Affiliations: [a]
Information Technology Faculty, Electric Power University, HaNoi, Viet Nam
| [b]
Information Technology Faculty, Posts and Telecommunications Institute of Technology, HaNoi, Viet Nam
| [c]
Institute of Information Technology, Vietnam Academy of Science and Technology, Ha Noi, Viet Nam
Correspondence:
[*]
Corresponding author. Quynh Nguyen Huu, Information Technology Faculty, Electric Power University, HaNoi, Viet Nam. E-mail: [email protected].
Abstract: Over the years, many content-based image retrieval (CBIR) methods, which use SVM-based relevant feedback, are proposed to improve the performance of image retrieval systems. However, the performance of these methods is low due to the following limitations: (1) ignore the unlabeled samples; (2) only exploit the global Euclidean structure and (3) not taking advantage of the various useful aspects of the object. In order to solve the first problem, we propose a graph-based semisupervised learning (GSEL), which can add positive samples and construct balanced sets. With the second problem, we propose a manifold learning for dimensional reduction (MAL), which exploits the geometric properties of the manifold data. With the third problem, we propose a combination of classifiers by aspect (CCA), which exploits the various useful aspects of the object. Experimental results reported in the Corel Photo Gallery (with 31,695 images), which demonstrate the accuracy of our proposed method in improving the performance of the content-based image retrieval system.
Keywords: Content-based image retrieval (CBIR), relevance feedback, support vector machines (SVM), Graph-based Semisupervised learning and manifold learning
