No-Reference Image Quality Assessment by Wide-Perceptual-Domain Scorer Ensemble Method

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Assessing image quality without a reference image is crucial for various applications.
• Existing methods often struggle with diverse and authentic image distortions.

Purpose of the Study:

• To develop a robust no-reference (NR) image quality assessment (IQA) model.
• To improve the accuracy and reliability of IQA across a wide range of distortions.

Main Methods:

• Feature extraction from five perceptual domains: brightness, contrast, color, distortion, and texture.
• Training a predictive model (scorer) using these features.
• Employing scorer selection algorithms and an ensemble method for combining predictions.
• Developing single-scale and multiple-scale versions of the approach.

Main Results:

• Multiple-scale versions outperformed the single-scale method.
• The model demonstrated robustness against more than 24 types of image distortions.
• Effective evaluation of images with authentic distortions was achieved.

Conclusions:

• The proposed NR-IQA model, leveraging multi-domain features and ensemble techniques, offers high robustness.
• The multiple-scale approach enhances performance and generalizability.
• The method is suitable for evaluating images with both synthetic and authentic distortions.