Automatic content description and annotation of sport images using classification techniques

Abstract

Multimedia enhances the applications with rich content. The uses of media tools moreover make communication more effective. Consequently, automatic image annotation and retrieval has recently occupied a large space in the multimedia research area. This paper presents a framework for automatic sport image annotation and content description for search and retrieval purpose. The images are segmented into semantic concepts based on their color and texture. Next, low-level features of color, texture and position are extracted from each segment, followed by manual annotation for supervised training. Finally, classification is applied using two different classifier techniques: Support Vector Machine (SVM) and Decision Tree. Experiments have been conducted on the "Leeds Sports Pose" dataset to test the performance of the given techniques. The comparison between the classifiers has shown that the SVM outperforms the Decision Tree, with an average measured accuracy of 73% for SVM versus 45% for the Decision Tree.