Identification of starch granules using image analysis and multivariate techniques

Abstract

The full potential of ancient starch analysis will not be fully realised until comprehensive identification keys have been established. To address this problem multivariate analysis of data recorded from digital images can be used to construct an automatic system of classification. As a first step we used digital image analysis to record 18 variables for 1998 starch granules representing 29 species which were mainly collected within one area of Papua New Guinea and largely used for food or craft production. High success rates for classifying these into their correct plant taxa were achieved by three different methods of discriminant analysis. The results confirm earlier studies that have shown that the morphology of many starch granules, especially those derived from storage organs, is distinctive to a specific plant taxon. We also demonstrate that multivariate analyses can play an important part in the establishment of starch granule classification keys.