Abstract
This paper proposes a novel video image clustering method based on human face and shirt color. Clustering has been performed in unsupervised manner where no prior knowledge about the input video clip is required. For face detection, multi-resolution templates based pattern matching method has been used. We have experimented on 11 video clips of various durations in order to evaluate the performance of the proposed system. The ability of the system to generate a meaningful number of clusters that exist in the input video clip is definitely well. The performance of the system with respect to precision and recall rate is also highly satisfactory.