(a) (b) 
  
(d) (e) (0 
Figure 6. Textures of SAR imagery with different 
orientations(a to f stand the texture images of g = 7, 4 to g2z 
with every 7 
ry 7 ) 
This paper will establish the transformation of horizontal 
texture (0 = ) since the two types of images have more 
explicit horizontal and vertical texture. 
Since the amount of calculation of the algorithm proposed by 
Kain is very great, this paper will take the following two steps 
to reduce the amount of calculation. On the one hand, it will use 
a local horizontal texture (9 - 7) of figure4 and figure6 with 
50 X 50 instead of the whole texture image. On the other hand, 
it will use the K-means algorithm before establishing the texture 
mapping. 
The registered local texture of QuickBird imagery and SAR 
imagery used in this experiment are shown in figure7 and 
figure8 
    
    
» ud à " 
Figure 7. QuickBird imagery Figure 8. SAR imagery 
figure9 shows the outputs of transformation with different m 
which is the number of single Gaussians in one GMM. 
Spatially, the output of transformation has a high similarity and 
expression with the SAR imagery in figure8, such as the 
building at the top of left corner both in figure7 and figure8. Its 
shape and scale have transformed to be more similar to the SAR 
imagery. Consequently, it demonstrates that the GMM can play 
a good performance in the transformational of texture features 
description under different imaging conditions. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
   
a 
  
(c) (d) 
Figure 9. The transformation results with different m 
(a, result with m4; b, result with m-8; c, result with m=16; 
d,result with m=64) 
Table! shows the results of the average of transformation error 
precision with different m in the experiment according to 
Eq.(19). And it demonstrated that the transformation error 
precision will reduce by the number of single Gaussian 
increasing. 
  
m-4 m-8 m=16 m=64 
Average error | 0.2602 0.2533 0.2493 0.2482 
precision 
  
  
  
  
  
  
  
Table 1. Average error precision 
Figurel0 shows the comparison of QuickBird imagery, SAR 
imagery and the transformed result with m=64. The X-axis 1s 
the point number from 1 to 50 and y-axis is the value of each 
pixel. There are three types of lines in Figurel0. The solid line 
with point represents the value of SAR imagery texture. The 
solid line with star stands the value of transformed texture and 
dotted line stands the value of Quickbird imagery texture. 
The Sua eh Vey bed a GI 
  
  
EGG ud Eo s RTE RBANLG WEEE ERR 
  
FRE RTE RRR NY 
Figure 10. The comparison of results