Abstract
In computer science, and especially in computer vision, the contour of an object is used to describe its features; thus, the shape descriptor plays an indispensable role in target detection and recognition. Further, Fourier is an important mathematical description method, and the Fourier transform of a shape contour has symmetry. This paper will demonstrate the symmetry of shape contour in the frequency domain. In recent years, increasing numbers of shape descriptors have come to the fore, but many descriptors ignore the details of shape. It is found that the most fundamental reason affecting the performance of shape descriptors is structural restraints, especially feature structure restraint. Therefore, in this paper, the restraint of feature structure that intrinsically deteriorates recognition performance is shown, and a fast shape recognition method via the Bone Point Segment (BPS) restraint reduction is proposed. An approach using the inner distance to find bone shapes and segment the shape contour by these bones is proposed. Then, Fourier transform is performed on each segment to form the shape feature. Finally, the restraints of the shape feature are reduced in order to build a more effective shape feature. What is commendable is that its discriminability and robustness is strong, the process is simple, and matching speed is fast. More importantly, the experiment results show that the shape descriptor has higher recognition accuracy and the matching speed runs up to more than 1000 times faster than the existing description methods like CBW and TCD. More importantly, it is worth noting that the recognition accuracy approaches 100% in the self-build dataset.
Funder
National Natural Science Foundation of China
Natural Science Basic Research Program of Shaanxi Province of China
Subject
Physics and Astronomy (miscellaneous),General Mathematics,Chemistry (miscellaneous),Computer Science (miscellaneous)
Reference43 articles.
1. Learning context-sensitive shape similarity by graph transduction;Bai;IEEE Trans. Pattern Anal. Mach. Intell.,2009
2. Sparse Contextual Activation for Efficient Visual Re-Ranking
3. Shape matching and object recognition using shape contexts
4. Applying Machine Learning to Ultrafast Shape Recognition in Ligand-Based Virtual Screening
5. A multi-angle shape descriptor with the distance ratio to vertical bounding rectangles;Li;Proceedings of the 2021 International Conference on Content-Based Multimedia Indexing (CBMI),2021
Cited by
2 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献