Gjorgji Madzarov, Ivica Dimitrovski, Suzana Loskovska, Dejan Gjorgjevikj

Abstract: In this study we propose a novel architecture of Support Vector Machines (SVM) classifiers for classifying video documents in classes organized in a predefined class hierarchy. This architecture of SVM classifiers was designed to provide superior recognition speed and accuracy utilizing a binary classification tree. It solves a four-class video classification problem, having a two level hierarchy. The classification data consisted of automatically detected key frames, which were described by the MPEG-7 standard. The experimental results show that the classification architecture achieves high classification accuracy, especially in the first level of the hierarchy where the classification was made between different genres of videos.