Workshop proceeding - final.pdf - Faculty of Information and ...
Workshop proceeding - final.pdf - Faculty of Information and ...
Workshop proceeding - final.pdf - Faculty of Information and ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Table 2 experiment results <strong>of</strong> several different algorithms<br />
Number <strong>of</strong> Image Success* Registration error<br />
pairs<br />
number (max, min, average)<br />
Changeable resolution 10 10 (6,0,2.6)<br />
algorithm<br />
NMI algorithm 10 8 (9,0,2.7)<br />
CR algorithm 10 0 Null<br />
Remark: *here a successful registration is whose position error is less than 10 pixels.<br />
(a) Image pair 1(Optical(left), IR(right))<br />
(299,120,-0.8)<br />
(b) registration result 1 <strong>of</strong> Changeable Res Algo<br />
(c) registration result 1 <strong>of</strong> NMI alg (304,125,-0.5) (d) registration result 1 <strong>of</strong> CR Alg (318,257,0.5)<br />
(e) Image pair 2(Optical(left), IR(right)) (f) registration result 2 <strong>of</strong> Changeable Res Algo (364,153,1.5)<br />
(g) registration result 2 <strong>of</strong> NMI alg (363,157,1.0) (h) registration result 2 <strong>of</strong> CR Alg (283,93,3.9)<br />
Fig 4 Raw image pairs <strong>and</strong> their registration results using different algorithms<br />
5. Discussion<br />
From the experiment results we can find that NMI <strong>and</strong> CR algorithms are unsuitable to the<br />
different source image registration because these two methods are dependent on the intensity<br />
distributions <strong>of</strong> the images, while the changeable resolution algorithm is related to the edge <strong>of</strong> the<br />
images, that is to say, the latter algorithm is dependent on the image features <strong>and</strong> independent with the<br />
image intensities. Though they have good results in many applications, the image registration methods<br />
31