In Table 4, as the smoothing times increase, the entropies and 
information amounts are correspondingly reduced, which 
reveals information loss during the edge protection smoothing 
process. Meanwhile, the gradually blurred images (in Figure 6) 
after increased times of smoothing also suggest the information 
loss to some extent. Therefore, the edge preserved smoothing 
method is proposed to be integrated with the district forecast 
differential encoding so as to further improve compression rates. 
But the spatial resolution of images should be considered. For 
example, urban areas are characterized by point and linear 
details inherent to images. If the spatial resolutions of images 
are relatively higher, the times of edge protection smoothing are 
supposed not to be too much in order to avoiding large 
information loss. Otherwise, the land cover types with large 
redundancy, exemplified by water, would be compressed with a 
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 
    
significantly improved compression rate because of the edge 
protection smoothing. 
Besides, a sectional drawing is plotted along with the A-A 
direction, as is shown in Figure 7. The black curve represents 
the gray values of the A-A section pertaining to the original 
image, while the red, blue and green curves respectively 
represent the gray values of the A-A section after once, two 
times and three times edge protection smoothings. It is clear 
that the edges are relatively enhanced by the smoothing. 
Moreover, the positions of edges are relatively permanent. 
Hence, this kind of smoothing is immune to edge destruction, 
which will maintain the accuracies of stereo matching. 
  
  
  
  
  
  
  
  
Urban areas | Farmland | Mountain areas | Water 
Entropy (bits) 6.12 4.89 4.13 3.27 
Information amount (bits) 4.31 2.94 2.24 1.65 
Average code length 6.16 491 4.17 3.31 
Huifinan Coding Compression rate 1.30 1.63 1.92 2.42 
Uni-directional Average code length 4.97 3.62 3.18 2.41 
differential coding Compression rate 1.61 2.21 2.52 3.32 
Bidirectional Average code length 3.93 331 2.78 2.32 
differential coding Compression rate 2.04 2.42 2.88 3.45 
  
Table 3. Huffman coding and differential coding of images of four land cover types 
  
(b) farmland 
    
  
(c) mountain area 
Figure 5. Remote sensing images of for typical land cover types 
  
  
(b) 
(c) 
  
(d) water 
  
Figure 6. Resulting images after different times of edge protection smoothing: (a) one time; (b) twice; (c) three times