Fast and Accurate Home Photo Categorization for Handheld Devices using MPEG-7 Descriptors

Home photo categorization has become an issue for practical use of photos taken with various devices. But it is a difficult task because of the semantic gap between physical images and human perception. Moreover, the object-based learning for overcoming this gap is hard to apply to handheld devices due to its computational overhead. We present an efficient image feature extraction method based on MPEG-7 descriptors and a learning structure constructed with multiple layers of Support Vector Machines for fast and accurate categorization of home photos. Experiments on diverse home photos demonstrate outstanding performance of our approach in terms of the categorization accuracy and the computational overhead.