Multi-Frame Image Processing Training Framework
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile electronic devices' cameras suffer from noise and image blur in low-light settings due to increased light sensitivity and hand shake during long exposure scenarios, degrading image quality.
Innovation Solution
A machine learning model training framework for multi-frame image processing that generates alignment and blending maps to align and combine multiple image frames, creating a ground truth image pair for training, which reduces image distortion and noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple image frames are captured and combined using traditional multi-frame processing, then image quality is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent pre-generates alignment maps and blending maps from a first set of image frames before processing the second set of image frames. This preliminary action allows subsequent image processing to reuse these maps, significantly reducing processing time while maintaining image quality improvements from multi-frame processing
Solution Approach 2:
The patent implements a hybrid processing approach where alignment and blending maps are dynamically generated for the first image set, then reused for subsequent image sets. This dynamic strategy adapts the processing intensity based on whether images are from the first or subsequent captures, optimizing the balance between quality and processing time
2Manufacturing precision
If multiple image frames are captured and combined using traditional multi-frame processing, then image quality is improved, but device complexity increases
Solution Approach 1:
The patent pre-generates alignment maps and blending maps from a first set of image frames before processing the second set of image frames. This preliminary action allows subsequent image processing to reuse these maps, significantly reducing processing time while maintaining image quality improvements from multi-frame processing
Solution Approach 2:
The patent divides the multi-frame processing into distinct stages: first generating alignment and blending maps from one set of images, then applying these maps to process additional image sets. This segmentation separates the complex map generation from the simpler map application, reducing overall processing complexity
3Manufacturing precision
If machine learning model training is performed to reduce noise and distortion, then image quality is improved, but training data generation complexity increases
Solution Approach 1:
The patent pre-generates alignment maps and blending maps from a first set of image frames before processing the second set of image frames. This preliminary action allows subsequent image processing to reuse these maps, significantly reducing processing time while maintaining image quality improvements from multi-frame processing
Solution Approach 2:
The patent uses alignment maps and blending maps as intermediary data structures that facilitate the creation of training datasets. These maps serve as mediators between the raw image frames and the final training data, simplifying the overall process by providing a structured intermediate representation
Data Source
AI summary
A method for training data generation includes obtaining a first set of image frames of a scene and a second set of image frames of the scene using multiple exposure settings. The method also includes generating an alignment map, a blending map, and an input image using the first set of image frames. The method further includes generating a ground truth image using the alignment map, the blending map, and the second set of image frames. In addition, the method includes using the ground truth image and the input image as an image pair in a training dataset when training a machine learning model to reduce image distortion and noise.