摘要
随着深度学习技术的不断进步,特征匹配算法在计算机视觉领域的重要性日益凸显。传统的SuperGlue算法在特征匹配准确度上已经表现出了优异性能,但在处理低光照和纹理复杂的口腔图像时,其效率和准确性仍有提升的空间。针对上述问题,文中提出一种基于置信度策略优化的SuperGlue口腔特征匹配算法。首先,通过引入一个置信度评分机制,可以更准确地评估特征点对之间的匹配可能性,让算法聚焦于更可能正确的匹配点对;其次,提出动态置信度阈值调整策略,根据口腔图像对的特性和特征点分布自动调整阈值,以达到匹配数量与质量平衡的目的。经过一系列实验验证,改进后的算法在效率和准确性方面都取得了显著提升,尤其是在特征点多样性和图像质量不一的情况下,展现了更好的鲁棒性。设计算法的成功实现,为口腔视觉领域中的特征匹配问题提供了一种新的解决思路,具有重要的理论价值和实际应用前景。
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.
作者
郭瑀璐
李占利
GUO Yulu;LI Zhanli(College of Computer Science and Technology,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《现代电子技术》
北大核心
2024年第23期22-28,共7页
Modern Electronics Technique
