Structuring home video by snippet detection and pattern parsing

Hand-held camcorders have been popularly used in capturing and documenting daily lives. Nonetheless, searching for personal memories in home videos is still a laborious task. This paper describes novel approaches in detecting snippets and patterns in home videos for content indexing. To deal with the fact that most shots are long and with handshake artifacts, a motion analysis algorithm based on Kalman filter and finite state machine is proposed to decompose videos into tables of snippets. Each snippet is represented by a set of moving and static patterns. The moving patterns are automatically detected and tracked, while the static patterns are manually input by users. A MWBG pattern matching algorithm is then proposed to effectively detect and parse the patterns in snippets. Home videos are ultimately albumed and indexed according to the moving and static patterns to facilitate content search. Copyright 2004 ACM.