Reconstruction of spectral emittance of fluorescent dyes using image color and principal component analysis

Abstract

We reconstruct the spectral emittance of commercial fluorescence probes from the RGB (red (R), green (G) and blue (B) components) of their colored images and Principal Component Analysis (PCA) of their spectra. We show that there exists a transformation matrix that can map RGB into basis spectra coefficients (a_i). It was found that 12 basis spectra (e_i) give about 99% cumulative contribution and upon the inclusion of camera parameters it was found that six e_i having a cumulative contribution of 91.3004% is the optimum number that will give the most accurate reconstruction. In theory, it is expected to have better reconstruction with increasing number of e_i's, however the method employed in this study is restricted by residual error that propagates as the number of e_i increases. Reconstruction was tested using 3, 6, 9 and 12 e_i, and an equal number of camera channels. Error in reconstruction was analyzed using normalized root mean square (NRMS), cosine angle (θ) and correlation coefficient (μ).