Determination of human skin optical properties from spectrophotometric measurements based on optimization by genetic algorithms

Abstract

We present an initial study on applying genetic algorithms (GA) to retrieve human skin optical properties using visual reflectance spectroscopy (VRS). A three-layered skin model consisting of 13 parameters is first used to simulate skin and, through an analytical model based on optical diffusion theory, we study their independent effects on the reflectance spectra. Based on a preliminary analysis, nine skin parameters are chosen to be fitted by GA. The fitting procedure is applied first on simulated reflectance spectra with added white noise, and then on measured spectra from normal and port wine stain (PWS) human skin. A normalized residue of less than 0.005 is achieved for simulated spectra. In the case of measured spectra from human skin, the normalized residue is less than 0.01. Comparisons between applying GA and manual iteration (MI) fitting show that GA performed much better than the MI fitting method and can easily distinguish melanin concentrations for different skin types. Furthermore, the GA approach can lead to a reasonable understanding of the blood volume fraction and other skin properties, provided that the applicability of the diffusion approximation is satisfied.