Analysing spectral data: comparison and application of two techniques

This study compares two different methods of condensing complex spectral data into interpretable indices or metrics of colour. Colour spectra were measured on standardized Munsell colour paint chips and estimates of the three fundamental components of colour (brightness, chroma and hue) were generated using principal component analysis (PCA) and segment classification (SC). We tested the accuracy of these techniques for generating colour component estimates from spectral data in four different 'colour classes' (red/orange, yellow/ green, green/blue, and blue/purple) and in an aggregate data set containing all measured spectra. We conclude that both analysis techniques will generally provide more information than more conventional colour assessment techniques. Neither technique was superior under all conditions; appropriateness of each technique depends on the system under study. Efficacy of both techniques depends on the variability present in the data set; greater variance in colour tends to reduce overall resolution. An application of tfiese techniques is also presented in which we characterize sex-specific differences in colour between different populations of the poecilliid fish, Limia perugia.