Neural Network Style Transfer Super-Resolution Branches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using deep learning technologies struggle to transform images into high-resolution outputs with a target artistic style, as previous solutions either produce low-quality style enhancements or are limited to processing lower resolution images.
Innovation Solution
The integration of a convolutional neural network with super-resolution technology to create a branch that simultaneously performs style transfer and super-resolution, enabling the output of both low-resolution and high-resolution stylized images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing deep learning image processing methods are used for style transfer, then artistic style enhancement is achieved, but image resolution remains low
Solution Approach 1:
The patent divides the image processing task into two separate branches: a first branch for style transfer and a second branch for super-resolution. This segmentation allows each branch to specialize in its function, with the first branch maintaining artistic style enhancement while the second branch independently improving image resolution, thereby resolving the contradiction between these two requirements.
Solution Approach 2:
The patent merges the outputs of the first branch (stylized low-resolution image) and the second branch (high-resolution image) through a addition operation. This combining approach integrates the style enhancement capabilities with the super-resolution capabilities, producing a final high-resolution stylized image that satisfies both requirements simultaneously.
2Productivity
If separate processing branches are used for style transfer and super-resolution, then processing efficiency is improved, but device complexity increases
Solution Approach 1:
The patent implements a multi-functional neural network framework where the first branch handles style transfer and the second branch handles super-resolution. Both branches share common components such as convolutional layers, activation functions, and processing pipelines. This universality reduces overall system complexity compared to implementing completely separate systems, while still providing specialized processing for each function to improve efficiency.
Data Source
AI summary
An image processing method implemented by a computing device is described herein, which includes acquiring an image to be processed and a target style of image, the image to be processed being an image of a second resolution level, and inputting the image to be processed and the target style into a trained image processing neural network for image processing to obtain a target image of the target style, the target image being an image of a first resolution level. The resolution of the image of the first resolution level is higher than that of the image of the second resolution level.