The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1270
conventional region-based image fusion using Canny
segmentation by both visual inspection and objective analysis.
Four objective measures are used in this evaluation, i.e.
Entropy, Mutual Information (MI), Spatial Frequency (SF) and
Relative Dimensionless Global Error in Synthesis (ERGAS)
(Eskicioglu, et al, 1995).
Built-up area data sets
Figure 2. Fusion of built-up 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
A visual inspection of results shown in Figure 2 reveal that the
Canny segmentation of Pan and MS images produces over
segmented results and the resultant regions are too tinny to tell
which one is useful. In contrast, the mean shift segmentation
produces sounder results. The segmentation trends of Pan and
MS images are nearly the same. The total number of regions in
MS image is slightly more than that in Pan image with the
mean shift segmentation but the relationship is reversed with
Canny segmentation . It is clear that the image fused with mean
sift segmentation is clearer than that with Canny segmentation.
The spatial texture of images fused with Canny segmentation is
disturbed due to over segmentation.
image
Entropy
MI
SF
ERGAS
1
B
6.8396
1.2108
20.2839
12.9820
G
7.5488
1.2141
25.9663
R
7.8040
1.2184
29.3771
Nir
8.1422
1.2126
32.6054
2
B
7.1512
1.1520
32.2683
12.0265
G
7.5669
1.1626
39.6594
R
7.6649
1.1792
42.1544
Nir
7.9947
1.1862
41.3579
Table 1. Quantitative results on built-up area. (Image 1 from
Canny segmentation fusion and image 2 from proposed fusion)
The objective evaluation is made to verify the visual analysis
and the results are given in Table 1. From Table 1 we can see,
that the Entropy in image 1 is larger than that in image 2 except
for the blue band. The MI in image 1 is also a slightly larger
than that in image 2. Of course, this doesn’t mean the quality of
image 1 is better than of image 2 as the noise will also increase
the information which can be verified by visual analysis. But
the SF in image 2 is much higher than in image 1. This means
the clarity in former image is higher than latter images which
are consistent with the visual inspection. The smaller ERGAS
implies error in all the bands is smaller.
Rural area data sets