A Snapshot Multi-Spectral Demosaicing Method for Multi-Spectral Filter Array Images Based on Channel Attention...

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Multi-spectral imaging (MSI) systems utilize Multi-Spectral Filter Arrays (MSFA) for capturing dynamic scenes across various spectral bands.
• Snapshot MSI cameras, while miniaturized, capture only one spectral band per pixel, necessitating image reconstruction.
• Existing demosaicing methods struggle with increasing spectral bands, leading to artifacts and aliasing due to spatial interpolation and shallow network structures.

Purpose of the Study:

• To develop an advanced demosaicing method for reconstructing full-resolution multi-spectral images (MSI) from raw MSFA images.
• To address the challenges of artifact and aliasing in MSI reconstruction, particularly with increasing spectral bands.
• To improve spatial detail and spectral fidelity in reconstructed multi-spectral images.

Main Methods:

• A novel deep convolution neural network (CNN) was designed for multi-spectral demosaicing.
• The CNN architecture incorporates a channel attention mechanism to preserve critical spectral channel features.
• The method was evaluated using 5x5 raw mosaic images from both synthetic and real-world datasets.

Main Results:

• The proposed deep learning method significantly outperforms existing demosaicing techniques.
• Superior reconstruction quality is demonstrated in terms of preserving spatial details.
• Enhanced spectral fidelity is achieved compared to conventional methods.

Conclusions:

• The novel CNN-based demosaicing method effectively reconstructs high-quality multi-spectral images from MSFA data.
• The integration of channel attention mechanism proves beneficial for preserving spectral information.
• This approach offers a promising solution for advanced applications requiring accurate multi-spectral image reconstruction.