Local Spectral Component Decomposition for Multi-Channel Image Denoising.

Area of Science:

• Image Processing
• Computer Vision
• Remote Sensing

Background:

• Multi-channel images often suffer from noise, degrading their quality and utility.
• Existing denoising methods can cause image deterioration, especially when spectral component correlations are imbalanced.
• Hyperspectral images present unique challenges due to their high dimensionality and spectral complexity.

Purpose of the Study:

• To propose a novel method for local spectral component decomposition to reduce noise in multi-channel images.
• To exploit the linear correlation in the spectral domain of local image regions for effective noise reduction.
• To develop a denoising algorithm that minimizes image deterioration and preserves contrast, particularly for hyperspectral imaging.

Main Methods:

• A spectral line feature is calculated from the spectral components of an M-channel image.
• The image is decomposed into three components: one M-channel image and two grayscale images.
• Noise is concentrated in the two grayscale images, which are then denoised independently of the number of channels.

Main Results:

• The proposed method successfully reduces noise in multi-channel images.
• Image quality is improved with less deterioration compared to existing methods.
• Vivid contrast is preserved, and the method demonstrates particular effectiveness for hyperspectral images.

Conclusions:

• The spectral component decomposition method offers an effective approach to image denoising.
• The algorithm's ability to denoise grayscale components simplifies the process regardless of channel count.
• The method achieves state-of-the-art performance, competing with existing advanced denoising techniques.