International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
quantized with much coarse level to obtain higher compression. 
Figure 2 gives the quantification policy of the wavelet 
coefficient based on the edge reservation: 
big redundancy among the multispectrum images. Only one 
edge image is kept to obtain high compression ration during the 
multispectrum image compression course. 
  
  
  
    
  
  
  
  
     
  
    
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Figure 4 give the multipsectrum images compression flowchart 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
    
  
  
  
  
  
  
  
   
  
Wevelet TH Bdge Huffman Code| | based on 3D wavelet transformation. 
Transform Extraction : 
foy) fu(x.y) r(x) m Image Band 1 | Image Band 2 | so | Image Band N 
A | ] 
Js(x.y) 
Fan X M —_— Correlation Calculation 
fue y) mb Huffman Code| | Quantization A 
to determine reference 
image 
Figure 2 Wavelet Coefficient Quantization Flow $5 
Where f(x,y) is the original image. /;;(x,y) is the low frequency | Reference [mage | | Image Band i | 
coefficient image. fy(x,y) is the high frequency coefficient. T 
fs) i 
is the edge image. /(x,y)is the left wavelet coefficient without 3D Wavelet 3D Wavelet 
edge. Transformation Transformation 
e - . 
Fu Gy). f ny) fu Gy) f uy) 
In this paper, the sobel operator(as equation shown) is adopt to rT} 
extract edge from the wavelet coefficient. Y Y Y 
Band Reference || Band Reference Band i Low 
Sai lati wd Low Frequency || High Frequency > Frequency 
Coefficient Coefficient Coefficient 
-2 0 —2 0 0 0 l v Remove edge from f(x,y 
-1 0 -1I -1 -2 -1 [Huffman Code || Arithmetic Code | Huffman Code 
  
Figure 3 show the edge detection result from the wavelet 
Figure3 Edge Extraction Result With Sobel Operator 
Minus the edge coefficient from the wavelet coefficient, we can 
obtain the wavelet coefficient image without edge. 
2.4 Technique flowchart of Multi-band wavelet based 
multispectrum image compression 
After the 3D wavelet transformation, the multispectrum images 
can obtain high frequency wavelet coefficient image. As the 
above section analyse, the edge and contour can be extracted 
from the wavelet coefficient. For each spectrum image, the edge 
and contour is almost the same for they present the same region. 
Thus, the edge image extracted from the wavelet coefficient is a 
      
  
+ Band i Keep High frequency 
  
B [Reference Band 
  
  
coefficientwithout Edge 
  
  
Compressed 
Data 2 of 
  
Compressed 
Data 2 of 
  
  
  
  
Compressed [Arithmetic Code | 
Data 1 of T 
Reference Band 
Band ; 
  
  
  
  
  
  
  
Compressed 
Data 2 of Band i 
| 
Y 
Figure 4 Compression Technqiue Flow of Multispectrum Images 
  
  
  
  
  
As figure 4 shown, The multispectrum image compression 
technqiue is summarized as the following steps: 
(1) Determining the reference image 
How to determine the reference image is very important to the 
multispectrum image compression with high compression ratio 
and high reconstruction image quality. In this paper, the 
reference image is determined through the correlation analyse. 
Each spectrum image is selected to calculate the correlation 
coefficient to any other spectrum image. The spectrum image 
with the maximize average correlation coefficient is determined 
as the reference image. The reference image determination 
course can be explained as: 
For i 1 to M 
/ 
For j *1 toM 
f 
i 
If i!=j Cor(i,j) = CalCor(Image(i), Image(j)) 
/ 
AverCor(lj) = Avreage(Cor (i,j) 
If (Max(AverCor(l,j))) 
58 
Internati 
1 
J 
(2) Con 
At first 
image. 
wavelet 
compre 
(3) Con 
Decom 
transfor 
low fre 
wavelet 
coding. 
wavelet! 
quantifi 
taken t 
arithme 
coeffici 
(4) Rep 
The cor 
illustrat 
Compres: 
Data 2 c 
Reference I 
  
D... 
Huffme 
Decod 
v 
Band Refe 
Low Frequ 
Coeffici 
| 3D Inverse W 
Transforma 
wie 
Referen 
Reconstru 
Image 
  
  
  
  
Fidility 
quality