Component Pattern Separation of Completely Unknown Fluorescent Images by Double Eigenvector Analysis

We have proposed a new pattern recognition method, for interpreting image data from fluorescence microscopes. In this method, component patterns included in multispectral images of completely unknown mixture samples can be evaluated, without knowing the spectra or spatial patterns of the components [1,2]. Principal component analysis and optimization theory are used with nonnegativity constraints and entropy minimization. This method discovers unpredicted or unknown components in various microscopic environments. Even completely new material to the human beings could be found by this method. However, the method does not guarantee to give the true solution of component patterns, but instead an optimal one under the criterion of entropy minimization.