Performance enhancement of the branch length similarity entropy descriptor for shape recognition by introducing critical points

详细信息    查看全文

• 作者：Sang-Hee Lee ; Seung-Ho Kang
• 关键词：Branch length similarity entropy ; Critical points ; Shape descriptor
• 刊名：Journal of the Korean Physical Society
• 出版年：2016
• 出版时间：October 2016
• 年：2016
• 卷：69
• 期：7
• 页码：1254-1262
• 全文大小：927 KB
• 刊物主题：Physics, general; Theoretical, Mathematical and Computational Physics; Particle and Nuclear Physics;
• 出版者：Springer Netherlands
• ISSN：1976-8524
• 卷排序：69

文摘

In previous studies, we showed that the branch length similarity (BLS) entropy profile could be used successfully for the recognition of shapes such as battle tanks, facial expressions, and butterflies. In the present study, we introduce critical points defined as a set of distinguishing points with high curvature to the BLS entropy profile in order to improve the shape recognition. In order to generate a given number of critical points from the shape, we propose a critical point detection method. Furthermore, we show the invariant properties of the BLS entropy descriptor. To evaluate the effects of critical points on the shape recognition of the BLS entropy descriptor, we performed a butterfly classification experiment against a real image data set, and we conducted performance comparisons with other point detection methods. In addition, the performance of the BLS entropy descriptor computed using the critical points was compared with those of other well-known descriptors such as the Fourier descriptor using three machine learning techniques, the Bayesian classifier, the multi-layer perceptron and the support vector machine. The results show that the BLS entropy descriptor outperforms other well-known descriptors.