2003 Volume 65 Issue 2 Pages 94-100
We presented a Holstein cow identification method and considered its robustness against various lighting conditions. The method divided a pattern region into many blocks and made the feature vector of each cow using the 0 or 1 number coded by a sign of the brightness difference between neighboring blocks. In the experiment, Euclid distance, cosine and normalized correlation was used for identification on eigenspace spanned by KL expansion. As the result, the sign coding method was more robust against various lighting conditions than the non-coding method, but a small positioning error of an identified pattern region caused its recognition performance to degrade. A new method with high robustness against the positioning error as well as various lighting conditions must be developed for practical use in natural scene.