摘要
针对传统目标识别算法在目标遮挡情况下识别准确率较低的问题,提出一种基于轮廓分段特征可信度的遮挡目标识别算法。为获取有效分段,提出了基于弯曲度的轮廓分段优化算法,首先根据轮廓曲率分布进行初步分段,再根据分段的弯曲度进行分段优化,得到有效分段。然后用重要性和局部性两个参数对分段的特征可信度进行评价。在此基础上提出加权相似度匹配,将分段之间的相似度与其特征可信度结合起来,最后得到衡量识别准确率的加权相似度,获得遮挡目标的最终识别结果。通过对MPEG-7数据库进行仿真,实验证明所提出算法对于遮挡目标的识别可以获得稳定准确的识别结果,有效的提高了目标识别率。
Aiming at the problem that the traditional target recognition algorithm has low recognition accuracy in the case of target occlusion, an occlusion target recognition algorithm based on contour segmentation feature credibility is proposed. In order to obtain effective segmentation, a contour segmentation optimization algorithm based on curvature is proposed. First, preliminary segmentation was performed according to the contour curvature distribution, and then segmentation optimization was performed according to the curvature of the segmentation to obtain effective segmentation. Then the importance and locality of two parameters were used to evaluate the credibility of the segmented features. On this basis, weighted similarity matching was proposed, which combined the similarity between the segments and the feature reliability of the segments. Finally, the weighted similarity of recognition accuracy was obtained, then the final recognition result of occlusion target was obtained. By simulating the MPEG-7 database, the experimental results show that the proposed algorithm can obtain stable and accurate recognition results for the occlusion targets, and effectively improve the target recognition rate.
作者
宋建辉
杨明树
于洋
刘砚菊
SONG Jian-hui;YANG Ming-shu;YU Yang;LIU Yan-ju(School of Automation and Electrical Engineering,Shenyang University of Science and Technology,Shenyang Liaoning 110159,China)
出处
《计算机仿真》
北大核心
2022年第2期436-440,共5页
Computer Simulation
基金
辽宁省自然科学基金指导计划项目(2019-ZD-0252)。
关键词
轮廓分段
分段优化
特征可信度
加权相似度匹配
遮挡目标
Contour segmentation
Segmentation optimization
Feature credibility
Weighted similarity matching
Occlusion target
