At figure 13, we present matching results on two images of a scene differing from viewpoint and from change of intensity. We have locally normalized the two images to make them independent from intensity variation according to the diagonal model (equation 4) and then computed the eight Euclidean invariants. The matches have been obtained using the iterative method described in details in [5] and summerized at section 4.2. There were 1170 Color Harris points found on the left image and 1035 on the right image (figure 12). The iterative matching process has found 403 matches (figure 13). The figure 14 shows the final epipolar geometry estimated after matching as explained at section 4.3. The reader can see that the results are very accurate in spite of the differences between the images. Some other results can be found in [6], [5], [12].
Figure 12: Points used for matching: 1170 Color Harris points on the left image (l) and 1035 on the right image (r)
Figure 13: Our matching algorithm: Matching scores computed with Euclidean invariance and invariance to change
of intensity, then improved by iterative constrained relaxation.
(l) left image, (r) right image
Figure 14: We obtain a precise epipolar geometry.
We have selectioned four corners of the roof of the house
at the right of one image,
the bottom of the tent at the top of the image,
a corner of the pannel down the image,
and we have drawn epipolar line on the other one.
(l) left image, (r) right image