From image processing to classification: III. Matching patterns by shifting and stretching

Abstract

A method for the classification of electrophoretic patterns is described and tested on a data set representing ten wheat varieties. The method attempts to match each electropherogram to each variety by a transformation involving displacement and stretching along the x-axis. This is done essentially by the method of least squares, which uses only the information contained in the electropherogram itself to adjust it to the variety in question. The method is completely automatic and works extremely well by classifying 98% of the spectra correctly, judged by cross validation.