Deep Likelihood Network for Image Restoration With Multiple Degradation Levels.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Convolutional neural networks (CNNs) excel at image restoration.
• Current CNNs struggle with varying degradation levels, limiting their real-world application.
• A need exists for robust image restoration models adaptable to diverse degradation scenarios.

Purpose of the Study:

• To develop a Deep Likelihood Network (DL-Net) that generalizes image restoration models.
• To enable off-the-shelf networks to perform effectively across a spectrum of degradation levels.
• To improve the adaptability and robustness of deep learning models in image restoration.

Main Methods:

• Proposed a Deep Likelihood Network (DL-Net) by modifying existing CNN architectures.
• Introduced a simple recursive module derived from a fidelity term.
• Designed the module to disentangle computations for multiple degradation levels.

Main Results:

• DL-Net demonstrated effectiveness in generalizing image restoration.
• Experiments showed successful application on image inpainting, interpolation, and super-resolution tasks.
• The proposed method significantly improved performance across various degradation settings.

Conclusions:

• DL-Net successfully generalizes image restoration networks to multiple degradation levels.
• The recursive module effectively handles varying degradation, enhancing model adaptability.
• This work offers a promising direction for more versatile and robust image restoration solutions.