The research in the various research and development teams of the project (see below) has led to various developments that are incrementally tested and embedded in the testbed system. We describe a few of the contributions that have been embedded in the testbed over the past year.
Image Processing Team: Browsing and Progressive Delivery
The Testbed currently supports a demonstration of progressive delivery of wavelet-encoded imagery, developed by the Image Processing Team. This method stores the wavelet coefficients in separate files, one for each level of decomposition, and sends the next level to a Web browser upon request. The wavelet data arrives at the browser with a special MIME type that triggers the execution of a client-side "helper application" to perform the inverse wavelet transform and image reassembly. Currently, this client is only supported on Solaris and Digital UNIX, and requires the shareware program "xv" to display the reconstructed images. The Image Processing team is rewriting the helper application in Java, so it will be portable to any client environment.
Image Processing Team: Texture-Based Retrieval
The Testbed currently supports a demonstration of texture-based image retrieval, developed by the Image Processing Team. Images to be subject to texture-based retrieval are preprocessed into fixed-size tiles, for each of which a vector of texture metrics is computed. These vectors are clustered and quantized into a codebook of most-typical textures, and each tile is mapped to its most-similar texture in the codebook. Texture-based search thus becomes a simple one- or two-step process: the codebook may be browsed, and any tiles matching a specific texture retrieved, or the codebook may be entered indirectly by selecting an arbitrary image tile and thus its corresponding codebook entry. In the context of the Testbed, this translates to either one or two database table accesses, so it is quite fast.
Performance Parallel Processing Team
The Testbed currently supports a demonstration of browsing and retrieving arbitrary subsets of wavelet-encoded imagery, developed by the Parallel Processing Team. This method supports the specification of an arbitrary rectangular region in a Java-based client at each level in the wavelet reconstruction hierarchy; only the specified region is delivered at the next level as a GIF image viewable in any Web browser. The inverse wavelet transform and image reassembly is currently performed on the server, but will switch to using the Java-based client-side wavelet code as soon as it becomes available.