Synthetic Bracketing for Real-Time Edge Exposure Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-cost cameras often produce visually unappealing images and videos due to poor exposure, especially in challenging lighting conditions, and existing exposure correction methods are computationally inefficient, requiring multiple captures or costly sensors, and fail to correct exposure in real-time on edge devices.
Innovation Solution
A deep learning-based method synthesizes multiple images with different exposures from a single input using local and global transformations, avoiding motion artifacts and computational overhead, and allows for real-time exposure correction on edge devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple images are acquired by camera for exposure bracketing, then exposure correction quality is improved, but acquisition time increases and motion artifacts occur
Solution Approach 1:
The system performs exposure correction by synthesizing multiple exposure images from a single captured image through computational processing. The encoder extracts features from the single input image, and the decoder generates multiple exposure versions by applying learned transformations, eliminating the need for actual multiple captures and thus resolving the time loss contradiction.
Solution Approach 2:
The system creates synthetic copies of the captured image with different exposure levels by learning the mapping from a single image to multiple exposure images. The neural network generates exposure-bracketed versions as computational copies, avoiding the need for physical multiple captures while maintaining exposure correction quality.
2Manufacturing precision
If multiple sensors are used for exposure bracketing, then exposure correction quality is improved, but device cost increases
Solution Approach 1:
The system replaces the mechanical/optical approach of using multiple physical sensors with a computational approach using a single sensor and neural network processing. The encoder-decoder architecture learns to synthesize multiple exposure images from one captured image, substituting hardware complexity with software intelligence.
Solution Approach 2:
A single camera sensor performs multiple functions by capturing one image that is then computationally transformed into multiple exposure versions. The neural network enables the single sensor to effectively serve the role of multiple sensors through learned image synthesis, reducing device complexity while maintaining quality.
3Reliability
If traditional exposure correction methods are used, then exposure issues are addressed, but computational efficiency is poor for real-time processing
Solution Approach 1:
The system performs preliminary learning during a training phase where the encoder-decoder network learns the mapping between single images and multiple exposure images. During real-time operation, the pre-trained model efficiently processes images through forward propagation, achieving both reliability and speed by separating the computationally intensive learning phase from the efficient inference phase.
Data Source
AI summary
Systems and methods are disclosed related to synthetic bracketing for exposure correction. A deep learning based method and system produces a set of differently exposed images from a single input image. The images in the set may be combined to produce an output image with improved global and local exposure compared with the input image. An image encoder applies learned parameters to each input image to generate a set of image features including local exposure estimates for each of two or more regions of the input image and a low resolution latent representation of the input image. A decoder receives the local exposure estimates, the latent representation, and target enhancements that are processed to generate synthesized transformations. When applied to the input image, the synthesized transformations produce the set of transformed images. Each transformed image is a version of the input image synthesized to correspond to a respective target enhancement.