Attention-guided CNN for image denoising.

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Deep convolutional neural networks (CNNs) are widely used in low-level computer vision tasks.
• Increasing CNN depth can weaken shallow layer influence, impacting performance.
• Existing methods often focus on increasing network depth for improved denoising.

Purpose of the Study:

• To propose an attention-guided denoising convolutional neural network (ADNet) for improved image denoising.
• To address the performance degradation in very deep CNNs by incorporating an attention mechanism.
• To develop an efficient and effective model for synthetic, real noisy, and blind image denoising.

Main Methods:

• The proposed ADNet comprises a sparse block (SB), feature enhancement block (FEB), attention block (AB), and reconstruction block (RB).
• The SB uses dilated and common convolutions for noise removal, balancing performance and efficiency.
• The FEB integrates global and local features, enhanced by the AB for complex noise extraction, improving efficiency and reducing training complexity.

Main Results:

• ADNet demonstrates strong performance in quantitative and qualitative evaluations across synthetic, real noisy, and blind denoising tasks.
• The integrated FEB and AB effectively extract noise from complex backgrounds, particularly in real-world noisy images.
• The model achieves a good tradeoff between performance and efficiency.

Conclusions:

• The proposed attention-guided denoising convolutional neural network (ADNet) is effective for various image denoising challenges.
• ADNet offers an efficient and robust solution for complex noisy image scenarios.
• The model's architecture successfully mitigates the impact of increased depth on shallow layer influence.