A feature-enriched completely blind image quality evaluator.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Existing blind image quality assessment (BIQA) methods are predominantly opinion-aware, requiring extensive human-scored training data.
• Opinion-aware BIQA models often exhibit limited generalization due to diverse distortion types and sample size constraints.
• Current opinion-unaware BIQA methods lack the accuracy of opinion-aware approaches.

Purpose of the Study:

• To develop a highly accurate opinion-unaware blind image quality assessment (BIQA) method.
• To overcome the limitations of data dependency and poor generalization in existing BIQA models.
• To achieve quality prediction performance competitive with or superior to state-of-the-art opinion-aware methods.

Main Methods:

• Integrates natural image statistics from multiple cues.
• Learns a multivariate Gaussian model of image patches from pristine images.
• Employs a Bhattacharyya-like distance for patch quality assessment and average pooling for overall score.

Main Results:

• The proposed opinion-unaware BIQA method demonstrates superior quality prediction performance.
• Achieves better results than state-of-the-art opinion-aware BIQA methods in extensive experiments.
• Requires no distorted images or subjective scores for training, enhancing practical usability.

Conclusions:

• The developed opinion-unaware BIQA method offers a robust and generalizable solution for image quality assessment.
• This approach eliminates the need for subjective scoring, simplifying the training process.
• The method provides a competitive and potentially superior alternative to current BIQA techniques.