9.1B
Multiresolution Tiling for Interactive Viewing of Large Datasets
K. Palaniappan, University of Missouri, Columbia, MO; and J. B. Fraser
The size of remotely sensed imagery continues to grow at rapid rates with the advent of new high resolution sensors such as one meter imagery from Ikonos or 36 spectral bands from Terra/MODIS. Image mosaics covering large geographical regions are also becoming more prevalent. For example, a Landsat TM mosaic of the conterminous United States (US) at 30 m resolution results in an image with dimensions of 218,000 x 95,000 pixels and uses 160 gigabytes (GB) of storage space at multiple resolutions to facilitate interactive visualization. New techniques are needed to efficiently store and quickly retrieve such large datasets for interactive viewing on computers with physical memory many times smaller than the size of the data to be viewed. Additionally, these datasets often reside on a remote machine further complicating the ability to view and manipulate this data. We present a system for interactively viewing such datasets that not only exceed available memory resources, but potentially exist only in secondary storage of a remote system.
We define optimization criteria to provide for an efficient and truly interactive system. In particular, our goal is to minimize network and disk I/O, while maximizing both on-disk data compression and viewer refresh rate. In order to meet these criteria, several schemes for maintaining the data at multiple resolutions and organizing the data into tiles are presented for providing scaled views and efficient retrieval of view-dependent subsets of the data. These tiles are used within a paging system to make efficient use of memory by caching and swapping sections of data from secondary storage to primary memory based on view in a way similar to that of a virtual memory operating system. Within this system, client-server approaches using existing internet protocols are used to provide similar caching of data to provide interactive viewing of remote datasets. Additionally, approaches for compressing this structured data using wavelet-based compression are given.
This system is designed to meet not only the needs of two-dimensional image data, but adapted for viewing multispectral remote sensing data products and to provide volumetric compression of multidimensional data. Finally, comparisons are made with existing techniques for compressing and viewing very large datasets such as LizardTech's MrSid, and the JPEG 2000 standard.
Session 9, Visualization (Parallel with Session 7 & 8)
Wednesday, 17 January 2001, 1:30 PM-4:15 PM
Next paper