Terrain chunk size is 256 by 256 elevation samples.
Two curvature preserving methods of mesh simplification has been tested so far:
Results in tabular form
Plots (mesh simplification threshold on x axis):
Screen dumps for different threshold values:
Terrain chunk size is 256 by 256 elevation samples.
Number of triangles for uncompressed and compressed DEM (Digital Elevation Model) vs. mesh simplification threshold: results in tabular form, Postscript file or gif image
Visual comparison of compressed (32 times) and uncompressed DEM for various thresholds, left terrain model with DEM compression, right one without:
Note: wavelet compressed DEMs are more sensitive to mesh simplification for higher thresholds (what can be seen for t=0.3).
The compression ratio varies depending on particular image features. The more ``detailed'' image, the harder to compress. The number of wavelet compression algorithm passes in used as a measure of image quality (values used in the GIS: 6-7 passes).
For satellite imagery data 7 passes correspond to compression ratios between 10 and 36 (see table). 6 passes correspond to compression ratios between 28 and 75 (see table).
Decompression time for 9 sample satellite images, 512 by 512 pixels: table or gif image
Note: Significant profits can be expected for slow network connection and fast CPU. For fast connections, time necessary to uncompress the image can be longer than time spent on downloading the uncompressed one, eg. during the tests for SGI Indigo w/R10000 and network connection at 260kB/s both times were roughly the same.
Screen dumps for image quality 6, left terrain model with texture compression, right one without (original):
Screen dumps for image quality 7, left terrain model with texture compression, right one without (original):
Note: satellite images of quality 7 are practically indistinguishable from originals (see example 3 above).