Back to Search Start Over

Probability fuzzy SVM for image retrieval

Authors :
Shenglan Liu
Yonghua Tang
Deshan Liu
De-qin Yan
Source :
FSKD
Publication Year :
2015
Publisher :
IEEE, 2015.

Abstract

In the scheme of relevance feedback (RF) of conventional content-based image retrieval (CBIR) systems, only the labeled samples will be trained for learning, which gives rise to the small sample size (sss) problem. To solve the problem, a variety of RF schemes have been developed with exploiting support vector machine (SVM) which concentrates on enhancing classification ability of the schemes. In fact, there is not only classification distribution but also probability distribution in both numeric value data and image data. In fact, probability distribution existing in the images with multi-semantic meaning is more crucial for building a better scheme of RF. For modeling the scheme of RF under the consideration of both classification distribution and probability distribution, a framework of probability fuzzy support vector machine (PFSVM) is developed in this paper. In the framework, every pseudo-label sample is assigned both a fuzzy and a probability membership which are integrated into mechanism of PFSVM for active learning. Experimental results with a database of 1,000 images demonstrate the effectiveness of the proposed method.

Details

Database :
OpenAIRE
Journal :
2015 12th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD)
Accession number :
edsair.doi...........149de647b50d402a3229580e0a1be50b
Full Text :
https://doi.org/10.1109/fskd.2015.7381929