715 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
Compared the three spanning trees of mosaicing graph, we can 
find out that MST-based method does not consider the routing 
cost between pairs of images although it is very efficient. The 
panorama will not be in high quality. The quality of panorama 
based on MRCT is best among the three methods, but it would 
consume much time when there are a large amount of images to 
be mosaiced. So when high quality of the panorama is required 
and the amount of the images is to be mosaiced, it is a proper 
choice. The method based on SPST with median as root, which 
balance the efficient and the quality of the panorama, will be a 
practical choice for large scale microscope image mosaicing. 
4. EXPERIMENTS AND RESULTS 
In the experiments, microscope images are collected from tissue 
slice in 1600*1200 pixels under 40 amplification ratios. For 
images in the same row and between adjacent columns, the 
horizontal search range is from 70% to 92% and the vertical 
search range is from -10% to 10%. For images in the same 
column and between adjacent rows, the vertical search range is 
from 70% to 92% and the horizontal search range is from -10% 
to 10%. 
In general, number of microscope images from tissue slice in 
1600*1200 pixels under 40 amplification ratios will arrive 
several hundred; some of them even are over one or two 
thousand. Images of a tissue slice with 22 rows and 17 columns 
are tested in this paper, which have 709 conjoint relationships 
and many blank and vague images. 
In the local registration, 109 pairs of images fail and the right 
rate of registration is 84.635%, in which most of the registration 
failures come from blank images. 
Based on local registration, we adopt three methods to construct 
spanning trees of mosaicing graph as described in section 3, 
including mosaicing method based on minimum cost spanning 
tree, mosaicing method based on shortest path spanning tree 
with media as root and mosaicing method based on minimum 
routing cost spanning tree of 2-star. 
Image mosaicing method based on MST has some serious 
shadows in the panorama(Figure2). The method based on SPST 
and MRCT improve the quality much and there almost doesn’t 
exist obvious mistake in the panoramas, but parts of panorama 
based on SPST are more blurry than that based on 
MRCT(Figure3). 
The comparison of the four methods is shown in Talbe.l. From 
the table, we can see that the quality of panorama based on 
shortest path spanning tree with media as root is very similar to 
that based on minimum routing cost spanning tree but the 
algorithm of minimum routing cost tree is very complex and the 
method to construct panorama based on minimum routing cost 
tree consumes much more time than the method based on 
shortest path spanning tree with media as root does. 
Figure 2. Mistakes in Panoramic Image 
Based on MST 
Figure 3. Comparison the Same Parts from the Panoramic Image 
based on 
SPST and MRCT: (a) Part of Panorama Based on SPST; (b) Part 
of Panorama Based on MRCT.