Storage and retrieval of multimedia objects has become a requirement for many contemporary systems．For example，given an image database，one may want to retrieve all images that are similar to a query image．Asmany c...Storage and retrieval of multimedia objects has become a requirement for many contemporary systems．For example，given an image database，one may want to retrieve all images that are similar to a query image．Asmany content-based retrieval techniques for digital imagery use a feature vector approach to represent image contents，it is desirable to reduce the dimensionality of the data，whilst maintaining as much of its original structure．Several dimensionality reduction techniques are available．The most popular one is PCA，which works well for static databases．In this paper，we present a novel scheme for performing PCA-based dimensionality reduction in dynamic databases．Instead of using the entire dataset，we only recompute the PCA transform matrix on the updating dataset．Note that the size of the updating dataset is usually much smaller than that of the entire data，this technique may reduce the PCA computation time complexity without losing exactness．In addition，the updates to the database are based on the existing dimensionality-reduced data vectors rather than the original high-dimensional data vectors，which may relieve the system overhead for the management of the original high-dimensional data．展开更多
文摘Storage and retrieval of multimedia objects has become a requirement for many contemporary systems．For example，given an image database，one may want to retrieve all images that are similar to a query image．Asmany content-based retrieval techniques for digital imagery use a feature vector approach to represent image contents，it is desirable to reduce the dimensionality of the data，whilst maintaining as much of its original structure．Several dimensionality reduction techniques are available．The most popular one is PCA，which works well for static databases．In this paper，we present a novel scheme for performing PCA-based dimensionality reduction in dynamic databases．Instead of using the entire dataset，we only recompute the PCA transform matrix on the updating dataset．Note that the size of the updating dataset is usually much smaller than that of the entire data，this technique may reduce the PCA computation time complexity without losing exactness．In addition，the updates to the database are based on the existing dimensionality-reduced data vectors rather than the original high-dimensional data vectors，which may relieve the system overhead for the management of the original high-dimensional data．
