HDR Image Processing Using Dense Residual and Gate-Control Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for obtaining high-dynamic-range (HDR) images face challenges due to structural artifacts caused by movement during multi-frame exposure, leading to suboptimal image quality.
Innovation Solution
An image processing method utilizing a pre-trained network model combining dense residual modules and gate-control-channel conversion modules for HDR image processing, which includes fuzzy processing and characteristic matrix analysis to enhance image detail and dynamic range, effectively addressing movement artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If multi-frame exposure scheme is used to obtain HDR images, then image dynamic range is improved, but structural artifacts appear due to lens or object movement during shooting
Solution Approach 1:
The patent segments the HDR image processing into multiple independent network modules: a first network model for HDR processing, a second network model for detail information extraction, and a third network model for low-frequency information extraction. This segmentation allows each module to specialize in specific tasks, reducing artifacts while maintaining dynamic range.
Solution Approach 2:
The patent introduces an auxiliary characteristic matrix as an intermediary that bridges the HDR characteristic matrix and the final target image. This intermediary incorporates both detail information and low-frequency information, acting as a mediator that reconciles the conflicting requirements of dynamic range and structural accuracy.
2Manufacturing precision
If multi-frame exposure shooting is performed, then HDR image quality is improved, but shooting time is extended causing movement artifacts
Solution Approach 1:
The patent performs preliminary processing by generating an auxiliary characteristic matrix from a single-frame blurred image before final HDR image synthesis. This preliminary action extracts detail and low-frequency information in advance, eliminating the need for prolonged multi-frame exposure while maintaining image quality.
Solution Approach 2:
The patent replaces the mechanical multi-frame exposure shooting system with a computational approach using deep learning networks. Instead of physically capturing multiple frames over time, the system uses neural networks to synthesize HDR images from single-frame inputs, substituting mechanical capture with computational generation.
3Measurement precision
If dense residual modules and gate-control-channel conversion modules are used, then image processing accuracy is improved, but computational complexity increases
Solution Approach 1:
The complex network model is segmented into three separate specialized networks: first network model for HDR processing, second network model for detail extraction, and third network model for low-frequency extraction. This segmentation distributes computational complexity across multiple simpler modules rather than one monolithic complex model.
Solution Approach 2:
The patent designs network modules with multi-functional capabilities. The dense residual modules and gate-control-channel conversion modules serve multiple purposes: HDR transformation, detail preservation, and frequency separation. This multi-functionality reduces overall system complexity by eliminating the need for separate specialized components.
Data Source
AI summary
An image processing method, an apparatus, an electronic device and a non-transient computer-readable storage medium. The image processing method includes: acquiring an original image; performing a fuzzy processing to the original image to obtain a fuzzy image; performing a high-dynamic-range image to the original image by using a first network model obtained by pre-training, to obtain a first characteristic matrix, wherein the first network model includes a dense residual module and a gate-control-channel conversion module; obtaining an auxiliary characteristic matrix of the original image according to the fuzzy image, wherein the auxiliary characteristic matrix includes detail information of the original image and/or low-frequency information of the original image; obtaining a target image according to the first characteristic matrix and the auxiliary characteristic matrix.