Shape recognition through multi-level fusion of features and classifiers
- PDF / 888,846 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 18 Downloads / 206 Views
ORIGINAL PAPER
Shape recognition through multi‑level fusion of features and classifiers Xinming Wang1 · Weili Ding1 · Han Liu2 · Xiangsheng Huang3 Received: 14 December 2018 / Accepted: 28 March 2019 © Springer Nature Switzerland AG 2019
Abstract Shape recognition is a fundamental problem and a special type of image classification, where each shape is considered as a class. Current approaches to shape recognition mainly focus on designing low-level shape descriptors, and classify them using some machine learning approaches. To achieve effective learning of shape features, it is essential to ensure that a comprehensive set of high quality features can be extracted from the original shape data. Thus, we have been motivated to develop methods of fusion of features and classifiers for advancing the classification performance. In this paper, we propose a multi-level framework for fusion of features and classifiers in the setting of granular computing. The proposed framework involves creation of diversity among classifiers, through adopting feature selection and fusion to create diverse feature sets and to train diverse classifiers using different learning algorithms. The experimental results show that the proposed multilevel framework can effectively create diversity among classifiers leading to considerable advances in the classification performance. Keywords Machine learning · Ensemble learning · Image classification · Shape recognition · Feature extraction · Granular computing
1 Introduction Shape recognition is a critical part of pattern recognition due to its wide applications in image retrieval, object detection surveillance systems and any related areas. In the recent years, machine learning gains its popularity due to * Weili Ding [email protected] Xinming Wang [email protected] Han Liu [email protected] Xiangsheng Huang [email protected] 1
Department of Automation, Institute of Electrical Engineering, Yanshan University, 438 West of Hebei Avenue, Haigang District, Qinhuangdao 066004, China
2
School of Computer Science and Informatics, Cardiff University, Queen’s Buildings, 5 The Parade, Cardiff CF24 3AA, UK
3
Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
its modeling ability. In shape recognition tasks, the same kind of shapes is assigned a specific label, which consists of data samples, and effective feature descriptors extracted from these data combined with powerful machine learning algorithms usually lead to good recognition results. Feature extraction and classification are two significant steps in shape recognition, which can directly affect the recognition results. In the past few years, effective shape features and classification methods have been studied. In general, a high dimensional feature set usually contains redundant information which has negative effects on the recognition result. Specifically, the redundant information exists in local features and global features. In addition, classification performance can be varied d
Data Loading...
