F. Brent Neal

Shape Analysis and the Classification of Objects

Quantification of shape remains an area of active study in the field of image analysis and machine vision. We present a comparative survey of three approaches to shape measurement: classical dimensionless ratios, harmonic analysis, and invariant moments, showing their suitability for classification of objects and other statistical analyses, including quantitative structure-property relationships. We show that for topologically simple shapes and well controlled imaging conditions, all three methods can provide robust classification of objects.