Compressed Spatio-temporal Descriptors for Video Matching and Retrieval
Orkun Alatas, Omar Javed, Mubarak Shah · 2004
The contents of a video can be described in terms of appearance and motion of the scenes. In this paper, we propose wavelet-based appearance descriptors and splinebased motion descriptors that can together represent the videos, and be used for matching and retrieval. The optical flows give the direction along which the gray levels have regular variations in time, and the wavelet decomposition can take advantage of this by decomposing the sequence along these directions, exploiting the redundancy. Such a compact representation of video is very crucial for storage and retrieval purposes. Therefore we show that these spatio-temporal descriptors can be used for 1) video retrieval and matching, and 2) calculating spatiotemporal similarity between articulated objects. The results are demonstrated on various kinds of sequences.