correlation 
coefficient 
c 'c=ZZl/('j)-/v fcoj)-/',]/ 
y-0 /-0 / 1 
M-l N-l _ . M-lN-l. , 
YLVV'fi-PrìYLWfi-rì 
7=0 i=0 7=0 i=0 
PSNR 
PSNR = -101og 10 — MSE = f-t i f j \f(i,j)-g(i,j)Y 
¿55 MN j —Q l= o 
Table 2 Definition of image similarity, image fidelity, 
correlation coefficient, and PSNR 
3.1.3 Image application analysis 
Image application analysis refers to study the effects of lossy 
compression on image quality from the aspect of image 
application. The common image applications are image 
classification, image target recognition, image character 
extraction and spatial pattern analysis. 
3.2 Geometric quality assessment 
The geometric quality of remote sensing image reflects the 
reconstructed image in keeping geometric characters, such as 
feature’s shape, size, etc. The goodness of geometric quality 
determines the degree of image measurability. In many high 
precision image processing application, such as digital 
photogrammetry, computer graphics, it is very important to 
study the effects of compression on geometric quality. 
In surveying and mapping area, several methods, like image 
matching accuracy analysis, automated DSM extraction 
accuracy analysis, photogrammetric point determination 
accuracy analysis, etc, can be used to assess the effects of 
compression on geometric quality. 
3.2.1 Image matching accuracy analysis 
Image matching accuracy analysis shows how degradation in 
image quality associated with lossy compression can affect 
matching accuracy. 
Let Ax and Ay refer to image matching error, then: 
Ax = x / -x g > Ay = y f -y g (7) 
(xf,yf) is the pixel value of original image, and (x g , ) is 
the corresponding pixel value of reconstructed image. The 
matching algorithm - least square matching is a good choice 
since it can reach the accuracy of 1/10-1/100 pixel. 
3.2.2 DSM extraction accuracy analysis 
As to stereopairs, Automated DSM extraction accuracy analysis 
and photogrammetric point determination accuracy analysis are 
necessary [Tian-Yuan Shih et al, 2005; Zhilin Li et al, 2002]. 
A DSM is extracted from the original stereopair. This DSM is 
held as the reference against which the terrain heights obtained 
from compressed imagery are compared. The DSMs can also be 
derived for each of the compressed pairs, with the same exterior 
orientation parameters. DSM extraction accuracy analysis is 
conducted through comparing two DSM data-sets with X, Y, Z 
format that are generated from original images and compressed 
images, respectively. The relative error in elevation can be 
measured at the identical points by comparing the 
uncompressed and compressed images with root mean square 
error (rmse) [Tian-Yuan Shih et al, 2005]. 
3.2.3 photogrammetric point determination accuracy 
analysis 
Similar to DSM extraction accuracy analysis, photogrammetric 
point determination accuracy analysis is to compare the 
accuracy of two sets of 3D coordinates of the feature points that 
are from original images and reconstructed images [Zhilin Li et 
al, 2002]. 
4. A PROTOTYPE FOR IMAGE COMPRESSION 
QUALITY ASSESSMENT SOFTWARE 
Based on the above research, we have designed and 
implemented a prototype for image compression quality 
assessment software. The major functions include image 
character analysis, image comparison analysis, and image 
matching accuracy analysis, etc. Figure 1 shows the interface. 
(a) 
Figure 1 Image Compression Quality Assessment Software 
Interface 
5. CONCLUSION 
Quality assessment for remote sensing image compression is 
critical to compression algorithm designers and image products 
users. Through the above research, a series of experiments were 
conducted. As to the compression ratio adopted in Resources 
Satellite-3 (a high resolution stereo mapping satellite), 
compression ratio should be no more than 4:1 if JPEG2000 or 
SPIHT algorithm were employed on board. 
REFERENCES 
Ahmet M. Eskicioglu, 2000. Quality Measurement for 
Monochrome Compressed Images in the Past 25 Years. IEEE 
International Conference on Acoustics, Speech, and Signal 
Processing, Istanbul, Turkey, pp. 1907-1910 
Ferwerda, J.A., Pellacini, F., 2003. Functional difference 
predictors (FDPs): measuring meaningful image differences. 
Conference Record of the Thirty-Seventh Asilomar Conference