In this thesis, a method for retrieving an image sequence in a database based on its contents is proposed. The method evaluates the similarity between image sequences based on the similarity of trajectories in eigenspaces. Two criteria to evaluate the similarity are employed. One concerns the distance between an input image sequence and the eigenspaces, each of which is previously organized from an image sequence. The other concerns the similarity of trajectories in the eigenspaces. Experimental results show the feasibility of the proposed retrieval method with real image sequences which represent hand gestures.
E-mail: yasuhu-h@is.aist-nara.ac.jp