Dual-Channel Reconstruction Network for Image Compressive Sensing.

Area of Science:

• Signal Processing
• Image Reconstruction
• Machine Learning

Background:

• Existing compressive sensing (CS) reconstruction algorithms suffer from high computational costs and suboptimal image quality.
• There is a need for more efficient and effective CS reconstruction techniques.

Purpose of the Study:

• To propose a novel Dual-Channel Reconstruction Network (DC-Net) module for compressive sensing image reconstruction.
• To develop two CS reconstruction networks, RDC-Net and FDC-Net, utilizing the DC-Net module.
• To enhance both the quality and speed of CS image reconstruction.

Main Methods:

• Developed a Dual-Channel Reconstruction Network (DC-Net) module integrating residual and dense block channels.
• Implemented two networks: RDC-Net for traditional random under-sampling and FDC-Net for fully connected measurement matrices.
• Utilized a fully connected layer for initial image reconstruction followed by DC-Net refinement.

Main Results:

• The proposed DC-Net module significantly improves preliminary reconstructed image quality.
• FDC-Net, using a fully connected under-sampling method, represents original images more effectively.
• Both RDC-Net and FDC-Net demonstrated superior performance over state-of-the-art methods in PSNR and visual quality.
• The fused channel approach in DC-Net simultaneously enhances reconstruction performance and reduces runtime.

Conclusions:

• The proposed DC-Net module and the derived RDC-Net and FDC-Net offer a significant advancement in compressive sensing image reconstruction.
• The novel approach achieves a superior balance between reconstruction accuracy and computational efficiency.
• This work provides a promising direction for future research in efficient and high-quality CS image recovery.