Abstract

It is a challenging but important task to design an abstract and robust shape descriptor for deformable shapes in a highly discriminative way. However, most of the existing low-level descriptors are designed using hand-crafted features and are sensitive to local variations and large deformations. To solve this problem, in this Letter, we propose a shape descriptor based on the Radon transform and SimNet for shape recognition. It effectively overcomes structural barriers, such as rigid or non-rigid transformations, irregular topologies between shape features, and similarity learning. Specifically, we use the Radon features of the objects as the input to the network and calculate the similarity using SimNet. Deformation of objects may affect Radon feature maps, and SimNet can overcome such effects to reduce the loss of information. Compared with SimNet, which consumes the original images as the input, our method shows higher performance.

© 2022 Optica Publishing Group

Digital image correlation based on the pre-deformation assist strategy and an optimized checkerboard-like speckle pattern for large deformation measurement

Jingchen Ye, Xiaochuan Zhang, Hongwei Ji, and Yujiang Liu
Opt. Lett. 47(24) 6456-6459 (2022)

Anti-motion blur single-pixel imaging with calibrated radon spectrum

Wei Yang, Dongfeng Shi, Kai Han, Zijun Guo, Yafeng Chen, Jian Huang, Hubin Ling, and Yingjian Wang
Opt. Lett. 47(12) 3123-3126 (2022)

Image-free multi-character recognition

Liheng Bian, Huayi Wang, Chunli Zhu, and Jun Zhang
Opt. Lett. 47(6) 1343-1346 (2022)

Invariant pattern recognition, symmetry, and Radon transforms

David A. Pintsov
J. Opt. Soc. Am. A 6(10) 1544-1554 (1989)

Deep-learning-based 3D object salient detection via light-field integral imaging

Ying Li, Tianhao Wang, Yanheng Liao, Da-Hai Li, and Xiaowei Li
Opt. Lett. 47(7) 1758-1761 (2022)