用于高分辨率人脸图像的多样化编辑方法及系统
By generating multiple sampling noise and low-resolution image generation modules, combined with neural networks and face attribute segmentation networks, the problem of diversified editing of high-resolution face images is solved, realizing diversified editing of high-resolution face images, especially the editing of glasses and bangs attributes.
CN116152391BActive Publication Date: 2026-07-17SHANGHAI ARTIFICIAL INTELLIGENCE INNOVATION CENT
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI ARTIFICIAL INTELLIGENCE INNOVATION CENT
- Filing Date
- 2022-10-25
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing technologies struggle to achieve diverse editing of high-resolution facial images, especially through generative adversarial networks.
Method used
By setting target attributes and generating multiple sampling noises, a neural network is constructed using a low-resolution image generation module and a high-resolution image generation module for supervised learning. This is then combined with a face attribute segmentation network for optimization, enabling diverse editing of high-resolution face images.
Benefits of technology
It enables diverse editing of high-resolution face images, especially for editing glasses and bangs attributes, while keeping other image attributes unchanged.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116152391B_ABST
Abstract
本发明涉及人脸图像编辑技术领域，提出一种用于高分辨率人脸图像的多样化编辑方法及系统。该方法包括设置目标属性，并且根据所述目标属性生成多个采样噪声；提供低分辨率图像生成模块，向所述低分辨率图像生成模块中输入人脸图像和所述多个采样噪声以生成多个低分辨率结果图像；提供高分辨率图像生成模块，将所述多个低分辨率结果图像反演至所述高分辨率图像生成模块中以生成训练数据集；构造神经网络，通过所述神经网络从所述采样噪声拟合残差量；以及通过所述训练数据集对所述神经网络进行监督学习。通过本发明可以实现对于高分辨率人脸图像的多样化编辑并且可以保持生成图像的面部目标属性以外的其他属性不变化。
Need to check novelty before this filing date? Find Prior Art