A natural yet relatively unexplored way to classify hand gestures is to first count the number of fingers identifiable in a hand image. Feature variations introduced through imperfect hand segmentation or unconstrained gesture input hampers the performance of present boundary-based finger identification techniques. In this paper, we describe an adaptive parameterless procedure that strategically decomposes hand boundary into meaningful finger tip and wrist segments. Experimental results showed that a Fourier based approach to decomposing boundaries outperforms Least-Squared based methods in the presence of simulated variances. This way of hand modelling builds up to a solid foundation for robust untrained hand gesture classification.