Model-based similarity estimation of multidimensional temporal sequences

Content-based queries in multimedia sequence databases where information is sequential is a tough issue, especially when dealing with large scale applications. One of the key points is similarity estimation between a query sequence and elements of the database. In this paper, we investigate two ways to compare multimedia sequences, one – that comes from the literature – being computed in the feature space while the other one is computed in a model space, leading to a representation less sensitive to noise. We compare these approaches by testing them on a real audio dataset, which points out the utility of working in the model space.