SPSIM: A Superpixel-Based Similarity Index for Full-Reference Image Quality Assessment

Area of Science:

• Computer Vision
• Image Processing
• Perceptual Computing

Background:

• Traditional image quality assessment relies on square patches lacking perceptual meaning.
• Superpixels, representing perceptually meaningful image regions, offer a promising alternative for feature extraction.

Purpose of the Study:

• To develop a superpixel-based image quality assessment (IQA) index.
• To enhance IQA by incorporating perceptually meaningful features and revised similarity measures.

Main Methods:

• Extracted features from superpixels, focusing on luminance and chrominance similarity.
• Quantified structural variations using pixel gradient similarity.
• Revised similarity measures by incorporating regional gradient consistency and texture complexity.

Main Results:

• The proposed method evaluates image quality based on superpixel luminance, chrominance, and pixel gradient similarities.
• Regional gradient consistency was analyzed and integrated to improve quality assessment.
• Experiments on benchmark databases showed competitive performance against state-of-the-art metrics.

Conclusions:

• Superpixel-based feature extraction significantly improves image quality assessment.
• The novel index, incorporating regional gradient consistency and texture complexity, offers a robust and accurate approach to IQA.
• The method demonstrates strong potential for practical applications in image quality evaluation.