The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part BI. Beijing 2008
1132
269 input images with 65 X 65 pixels cropped from IRS-C source
images are used for test. Their signal-noise ratio is between 0.3 —
0.5, and different scene area is covered. Signal-noise ratio between
input image and its corresponding SPOT image is calculated,
shown in Table 1 and Figure 3.
Table 1. The relationship between signal-noise ratio from 0.3 to 0.5
and rate of successful matching
Signal-noise
ratio
Number of
input images
Number of
correct
matching
Rate of
successful
matching (%)
0.30-0.32
16
9
56.25
0.32-0.34
18
12
66.67
0.34-0.36
21
18
85.71
0.36-0.38
34
21
61.76
0.38-0.40
45
36
80.00
0.40-0.42
28
25
89.28
0.42-0.44
29
24
82.76
0.44-0.46
27
24
88.89
0.46-0.48
27
25
92.59
0.48-0.50
24
24
100.0
4.2 Summation of Image Gradient
Image gradient reflects image’s information content and the
amount of features the image contained, and it is the key factor for
feature-based matching. The value of gradient is great when image
contains rich prominent features. However, the value of gradient is
small for flat area, and it is zero for the area, whose grey-level is
invariable. Gradient calculator in common use is Robert, Prewitt,
Krisch arithmetic, etc. Following is an example for calculating
gradient located at (x, y) for image I(x, y) using Sobel arithmetic,
viz.:
G x = (/(x - \,y +1) + 2/(x,y +1) + I(x + l,y +1)
- (7(x -1, y -1) + 2/(x, y -1) + /(x +1, y -1))
G y = (/(x - l,y -1) + 21 (x -1 ,y) + /(x - l,y +1)
- (/(x +1, y -1) + 21 (x +1 ,y) + /(x +1, y +1))
Gradient at (x,y) for image I(x,y) is a vector, as follows:
V/ =
Gx
dl_
dx
a/
dy
(9)
Magnitude of gradient is mag(VI) = \G X +
From curve’s trend in Figure 3, we can see that the greater the
signal-noise ratio is, the greater the success-matching rate is in the
whole. But, there is some exception, for example, success
matching rate for images, whose signal-noise ratio is from 0.32 to
0.36, is greater than that of images whose signal-noise ratio is from
0.36 to 0.38. It is shown that images’ signal-noise ratio has no
strong relation to the success-matching rate for mutual information
based method. It is obvious that mutual information expresses
statistical characteristic of image’s grey value, good matching
result can still be obtained when non-linear change of image’s grey
value is taking place and images to be matched have lower signal-
noise ratio value.
The value of magnitude of gradient at very pixel is added together,
and then the summation of image gradient for whole image is
derived. It shows edges contained in the image and the change of
image’s grey-level. Therefore, summation of image gradient can
represent image’s information content. In this paper, experiment is
carried out for the relationship between the summations of image
gradient and success rate based on mutual information matching
method. It is also found out that there is no strong relation between
image gradient and matching success rate using mutual information
method. 267 input images, whose image gradient is from 4.0 e+005
to 6.0e+005, are used to test and detail statistic is obtained, shown
in Table 2 and Figure 4.
The curve in Figure 4 shows that image’s gradient magnitude has
Rate of successful matching
4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8 6.0
image gradient
Figure 4. Success rate of matching for images whose summation of gradient magnitude is from
4.0e+005 to 6.0 e+005