Fractal analysis for reduced reference image quality assessment.

Area of Science:

• Computer Vision
• Image Processing
• Signal Analysis

Background:

• Reduced-reference image quality assessment (RR-IQA) is crucial for digital image management.
• Existing RR-IQA methods often focus on pixel-level differences, potentially missing spatial pattern irregularities.

Purpose of the Study:

• To propose a novel RR-IQA method leveraging multifractal analysis.
• To assess image quality based on the spatial arrangement and regularity of image patterns.

Main Methods:

• Adapted multifractal analysis to the RR-IQA domain.
• Transformed images into the Log-Gabor domain.
• Computed fractal dimensions on Log-Gabor subbands to create feature vectors.
• Utilized l1 distance for feature pooling to derive the final quality score.

Main Results:

• The proposed method effectively captures spatial pattern differences between reference and distorted images.
• Evaluated on seven public benchmark datasets, demonstrating superior performance.
• Outperformed existing state-of-the-art RR-IQA approaches.

Conclusions:

• Multifractal analysis offers a unique perspective for RR-IQA by analyzing spatial regularity.
• The proposed Log-Gabor-based multifractal approach provides a robust and effective solution for image quality assessment.