Sport Video Classification using an Ensemble Classifier

Sport video classification is an application of videoanalysis which can be useful in video indexing and retrieval. Inthis article, a new method for sport video classification usingensemble classifier is proposed. The proposed method uses 6features: 3 dominant colors, dominant gray level, cut rate andmotion rate. These features are classified by 4 simple classifiersin an ensemble classifier: Nearest Neighbor (NN), LinearDiscriminant Analysis (LDA), Decision Tree (DT) andProbabilistic Neural Network (PNN). To combine the output ofsimple classifiers and make final decision, weighted majorityvote is used while the weight of each simple classifier is equalto corresponding correct classification rate (CCR).Experimental result shows that the CCR of proposed system is78.8%. In this experiment, 104 clips in 7 different sport classesare used: football, basketball, tennis, swimming, futsal, ski andbox.