual perception of denoised images.

Purpose of the Study:

• To develop a comprehensive IQA framework integrating multiple indicators for holistic image quality assessment.
• To create a robust and optimal indicator system for evaluating denoised images.

Main Methods:

• Development of a large-scale denoised image dataset with diverse distortions.
• Utilizing structural equation modeling to correlate structural similarity, information loss, and perceptual gain.
• Employing regressions and iterative refinements to select optimal IQA indicators.

Main Results:

• A robust and optimal system of IQA indicators was established.
• The proposed framework demonstrated high reliability and validity in assessments.
• Effective performance was shown across image quality prediction, IQA indicator comparison, and denoising algorithm optimization tasks.

Conclusions:

• The integrated IQA framework provides more reliable and valid assessments than single-indicator methods.
• The system offers valuable insights for analyzing and applying IQA indicators in image processing.
• This approach enhances the evaluation of image denoising algorithms and overall image quality.