No-reference image quality metrics for color domain modified images

Area of Science:

• Computer Vision
• Image Processing
• Perceptual Quality Assessment

Background:

• Assessing natural image quality without a reference (no-reference image quality assessment) is challenging.
• Existing image quality metrics (IQMs) excel at spatial/frequency degradations (blur, noise) but falter with color domain modifications.
• Color appearance attributes significantly impact perceived image quality, necessitating specialized metrics.

Purpose of the Study:

• To evaluate the performance of existing IQMs on color-domain distorted images.
• To develop and validate novel IQMs specifically designed for color domain image quality assessment.
• To investigate the effectiveness of color appearance attributes in predicting perceived image quality.

Main Methods:

• Conducted psychophysical experiments to gather subjective quality scores for color-altered images.
• Created a new dataset of color-modified images and incorporated existing datasets.
• Developed three new IQMs based on absolute, relative, and statistical color appearance attributes.
• Evaluated the proposed IQMs against five established spatial-domain IQMs using cross-database validation.

Main Results:

• The newly developed color-domain IQMs significantly outperformed existing spatial-domain IQMs.
• Models based on absolute and relative color appearance attributes demonstrated the highest performance when combined.
• Cross-database evaluation confirmed the robustness of the proposed color-domain IQMs.

Conclusions:

• Color appearance attributes are crucial for accurate no-reference image quality assessment.
• Existing spatial-focused IQMs are insufficient for evaluating images with color distortions.
• Further research into color-domain IQMs is essential for comprehensive image quality prediction.