An analysis technique for biological shape-II.

Abstract

A discrete function representing the curvature of closed contours has been derived. The function is easily computable and has properties desirable for use in shape analysis. Two curvature measures (as well as the traditional P2A) have then been applied to a set of shapes extracted from leukocyte nuclei in order to evaluate their effectiveness as numeric descriptors of shape complexity. Bending energy, (BEN), was shown to be, on the average, the most sensitive indicator of complex shape, normalized mean absolute curvature (NMAC) was next, and P2A was last. Bending energy could also have important biolgical significance due to its possible relation to developmental processes. Further studies on bending energy will be necessary in order to substantiate this possibility. Many of the discrepancies between P2A, NMAC, BEN, and the study subjects are explainable by the lack of "syntactic", or form, information used by the computer features. If an observer really classifies an object by decomposing it into simpler pieces, then a more sophisticated shape analysis procedure must be used. Such a procedure will be described in a sequel to this paper.