An efficient lightweight network for image denoising using progressive residual and convolutional attention feature...

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Deep learning excels at image denoising but often requires deep networks, leading to high computational costs.
• Existing methods struggle with balancing noise reduction and feature preservation, especially for real-world noise.

Purpose of the Study:

• To develop a lightweight, efficient deep learning network for superior image denoising.
• To address the limitations of excessive network depth and computational burden in current denoising techniques.

Main Methods:

• Proposed a novel lightweight progressive residual and attention mechanism fusion network.
• Utilized dense blocks (DB) for noise distribution discernment and local feature extraction.
• Implemented a progressive strategy to fuse shallow and deep features, incorporating a convolutional attention feature fusion module (CAFFM).

Main Results:

• The network demonstrated exceptional efficacy in denoising both Gaussian and real-world image noise across various levels (15-50).
• Achieved superior performance compared to over 20 existing methods on six diverse datasets, evidenced by higher PSNR, SSIM, and FSIMc values.
• Successfully preserved essential image features such as edges and textures.

Conclusions:

• The proposed network offers a significant advancement in image denoising, providing high performance with reduced computational load.
• The model's ability to maintain image fidelity makes it suitable for various image-centric applications.
• This research contributes a notable progression in image processing techniques through an innovative network architecture.