一种外观侵权的检测方法、装置及存储介质
By constructing an appearance infringement detection model and utilizing key point extraction and attention mechanism encoding, the problem of difficulty in infringement determination caused by appearance image deformation is solved, and efficient and accurate infringement detection is achieved.
CN115482398BActive Publication Date: 2026-07-17GONGDAO NETWORK TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GONGDAO NETWORK TECH CO LTD
- Filing Date
- 2022-09-08
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing technologies are insufficient to accurately detect copyright infringement of appearance images, especially when the image to be detected is distorted, leading to failure in the determination of copyright infringement.
Method used
By constructing an appearance infringement detection model, using key point extraction, attention mechanism encoding, and multi-layer neural networks, the model calculates the key point similarity between the object to be detected and the candidate object to determine whether appearance infringement exists.
Benefits of technology
It improves the accuracy of appearance image matching, enabling it to quickly and effectively identify infringing images from massive amounts of data, thus protecting the rights of designers.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115482398B_ABST
Abstract
本申请提供一种外观侵权的检测方法、装置及存储介质。该方法可以包括：获取待检测对象的外观图像，并提取该外观图像中的待检测关键点；构建初始待检测关键点局部特征集，其中包含每个待检测关键点对应的初始待检测关键点局部特征；基于注意力机制确定每个待检测关键点对应的注意力权重，根据注意力权重对初始待检测关键点局部特征进行编码，生成目标待检测关键点局部特征集；基于目标待检测关键点局部特征集和目标候选关键点局部特征集，计算每个待检测关键点与每个候选关键点之间的相似度；当相似关键点比例超过预设比例阈值时，判定待检测对象对目标候选对象形成外观侵权。通过本申请的技术方案，可以从海量数据中快速准确地寻找到侵权图像。
Need to check novelty before this filing date? Find Prior Art