2004 年 58 巻 12 号 p. 1835-1838
A method retrieving videos is presented by utilizing vector quantization and latent semantic analysis. Each video is represented by a sequence of signatures through the vector quantization of frame datasets. Latent semantic analysis is then applied to the signature with a video matrix. We verified through experiments that dimensionality reduction in latent semantic analysis increases the speed and precision of retrieval. Making vector quantization more robust further improved the performance of similarity searches.