Image Denoising With Edge-Preserving and Segmentation Based on Mask NHA

Area of Science:

• Signal Processing
• Image Analysis
• Digital Image Restoration

Background:

• Traditional frequency analysis methods like DFT and DCT suffer from sidelobes, complicating noise and image component separation.
• Non-harmonic analysis (NHA) offers high-resolution frequency analysis with reduced sidelobes, enhancing noise removal accuracy.
• Image signals can be non-stationary, leading to spectral distortions in NHA, necessitating region-specific analysis.

Purpose of the Study:

• To propose an advanced image denoising method for zero-mean white Gaussian noise removal.
• To address the spectral distortion issue in NHA for non-stationary image signals.
• To enhance noise removal accuracy and peak signal-to-noise ratio (PSNR) in image restoration.

Main Methods:

• Utilizing high-resolution frequency analysis for noise removal.
• Applying 2D non-harmonic analysis (2D NHA) for its sidelobe reduction capabilities.
• Developing an extended method, Mask NHA, to analyze homogeneous texture regions and non-uniform regions via segmentation.

Main Results:

• Mask NHA demonstrates superior performance in denoising simulation images, achieving higher PSNR values than state-of-the-art methods.
• The experimental results indicate the potential upper limit of PSNR for the Mask NHA method.
• Achieving optimal results is contingent on obtaining suitable segmentation of the input image.

Conclusions:

• Mask NHA is a promising technique for high-fidelity image denoising, especially for images with distinct textural regions.
• Further improvements in image segmentation techniques are expected to enhance the PSNR performance of Mask NHA.
• The method offers a significant advancement in accurately separating image components from noise using advanced frequency analysis.