The proposed method is tested over a dataset of hands from 100 subjects. The best obtained Equal Error Rate is . Results demonstrate that discarding features that are more prone to distortions allows lighter processing, but also produces better performance than using the full set of features. |