A universal denoising framework with a new impulse detector and nonlocal means

Area of Science:

• Image processing
• Computer vision
• Signal processing

Background:

• Impulse noise detection is crucial for effective image denoising.
• Existing methods struggle with impulse/Gaussian mixed noise and universal noise detection.

Purpose of the Study:

• To propose a novel impulse noise detection mechanism and a universal noise-filtering framework.
• To enhance image denoising performance for various noise types, including mixed noise.

Main Methods:

• Developed the Robust Outlyingness Ratio (ROR) for pixel impulsivity measurement.
• Implemented a two-stage (coarse-to-fine) detection and iterative filtering process.
• Extended Nonlocal Means (NL-means) filtering with a reference image for impulse noise removal, creating the ROR-NLM framework.

Main Results:

• The ROR detector outperforms existing methods in accuracy and robustness.
• The ROR-NLM framework achieves superior denoising results compared to existing filters for diverse noise models.
• Achieved high peak signal-to-noise ratio and excellent image quality, especially for impulse/Gaussian mixed noise.

Conclusions:

• The proposed ROR-NLM method offers a robust and effective solution for universal impulse noise detection and filtering.
• This approach significantly improves image quality in the presence of complex noise patterns.