HDR Image Reconstruction With Dual-Model Illumination Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for converting low dynamic range (LDR) images to high dynamic range (HDR) images fail to accurately restore illumination information of real scenes, resulting in low authenticity of HDR images.
Innovation Solution
A method involving two initial image processing models, where a low dynamic range image is input into each model to generate a high dynamic range image and a coefficient, with a loss function used to train the models, incorporating down-sampling and structural feature analysis to enhance image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single image processing model is used for LDR to HDR conversion, then the device complexity is low, but the authenticity and quality of the restored HDR image is insufficient
Solution Approach 1:
The patent divides the single image processing model into two separate models: a first image processing model for initial HDR reconstruction and a second image processing model for generating weight coefficients. This segmentation allows each model to specialize in specific tasks, improving overall HDR image authenticity while managing complexity through functional decomposition
Solution Approach 2:
The patent introduces a new dimension by adding weight coefficient generation as a separate processing stage. The second model operates in this new dimensional space to produce coefficients that modulate the output of the first model, thereby enhancing HDR image quality without directly increasing the complexity of the core reconstruction model
2Productivity
If down-sampling processing is applied to the low dynamic range image, then the processing speed is improved, but the image detail information is lost
Solution Approach 1:
The patent applies down-sampling as a preliminary action before feeding the image into the second image processing model. This pre-processing step reduces computational complexity for the coefficient generation task while the subsequent processing stages restore and preserve essential image details through the trained neural networks
Solution Approach 2:
The down-sampling operation is applied selectively to specific regions or aspects of the image processing pipeline. By downsampling only for certain processing stages (such as coefficient generation) while maintaining full resolution for others (such as final HDR output), the patent achieves a balance between processing speed and detail preservation
Data Source
AI summary
A method for training an image processing model includes: inputting a low dynamic range image into a first initial image processing model, so that a high dynamic range reconstruction processing is performed on the low dynamic range image to generate a first high dynamic range image; inputting the low dynamic range image into a second initial image processing model to generate a first coefficient; generating a second high dynamic range image according to the first high dynamic range image and the first coefficient; generating a loss function according to data pairs of the second high dynamic range image and a real high dynamic range image, the real high dynamic range image corresponding to the low dynamic range image; and training the first initial image processing model and the second initial image processing model with the loss function.