Thesis (Ph.D.)--Chulalongkorn University, 2010

Image classification with semantic contents is a challenging problem inrecent years. Various techniques such as content based image classification, textbased image classification, and combined content and text based imageclassification have been developed to solve this problem. To boost a high ofsemantic image classification performance, several techniques such as backgroundelimination, region of interest detection, and object recognition have beenincorporated. In an attempt to explore the applicability of such combined techniques,this research undertakes the semantic image classification problem by focusing onsport image domain. The proposed technique use player's posture as theidentification basis to classify the input image is the posture of player classification.For example, a baseball player image should show the 'hitting' and 'throwing'postures. To overcome this problem, the proposed method utilizes automaticcomplex background elimination and posture classification based on six differentsports to gauge the applicability of the approach. The proposed method is robust toscaling and rotation player image.