Generative Network Branching for Image Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies face challenges in enhancing image quality, particularly with low-quality images from mobile devices, due to limitations in hardware and environmental noise, which degrade image quality and contrast.
Innovation Solution
An image processing method utilizing a generative network with branching and weight sharing processes, including multiple sub-networks and convolutional layers, to enhance image quality by generating high-quality output images similar to those from digital single lens reflex cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a standard convolutional neural network is used for image processing, then the network can process images, but the number of parameters is large and processing speed is slow
Solution Approach 1:
The network is divided into multiple sub-networks, each responsible for processing different color channels (RGB and YUV). This segmentation reduces the computational burden on each sub-network, allowing for faster processing while maintaining overall network functionality.
Solution Approach 2:
The network implements multi-functionality by processing both RGB and YUV color spaces simultaneously through different sub-networks. This allows the system to leverage the advantages of both color representations (RGB for color accuracy, YUV for luminance-detail separation) without requiring separate processing pipelines.
2Manufacturing precision
If image quality enhancement is pursued to achieve DSLR-quality images, then image quality improves, but processing complexity increases
Solution Approach 1:
The network segments image processing into distinct sub-networks for different color channels and processing stages. This allows complex image quality enhancement to be achieved through multiple specialized, manageable sub-networks rather than one monolithic complex network.
Solution Approach 2:
The network transforms the problem from a single-color-space processing task to a multi-dimensional approach by simultaneously processing RGB and YUV color spaces through different sub-networks, then fusing the results to achieve superior image quality.
3Measurement precision
If more convolutional layers are added to improve feature extraction, then feature extraction capability improves, but training convergence becomes slower
Solution Approach 1:
The network divides deep feature extraction into multiple shallower sub-networks processing different color channels. Each sub-network can be trained independently and converges faster, while collectively they achieve the feature extraction capability of a much deeper network.
Data Source
AI summary
Disclosed are an image processing method and device, a training method of a neural network and a storage medium. The image processing method includes: obtaining an input image, and processing the input image by using a generative network to generate an output image. The generate network includes a first sub-network and at least one second sub-network, and the processing the input image by using the generative network to generate the output image includes, processing the input image by using the first sub-network to obtain a plurality of first feature images; performing a branching process and a weight sharing process on the plurality of first feature images by using the at least one second sub-network to obtain a plurality of second feature images; and processing the plurality of second feature images to obtain the output image.