Bidirectional Matching Algorithm for Target Tracking Based on SIFT - Advanced Research on Computer Education, Simulation and Modeling - page 255

3 Target Matching and Tracking
3.1 Target Match
3.1.1 Matching Improvement
There is an improvement on the match of the SIFT features in the article. Take the
example of matching two images, the original match is calculating all the Euclidean
distance between a definite feature in image one and every feature in image two, as
shows in figure 3,if the nearest distance is shorter than 0.6 times of the second nearest
then we accept this match, but we have a problem here is that there could be two or
more points in image one are matched with one point in image two at the same time,
obviously, there are wrong matched points in image one [5].To avoid the appearing of
this kind of circumstance, we have a reverse match in the base of the original match in
this article, as shows in figure 4. The result of the experiment shows that it can remove
the wrong matched points efficiently. Although it cuts down the quantity of the
matched points, it makes the match more accuracy and stable.
Fig. 3. The original features match
Fig. 4. The reverse features match that added