Even with access to very high speed networks, it is often impractical to transmit a large image as a single item, particularly if the user is in a browsing mode, trying to find items of interest. A simple solution to this problem is to maintain for each large image, a low-resolution (e.g., subsampled) ``thumbnail'' image for browsing purposes. While thumbnails consume storage space (or processing cycles, if generated on-demand), this overhead is typically insignificant compared to the advantages from their use. If the user finds a thumbnail of interest, additional high-resolution data may be downloaded.
It is clear from our experience with the RP and various other GIS systems that we should make thumbnails available as well as the original data. However, this only addresses the issue of browsing quickly through large numbers of images. In an interactive database system, users are likely to do much more than make binary ``go/no-go'' decisions based on simple thumbnail images. They may, for example, wish to zoom in on a given region. Such operations typically cannot be supported with low-resolution thumbnail images. Furthermore, different groups of users may have different requirements. A school teacher using a LANDSAT image for a certain demonstration may not need the same high-resolution image as a scientist trying to classify land cover types. The general solution is to have access to hierarchical, multiscale representations of image data.
An obvious solution to this requirement is the use of wavelet transforms, which provide multiscale decompositions of the image data [11]. Wavelets have been widely used in many image-processing applications, including compression, enhancement, reconstruction, and image analysis. Fast algorithms exist for computing the forward and inverse wavelet transforms, and desired intermediate levels can be easily reconstructed. The transformed images (wavelet coefficients) also map naturally into hierarchical storage structures. We can assume that low-resolution data are accessed more frequently than the finer information, and hence should be stored in faster devices for efficient browsing, while higher-resolution data may be placed in tertiary storage. A useful property of the decomposition is that the lowest-resolution, low-pass filtered and subsampled image (a by-product of the wavelet transform) may be used as a thumbnail for browsing.
Important issues related to wavelet-based storage include the choice of decompositions (i.e. choice of filters) that are appropriate for the different image databases. Image compression is important in storing large amounts of data. Many GIS and medical imaging applications often require lossless compression and this continues to be an active research problem in image-processing. Although the total number of wavelet coefficients equals the number of pixels in the images, their storage requirements differ. The original intensity data, in most cases, consists only of integer numbers. Wavelet coefficients are real numbers, thus requiring more memory. Even for the case of no compression, these coefficients need to be quantized and encoded appropriately to ensure that they do not take more space than the original image data. How to quantize these coefficients without losing near perfect reconstruction is an important research issue.