Image Restoration Quality Assessment Based on Regional Differential Information Entropy

Area of Science:

• Computer Vision
• Image Processing
• Information Theory

Background:

• Advanced image recovery models produce visually similar but not identical textures.
• Traditional image quality assessment (IQA) methods show inconsistencies between subjective perception and objective scores.
• Existing IQA methods struggle to accurately assess subtle textural differences in recovered images.

Purpose of the Study:

• To develop a novel image quality assessment method that addresses subjective-objective inconsistencies.
• To accurately evaluate the perceived quality of images with similar but not identical textural details.
• To improve the efficiency and speed of information entropy calculations for IQA.

Main Methods:

• Proposed a regional differential information entropy (RDIE) method for image quality assessment.
• Utilized neural networks to optimize the calculation of information entropy.
• Evaluated the method on a custom IQA dataset and the PIPAL dataset.

Main Results:

• The RDIE method demonstrated high agreement with human average opinion scores.
• RDIE outperformed other image quality assessment metrics in quantifying perceived quality.
• Neural network integration significantly enhanced the speed and efficiency of entropy computation.

Conclusions:

• The proposed RDIE method effectively quantifies perceived image quality, especially for images with subtle textural variations.
• RDIE offers a more reliable and consistent approach to image quality assessment compared to traditional methods.
• The integration of neural networks makes RDIE a computationally efficient and practical solution for modern image processing tasks.