Automated image acquisition and morphometric description

Abstract

Manual assembly of large morphometric data sets for investigations in population biology and biosystematics is extremely time consuming. Automated image handling allows rapid and accurate data collection and in addition makes it practicable to uncouple image acquisition and shape description. Variation can be characterized flexibly when these two phases of "scoring" are separated. Different descriptor systems can be used to describe a set of stored outlines, without having to return to the original objects. This paper introduces an integrated automated outline handling system and compares the relative performance of shape descriptor systems based on (i) landmark and perimeter measurements, (ii) chain codes, (iii) elliptic Fourier coefficients, and (iv) moment invariants. These methods were assessed by their ability to characterize the pattern of variation in a set of Betula leaf silhouettes with a hierarchical sample structure that allowed variation to be partitioned into within-tree, between-tree and between-species components. The descriptor systems using landmark and perimeter distances, elliptic Fourier coefficients, and moment invariants performed well, not only reflecting differences between species and between individual trees but also recovering the two-dimensional geometry of the within-tree sample positions. The elliptic Fourier method is judged to have performed best; the chain-code descriptor system performed poorly.