图像处理方法、模型的训练方法、装置、系统及介质
By processing the template image and replacing the attribute feature information of the target object, a target image with high clarity and realism is generated, thus solving the problems of distortion and aberration in image processing.
CN116229203BActive Publication Date: 2026-07-17JINGDONG TECH HLDG CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JINGDONG TECH HLDG CO LTD
- Filing Date
- 2023-01-10
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
The target image generated after image processing in existing technologies suffers from distortion and deformation.
Method used
By processing the template image, an image mask and attribute feature information are generated. The attribute feature information of the target object is replaced by the image processing model to generate the target image. The perfect reconstruction of the non-target object region is guaranteed by reversibility.
Benefits of technology
It achieves high clarity and realism of the target image, allows for precise editing of the identity-related regions of the target object, and preserves the original features of non-target object regions.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116229203B_ABST
Abstract
本公开提供了一种图像处理方法、图像处理模型的训练方法、装置、计算机系统、可读存储介质及计算机程序产品，涉及人工智能技术领域。该图像处理方法包括：对模板图像进行处理，得到包括第一目标对象的图像掩膜和模板图像的属性特征信息，第一目标对象为模板图像中的对象；基于模板图像、图像掩膜和模板图像的属性特征信息，得到目标隐变量，目标隐变量包括第一目标对象的属性特征信息和目标模板图像信息，目标模板图像信息包括模板图像中的除第一目标对象的属性特征信息外的信息；基于图像掩膜，将目标隐变量中的第一目标对象的属性特征信息替换为第二目标对象的属性特征信息，得到目标图像。
Need to check novelty before this filing date? Find Prior Art