> Hello!
>
> I have doing project, and I have a problem.
> I am working in pattern classification project to classify circle ,
> rectangle and triangle, and I use some feature such as compactness,
> roundness, eccentricity and no. of point of convex hull , but it not
> give me needed accuracy, so I looking for another features.
> Can somebody point me to some feature I can extract it?
>
>
If you're classifying standard shapes like these, 2 or 3 good shape features are more than enough, for
example number of corvers and principal inertia axes ratio (max/min) for symmetry. If you still have
problems with the accuracy, perhaps the classifier you're using is not robust enough (try using simple
NN after some tests with MDC).
--
Harris