No-Reference Image Blur Assessment Based on Discrete Orthogonal Moments

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Image quality is significantly impacted by blur, which alters image edges and shapes.
• Discrete orthogonal moments are effective for shape description and can represent blur's effect on image moments.

Purpose of the Study:

• To develop a blind image blur evaluation algorithm using discrete Tchebichef moments.
• To create a blur score that aligns with human visual perception.

Main Methods:

• Compute image gradients to capture shape information.
• Divide gradient images into blocks and calculate Tchebichef moments for shape characterization.
• Define blur score using variance-normalized moment energy, guided by a visual saliency model.

Main Results:

• The proposed algorithm achieves high consistency with subjective blur evaluations.
• The method outperforms state-of-the-art image blur metrics and general no-reference quality metrics.
• Experimental validation on four public image quality databases confirms the algorithm's effectiveness.

Conclusions:

• The discrete Tchebichef moment-based algorithm provides a robust method for blind image blur evaluation.
• The integration of visual saliency enhances the algorithm's relevance to human perception.
• This approach offers a significant advancement in no-reference image quality assessment.