Advanced Algorithm for Determining Component Spectra Based on Principal Component Analysis

Abstract
A method is proposed for determining the component spectra from a set of unknown two-component mixture spectra. This method is considered as an advanced version of the Lawton-Sylvestre self-modeling curve resolution. The disadvantage of the weakness against noise is overcome by the introduction of the concept of filtering; the uncertainty of the solution is removed by the reduction of the band for each curve to a single curve with aid of the concept of entropy minimization. Experimental results with infrared absorption spectra of xylene isomer mixtures and visible absorption spectra of dye mixtures demonstrate the power of this method.