Spectral image reconstruction using an edge preserving spatio-spectral Wiener estimation

Area of Science:

• Image processing
• Computational imaging
• Spectral imaging

Background:

• Spectral image reconstruction from camera responses is challenging due to noise and loss of spatial information.
• Existing methods often struggle to preserve fine details and edges during spectral reconstruction.

Purpose of the Study:

• To develop an advanced edge-preserving spatio-spectral Wiener estimation method for accurate spectral image reconstruction.
• To improve the handling of noise and preserve spatial details, especially at image edges.

Main Methods:

• A single spatio-spectral filter is created by combining Wiener denoising and spectral reconstruction filters.
• Local noise covariance is propagated, and bilateral weighting is used for estimating local mean and covariance to preserve edges.
• The method is derived using Bayesian inference.

Main Results:

• The proposed edge-preserving spatio-spectral Wiener estimation effectively reconstructs spectral images.
• Simulations on a six-channel camera system and multispectral test images demonstrate superior performance at edge regions.
• The filter converges to standard Wiener reflectance estimation under low noise conditions.

Conclusions:

• The developed edge-preserving spatio-spectral Wiener filter offers a robust solution for spectral image reconstruction.
• This technique significantly enhances the preservation of image edges, leading to more accurate spectral information.
• A MATLAB implementation is available for practical application.