Terrain Compression Experiments

The sample is the level-2 DEM of Bountiful Utah. Missing data values were replaced by the mean of the others. The standard deviation of the elevations is 1096 meters.

The compression method is Said and Pearlman's wavelet program, progcode. Lossless compression takes 5.234 bits per point (bpp) on this file.

The first display technique is a simple gray scale

Original


Lossy Compression to 1 Bit/Point, RMSE=6.0 meters


Lossy Compression to 0.1 Bit/Point, RMSE=51.7 m


Lossy Compression to 0.03 Bit/Point, RMSE=148 m


Lossy Compression to 0.01 Bit/Point, RMSE=535


Now, here are the same surfaces, plotted in 3D by Matlab.

Original


Lossy Compression to 1 Bit/Point, RMSE=6.0 meters


Lossy Compression to 0.1 Bit/Point, RMSE=51.7 m


Lossy Compression to 0.03 Bit/Point, RMSE=148 m


Lossy Compression to 0.01 Bit/Point, RMSE=535


Now, here are the differences between each level of approximation and the one below it. Note that the scales are different on each plot; otherwise the first plots would appear flat.






Wm. Randolph Franklin, Associate Professor
Email: wrfATecse.rpi.edu
http://wrfranklin.org/
☎ +1 (518) 276-6077; Fax: -6261
ECSE Dept., 6026 JEC, Rensselaer Polytechnic Inst, Troy NY, 12180 USA
(GPG and PGP keys available)