A Multilabel Approach Using Binary Relevance and One-versus-Rest Least Squares Twin Support Vector Machine for Scene Classification

Abstract

The classification of an image scene having multiple class labels produces significant challenge to the researchers. A semantic scene may be described by multiple objects or by multiple classes. For example, a beach scene may also contain mountain or buildings in the background. This research work proposes a multi-label scene classification model by using Binary Relevance (BR) based one-versus-rest Least Squares Twin Support Vector Machine (LSTSVM). Fifteen evaluation metrics have been used to analyze and compare the result of the proposed scene classification model with the six existing multi-label classifiers. Experimental results demonstrate the superiority and usefulness of the proposed model in the classification of multi-label scene over the existing multi-label approaches.