Decomposed Dissimilarity Measure for Evaluation of Digital Image Denoising

Area of Science:

• Image processing
• Computer vision
• Signal processing

Background:

• Digital image denoising is crucial for enhancing image quality.
• Existing evaluation metrics for denoising algorithms may not fully capture performance nuances.
• Impulsive noise removal presents unique challenges in image restoration.

Purpose of the Study:

• To introduce a new, comprehensive approach for evaluating digital image denoising algorithms.
• To decompose the mean absolute error (MAE) into components revealing specific denoising imperfections.
• To present a clear visualization tool (aim plots) for assessing denoising performance.

Main Methods:

• Decomposition of the mean absolute error (MAE) into three distinct components.
• Development and description of 'aim plots' for intuitive performance visualization.
• Application of the decomposed MAE and aim plots to evaluate impulsive noise removal algorithms.

Main Results:

• The decomposed MAE provides detailed information on error sources like pixel estimation errors and uncorrected distortions.
• Aim plots offer a clear and intuitive graphical representation of denoising performance.
• The method effectively evaluates algorithms designed for partial pixel distortion detection.

Conclusions:

• The decomposed MAE offers a hybrid measure combining image dissimilarity and detection performance.
• This new evaluation approach enhances the understanding of denoising algorithm effectiveness.
• The method is particularly valuable for assessing algorithms targeting specific noise patterns in images.