The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
663 
insignificant) 
IF THE ENTRY IN LIS REPRESENTS D(i,j) (every thing 
below node on tree) 
- decide if there will be any more significant pixels further 
down the tree and output the decision result 
- if it is significant, decide if all of its four children (0(i j)) 
are significant and output decision results 
•if significant, add it to LSP, and output sign 
•if insignificant, add it to LIP 
IF THE ENTRY IN LIS REPRESENTS L(ij) (not 
children but all others) 
- decide if there will be any more significant pixels in L(ij) 
further down the tree and output the decision result 
- if there will be one, add each child to LIS of type D(i,j) 
and remove it from LIS 
3. Refinement Pass: (all values in LSP are now 2n < | ci,j |) 
For all pixels in LSP, output the nth most significant bit 
4. Quantization-step Update: decrement n by 1 and do another 
pass at step 2. 
In the procedure above, each judgement generates an output 
sign, and put it in the output bit stream. We can directly adjust 
the length of the output bit stream to control the compression 
ratio of terrain data. We use the formula below: 
ITotalBits = nXDim x nYDim x BitRate 
ITotalBits: The length of current output stream (unit: bit) 
nXDim: the column number of the image pixels 
nYDim: the row number of the image pixels 
Bitrate: controls the compression ratio. 
In this paper, bit rate is proportional to terrain surface 
complexity (R), and is inversely proportional to terrain scale. 
So we add the step of terrain surface complexity calculation 
into the code procedure. The bit rate parameter direct ratios the 
terrain surface complexity (R). And the improved SPIHT 
coding method for DEM is as the figure below. 
Figure 6. Improved SPIHT coding method for terrain 
compression 
In the improved coding method, three levels of HAAR wavelet 
transform is adopted, then high frequency coefficients 
calculation and terrain complexity evaluation are proceeded. At 
last SPIHT code is carried through using the coefficients of the 
transformed data and bit rate generated in the complexity 
evaluation. The experiments of compression are in section 5. 
4. EXPERIMENTAL RESULTS 
In this section some experiments are carried out for the 
improved SPIHT compression method. 
Three types of terrain data are used for experiment, which are in 
high undulation, middle undulation and low undulation areas. 
And their compression results are shown in Figure 11. The x 
axis represents the bit rate adopted in the SPIHT encode. The 
higher the bit rate is, the lower the compression ratio is. And 
the y axis represents PNSR index after compressed. The higher 
PNSR is, the compressed data has finer fidelity .We can find 
that the low undulation area can also receive high PNSR using 
small bit rate, but the high undulation area need a big bit rate to 
receive satisfied compression quality. 
Figure 7. The relationship between PNSR and bit rate for three 
kinds of terrain area 
The compression results using the improved SPIHT coding 
method are shown in the figures below. In the flat area the bit 
rate calculated by the terrain surface complexity and terrain 
scale is 0.3, and we can find that when the bit rate adopts 0.3, 
the flat area can get an acceptable the compression result, (as 
shown in Figure 12). 
In the mountainous area the bit rate calculated by the terrain 
surface complexity is 1.1, and the compression result is shown 
in Figure 13. But if we use the bit rate 0.3 to compress a 
mountainous area. The compressed terrain will degenerate 
severely. 
We use the techniques adopted in this paper to realize global 
visualization The interface pictures of the global terrain 
visualization system are shown in Figure 14. We can find that 
these techniques can perform well.