摘要
Abstract
As the deep learning technology continues to advance,the significance of feature matching algorithms in the field of computer vision is increasingly prominent.The traditional SuperGlue algorithm has exhibited excellent performance in accuracy of feature matching,yet there is still room for improvement in efficiency and accuracy when dealing with oral images with low illumination and complex texture.In view of the above,an oral feature matching algorithm based on SuperGlue optimized by confidence strategy is proposed.Initially,by incorporating a confidence scoring mechanism,the possibility of feature point pair matching can be assessed more accurately,allowing the algorithm to focus on the most likely correct matching point pairs.Subsequently,a dynamic confidence threshold adjustment strategy is introduced,which adjusts thresholds automatically according to the characteristics and distribution of feature points of oral image pairs,so as to balance the quantity and quality of the matching.After a series of experimental validations,the improved algorithm has shown significant advancements in both efficiency and accuracy,particularly demonstrating better robustness in situations that the images have diverse feature points and are of uneven quality.The successful implementation of the designed algorithm provides a new approach for feature matching in the field of oral vision,so it holds significant theoretical value and practical application prospects.关键词
口腔图像/特征匹配/SuperGlue/置信度评分/动态阈值调整/深度学习Key words
oral image/feature matching/SuperGlue/confidence scoring/dynamic threshold adjustment/deep learning分类
电子信息工程
