The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1271
(0 (j)
Figure 3. Fusion of rural area images. (a)Original Pan image,
(b)Original MS image, (c) Canny segmentation of Pan image, (d)
Canny segmentation of MS image, (e) mean shift segmentation
of Pan image, (f) mean shift segmentation of MS image, (g)
Canny segmentation fused result, (i) Our proposed fused result
As figure 3 shows, the test on rural area shows a similar result
to the built-up area. The image in Figure 3(c) is over segmented.
In Figure 3(d), some parts are over segmented and some are
under segmented. The segmentation results shown in Figure 3(e)
and (f) are more reasonable than those in Figure 3(d) and (e).
The fused result based on Canny segmentation is blurred as can
be seen from Figure 3(g) and (h). The sharpness and spectral
reservation of the images fused by this new method are better.
More detailed quantitative evaluation result is given below:
image
Entropy
MI
SF
ERGAS
1
B
6.6494
1.2668
8.4064
12.0409
G
7.4298
1.2648
10.2679
R
7.4779
1.2725
10.9206
Nir
8.4651
1.2521
18.0159
2
B
6.9650
1.1759
26.8664
11.7954
G
7.4381
1.1892
30.5736
R
7.4379
1.2147
25.9148
Nir
8.5994
1.2240
37.7858
Table 2. The quantitative results on rural area. (Image 1 from
Canny segmentation fusion and image 2 from proposed fusion)
6. CONCLUSIONS
In this paper, a new technique for the fusion of high-resolution
images has been described. In this technique, mean shift
segmentation is adopted to extract the features from images.
SSIM is used to measure the region similarity which has more
physical meaning. The SSIM is then used to guide the decision
making in the fusion process. Experimental evaluations have
been conducted for built-up areas and rural areas. The results
show this new technique performs better than the conventional
technique with Canny detection operator. It has been found that,
for high resolution image, the Canny detection tends to produce
unstable segmentation, i.e. over segmentation in a sub-region
and under segment in another region of the same images. On
the other hand, mean shift segmentation is more reliable. The
sharpness and spectral reservation of images fused by our
proposed technique are better than those by conventional
method with Canny segmentation.
This conclusions made here are based on the limited tests.
More comprehensive tests will be conducted in the future.
ACKNOWLEDGEMENT
This research was supported by State 973 project grant
2006CB701304.
REFERENCES
Comaniciu, D., Meer, P.,1999. Mean shift analysis and
applications. The proceedings of the seventh IEEE international
conference on Computer Vision, pp. 1197-1203.
Eskicioglu, A.M., Fisher, P.S.,1995. Image quantity measures
and their performance. IEEE Transactions on Communications,
43 (12), pp.2959-2965.
Lewis, J.J., O’Callaghan, R.J., Nikolov, S.G., Bull, D.R.,
Canagarajah, N.,2007. Pixel- and region-based image fusion
using complex wavelets. Information Fusion (8), pp. 119-130.
Luo, J.B., Guo, C.E.,2003. Perceptual grouping of segmented
regions in color images. Pattern Recognition, 36(12), pp. 2781-
2792.
Mo, D.K., Lin,H., Li, J.P., Sun, H., Xiong, Y.J.,2006. VHR
Imagery Multi-Resolution Segmentation Based on Mean Shift.
Journal of Guangxi Normal University, 24(4), pp. 247-250.