Multi-Exposure Image Processing for Dark-Area Color Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods fail to effectively correct color cast in dark areas of images captured in low illumination or high-dynamic-range scenes, leading to degraded user experience.
Innovation Solution
An image processing method involving multiple frames of exposure images with varying exposure times, utilizing a deep learning network for noise reduction and demosaicing, followed by conversion to the YUV domain and style transfer using a trained style transfer network to correct color cast in dark areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple exposure images with different exposure times are captured to improve dark area quality, then color fidelity in dark areas improves, but processing complexity and time increase
Solution Approach 1:
The patent performs style transfer processing on multiple exposure images before final fusion. By pre-processing the images to transfer styles from long-exposure images to short-exposure images, the method prepares color-corrected versions in advance, which then can be easily fused with the original images. This preliminary style transfer action simplifies the overall processing pipeline while maintaining color fidelity in dark areas.
Solution Approach 2:
The patent introduces a style transfer network as an intermediary processing step between image capture and final fusion. This intermediary component extracts color information from long-exposure images and transfers it to short-exposure images, acting as a mediator that bridges the gap between different exposure images. The style transfer network serves as an intermediate representation that facilitates the fusion process while preserving color accuracy.
2Manufacturing precision
If deep learning network processing is applied to reduce noise and improve image quality, then image quality improves, but processing time and computational resources increase
Solution Approach 1:
The patent divides the image processing task into multiple specialized deep learning networks, each handling specific functions: one network for noise reduction, another for style transfer, and potentially others for different processing stages. This segmentation allows each network to be optimized for its specific function, improving overall efficiency while maintaining image quality. The modular approach enables selective application of processing steps.
Solution Approach 2:
The patent applies deep learning processing selectively rather than uniformly to all images. Style transfer is performed on multiple exposure images, but the full deep learning pipeline may be applied only when needed (e.g., when color cast is detected). This partial application reduces unnecessary computational overhead while maintaining image quality where it matters most.
3Measurement precision
If style transfer is performed on multiple exposure images to correct color cast, then color accuracy in dark areas improves, but computational complexity increases
Solution Approach 1:
The patent applies style transfer processing specifically to regions where color cast is problematic, particularly in dark areas of images. Rather than uniformly processing all pixels, the method focuses computational resources on areas that need correction, using the style information from long-exposure images to adjust colors in dark regions. This local application reduces overall computational complexity while maintaining color accuracy where needed.
Solution Approach 2:
The patent creates style representations from long-exposure images and applies these style copies to short-exposure images. Instead of performing complex direct color correction, the method copies color characteristics from the long-exposure image (which has accurate colors due to sufficient light) and transfers them to the short-exposure image. This copying approach simplifies the computational complexity while achieving color accuracy.
Data Source
Figure 1(a)~2
Figure 3(a)~4
Figure 5~7
AI summary
This application relates to the field of image processing, and provides an image processing method and a related device thereof. The method includes: displaying a first interface, where the first interface includes a first control; detecting a first operation on the first control; obtaining a plurality of frames of exposure images in response to the first operation; and performing style transfer on the plurality of frames of exposure images to obtain a target image. The image processing method provided in this application can effectively improve a color of an image obtained by fusing a plurality of frames of images captured in a scene with low illumination or a high-dynamic-range scene.