一种基于局部特征融合的图像多模态相似度识别方法
By constructing a multimodal similarity network, dynamically adjusting attention weights, and training model parameters step by step, the problems of feature fusion and modal differences in image similarity learning are solved, achieving more accurate and flexible image similarity recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2024-07-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively integrate different feature information in image similarity learning, and the differences and scarcity of data across different modalities limit the accuracy and scope of similarity calculation.
A multimodal image similarity recognition method based on local feature fusion is adopted. By constructing a multimodal similarity network, the attention weight is dynamically adjusted using global feature modules, regional feature modules, fusion feature modules, clustering modules, attention modules, and similarity modules. The model parameters are trained step by step, and the block information and overall vision of the image are fused to improve the accuracy of similarity recognition.
It improves the accuracy and flexibility of image similarity recognition, can adapt to different scenario requirements, dynamically adjusts attention weights, adapts to complex image similarity calculations, and enhances the model's expressive power and interpretability.
Smart Images
Figure CN118941901B_ABST