Multi-Scale Deep Residual Learning-Based Single Image Haze Removal via Image Decomposition.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Outdoor images are often degraded by atmospheric conditions like haze, impacting visibility.
• Existing methods often use end-to-end learning, which can be complex and prone to artifacts.

Purpose of the Study:

• To propose a novel deep learning architecture, MSRL-DehazeNet, for effective single image haze removal.
• To reformulate haze removal as the restoration of an image's base component.

Main Methods:

• The proposed MSRL-DehazeNet utilizes multi-scale residual learning (MSRL) and image decomposition.
• It decomposes hazy images into base and detail components, learning mappings for the base component restoration.
• A separate Convolutional Neural Network (CNN) enhances the detail component, preserving features across layers to prevent color distortion.

Main Results:

• The MSRL-DehazeNet framework successfully removes haze from single images.
• Experimental results show superior performance compared to existing state-of-the-art approaches.
• The method effectively integrates restored base and enhanced detail components for a final clear image.

Conclusions:

• The proposed MSRL-DehazeNet offers an effective solution for single image haze removal.
• The approach demonstrates robustness in restoring image quality without color distortion.
• This method advances the field of image restoration for degraded outdoor scenes.