Low-Light Image Enhancement via Reinforcement Learning and Aesthetic Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning techniques for low-light image enhancement primarily focus on underexposed regions, neglecting normally exposed or overexposed areas in backlit and uneven lighting scenarios, and rely on objective evaluation metrics that often disregard subjective user assessment.
Innovation Solution
A method utilizing reinforcement learning and aesthetic evaluation to enhance low-light images, which involves generating images under different lighting conditions, constructing a training dataset, initializing policy and value networks, updating them based on no-reference and aesthetic assessment reward scores, and completing model training to output enhanced images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep learning techniques use reference-based loss functions (L1, L2, SSIM) to guide model training, then the enhancement fidelity to normal-light images is improved, but the subjective user assessment quality deteriorates
Solution Approach 1:
The patent introduces aesthetic assessment feedback mechanisms that provide subjective quality evaluation signals during training. The system uses pre-trained aesthetic assessment models to generate feedback scores that guide the enhancement process, allowing the model to learn what produces subjectively pleasing results rather than just technical fidelity to reference images.
Solution Approach 2:
The patent employs aesthetic assessment models as intermediary components that bridge the gap between objective image enhancement metrics and subjective user perception. These intermediary models translate visual qualities into quantifiable scores that can be used as training signals, mediating between the enhancement process and user satisfaction.
2Illumination intensity
If deep learning techniques focus on enhancing underexposed regions, then the luminance of low-light images is improved, but the handling of normally exposed or overexposed regions deteriorates
Solution Approach 1:
The patent applies local quality enhancement by analyzing different regions of the image and applying appropriate enhancement strategies to each region based on its exposure characteristics. The system identifies underexposed, normally exposed, and overexposed regions separately and applies region-specific enhancement operations to preserve details in each area while maintaining overall image quality.
Solution Approach 2:
The patent implements dynamic enhancement that adapts to varying lighting conditions across different regions of the image. The enhancement parameters and operations are dynamically adjusted based on local exposure analysis, allowing the system to handle diverse lighting scenarios including underexposed, normally exposed, and overexposed regions within a single image.
Data Source
AI summary
Disclosed is a method for enhancement of a low-light image based on reinforcement learning and aesthetic evaluation. The method include: generating images of non-normal luminance under different lighting scenes, and constructing a training dataset for a reinforcement learning system based on the images; initializing the training dataset, a policy network, and a value network; updating, based on a no-reference reward score and an aesthetic assessment reward score, the policy network and the value network; completing model training and outputting an enhanced image result. By expanding the action space range defined in reinforcement learning, the enhancement operations for the input low-light image gain a greater dynamic range, offering higher flexibility for real-world scenarios and better meeting low-light image enhancement needs. Additionally, by incorporating the aesthetic quality assessment scores as part of the loss function, the enhanced image achieves better visual effects and higher subjective user evaluation scores.