A psychophysical performance-based approach to the quality assessment of image processing algorithms

Area of Science:

• Computer Vision
• Image Processing
• Human Perception

Background:

• Current image quality metrics often lack perceptual relevance and are numerically assessed.
• Existing perceptual metrics can be sensitive to specific criteria and have limitations.

Purpose of the Study:

• To propose and validate an objective, performance-based perceptual measure for image quality.
• To compare the effectiveness of various denoising algorithm filters using this new metric.

Main Methods:

• Utilized a two-alternative forced-choice (2AFC) paradigm to measure white noise detection thresholds in natural images.
• Assessed image pairs subjected to different configurations of a denoising algorithm.
• Derived Cartesian-separable log-Gabor filters with polar parameters.

Main Results:

• Objective perceptual measures of image denoising efficacy were obtained by comparing noise detection thresholds.
• A consistent performance factor of two (6dB) was observed across various filter types and bandwidths.
• Image quality thresholds converged on a common Peak Signal-to-Noise Ratio (PSNR) value, validating the metric.

Conclusions:

• The proposed 2AFC approach offers a robust, objective perceptual measure for evaluating image processing algorithms.
• This method can be extended to assess other algorithms like compression, deblurring, and edge-detection.
• The derived log-Gabor filters offer advantages for the biological vision community.