The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1272 
Fig.2. Distribution histograms of corrected data in B2 
The dispersion of the collected calibration data done with 
relative radiometric correction is called the relative calibration 
accuracy. Equation 1 is used to calculate it. 
£(DN-DN) 2 
y— 
RA = X xl00% (1) 
DN 
The calibration results of CBERS 02B CCD data are listed 
in Table 1. 
Radiance 
The average DN values 
The relative calibration 
level 
after corrected 
accuracy (%) 
1 
15.0582 
2.78 
2 
27.0772 
1.50 
3 
39.8562 
1.13 
4 
49.6421 
0.92 
5 
64.6399 
0.74 
6 
76.6266 
0.66 
7 
88.3551 
0.58 
8 
98.5708 
0.54 
9 
105.3885 
0.41 
10 
110.0155 
0.41 
11 
117.3167 
0.40 
12 
122.5048 
0.43 
Table 1. Calibration results of B2 
Judging from the histograms, the data follows the normal 
distribution. The j 2 test is used to checkout whether the 
distribution followed the normal distribution and the result 
shown that it satisfied the test condition. So the corrected 
calibration data of CBERS 02B CCD camera obeys the normal 
distribution. 
The object of relative calibration is to eliminate the 
nonuniformity of output DN values under the uniform input 
illumination. All output DN values should be the same after 
correction. But they can’t be identical actually because of the 
calibration error. The calibration accuracy is about 2% now and 
the range of DN values is about 2~3. So the data in a row of the 
average DN for each pixel can be described nearly by normal 
distribution. The distribution of the corrected and expected data 
in all radiance levels is shown as Fig.3. 
Fig.3. Distribution of the corrected and expected data 
In this figure, the symbol represents the corrected DN 
values and the lines correspond to the expected values in 
different radiance levels. 
From this figure, it can be seen that the corrected DN values are 
around the expected values. Some are bigger and others are 
smaller. According to the least-squares method, this 
phenomenon is correct. There is no certain rule between the 
corrected DN values and the expected values. The relationship 
depends on the original output data of each pixel. At the same 
time, there is no fixed relationship of the corrected values 
between different pixels. 
3 THE INFLUENCE ON IMAGE QUALITY UNDER 
DIFFERENT RELATIVE CALIBRATION ACCURACY 
3.1 Simulation method 
The difference between the ideal image (which is got in the 
situation that every pixel has the same output under the uniform 
incident radiance) and the actual image (which has been 
processed with relative correction) was simulated under 
different relative calibration accuracy in this paper. The results 
can be observed subjectively and evaluated by PSNR 
objectively. 
According to the analysis of corrected CBERS 02B CCD data, 
the simulation method is as follows (assuming it is 8-bit