Low Light Image Enhancement via External Memory Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for low light image enhancement, such as histogram equalization, Retinex models, and deep learning approaches, often result in undesirable outcomes like local illumination issues, under/over exposure, and high computational costs.
Innovation Solution
A system and method utilizing an external memory network for low light image enhancement, where an image memory stores sample-specific properties of a training dataset to facilitate adaptive adjustments during testing, improving the quality of enhanced images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep learning networks are used for low light image enhancement, then image quality is improved, but computational cost and time increase significantly
Solution Approach 1:
The network is divided into specialized sub-networks for illumination and reflectance components, allowing independent optimization and reduced computational complexity while maintaining image quality enhancement
Solution Approach 2:
The network uses pre-calculated hyperparameters and reference image correlations established during training, eliminating the need for complex real-time parameter adjustment and reducing computational burden during inference
2Ease of operation
If hyperparameters are pre-calculated and determined by users, then model application is simplified, but practical usability is limited due to extra computing requirements
Solution Approach 1:
The network automatically determines optimal hyperparameters through self-attention mechanisms and adaptive feature fusion, eliminating the need for manual pre-calculation and reducing computational overhead during deployment
3Adaptability or versatility
If transformer models with self-attention are applied to low light image enhancement, then handling of long range dependencies is improved, but complexity and computational expense increase
Solution Approach 1:
The patent extracts only the essential self-attention components needed for low light enhancement, removing unnecessary complexity from full transformer architectures while retaining the ability to handle long-range dependencies in image data
Data Source
AI summary
A system and a method for a computer implemented method of image enhancement includes the steps of receiving an input image, wherein the input image is a low light image, processing the input image, by a pre trained image enhancer, to generate an initial enhanced image, accessing, from an image memory, a response value corresponding to a sample specific property a normal image, generating an adjustment factor based on the response value from the image memory, generating a final enhanced image by applying the adjustment factor to the initial enhanced image.