Burst Image Restoration via Anchor-Based Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image restoration methods face challenges in efficiently restoring high-quality images from burst images, particularly in environments with low illuminance or when individual images have deteriorated quality, due to the need for center line alignment which increases calculation load.
Innovation Solution
An image restoration apparatus that selects an anchor image from a burst image set and uses its anchor information to execute a neural network model, thereby generating a restored image without the need for center line alignment, reducing calculation load and enhancing image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If center line alignment is performed to restore burst images, then image quality can be improved, but calculation load increases
Solution Approach 1:
The patent extracts and removes the center line alignment step from the traditional image restoration pipeline. By using anchor-based feature extraction and attention mechanisms, the method directly processes burst images without requiring precise geometric alignment, thereby reducing computational complexity while maintaining restoration quality
Solution Approach 2:
The patent replaces the mechanical/geometric alignment process (center line alignment) with a neural network-based feature extraction and attention mechanism system. This substitution allows the system to handle image restoration through learned features rather than rigid geometric constraints, reducing calculation load
2Manufacturing precision
If neural network is trained for specialized image restoration, then restoration accuracy improves, but generalization ability to unseen patterns decreases
Solution Approach 1:
The patent designs a neural network architecture that performs multiple functions: feature extraction, attention mechanism, and image restoration in a unified model. The anchor-based approach allows the network to handle various image quality issues (noise, blur, low light) simultaneously, enhancing both specialization and generalization capabilities
Solution Approach 2:
The patent employs dynamic anchor adjustment and attention weight optimization to adapt the network to different input conditions. By changing parameters such as anchor positions, attention masks, and learning rates during inference, the model maintains high accuracy across diverse image patterns without requiring separate specialized models
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The application concerns an image restoration method. The method includes determining an anchor image based on individual images of a burst image set, executing a feature extraction network based on the burst image set while using anchor information of the anchor image, and generating a restored image based on a feature map corresponding to an output of the feature extraction network. The application further concerns a computer program comprising instructions to perform said method and a non-transitory computer-readable storage medium storing said computer program. The application further concerns an apparatus comprising one or more processors configured for performing said method.