It is reasonable to refute the applicability of a skeletonizing approach in recognizing static hand gestures. The data set tested was representative of the domain, and so the interpretation of the results is sound. We don't believe that the applicability of slope density for finding an orientation is in question, nor the skeletonizing. Rather, we believe that the large errors from segmentation strongly influence the interpretation of hand features, namely the finger axes.

To overcome this difficulty, we propose applying LMS fitting for axes to help find a sensible measure. This sidesteps the need for any "eigenhand" description or pre-calibration, since LMS is insensitive to outliers up to half of the data. Exploring this statistical tool is currently in-progress.