Research On Skin Color Measurement Based On Digital Camera

The color characteristics of the skin are usually expressed by chromaticity value or spectral reflectance.Nowadays,it is becoming more and more common to research the measurement of skin color through spectral reflectance.The reflectance of the skin can generally be measured with an instrument,but it can be affected by various factors when using the instrument.In contrast,it is more practical to reconstruct the spectral reflectance of the skin from a camera image.Existing researches on skin reflectance reconstruction based on digital cameras are mostly carried out in a standard light environment in the laboratory.By using large-scale light boxes and other experimental instruments to control various interference factors including illumination uniformity,the accuracy of the result is satisfied,however,due to the limitations of experimental instruments,it is difficult for general laboratories to complete relevant experimental research.Therefore,this paper studies to reconstruct the skin spectral reflectance by using digital cameras,based on the target color characterization method in the open environment,that is,under the outdoor natural light conditions.This paper first uses JPG data and linearized JPG data to reconstruct the spectral reflectance and evaluates the reconstruction effect in the weighted polynomial regression algorithm of different orders.And it investigates the effect of the linearization of camera JPG data on the accuracy of spectral reflectance reconstruction based on weighted polynomial regression algorithm and demonstrates whether the JPG data needs to be linearized in the weighted polynomial regression algorithm.For the reconstruction of skin spectral reflectance under the natural light,this paper uses the spectrum of the standard D65 light source to calculate instead of the outdoor natural light spectrum.This paper improves four spectral reflectance reconstruction algorithms proposed by Amiri and Cao,and applies them to the spectral reflectance reconstruction of the skin.For the reflectance reconstruction algorithm,this paper first improves the weighted pseudo-inverse and weighted nonlinear regression algorithms proposed by Amiri,it uses the improved Gaussian function to weight instead of the inverse distance function which is used in the original method.The comparison result demonstrates that using the improved Gaussian function for weighting can improve the accuracy of outdoor skin reflectance reconstruction.Then,this paper uses the skin spectral reflectance data set as the training sample which is composed of personal skin spectral reflectance data and the reflectance data set of Munsell color patches as the training sample in the reflectance reconstruction algorithm and compares the effect with them in the algorithm.The results show that using the skin spectral reflectance data set as the training sample can significantly improve the accuracy of outdoor skin reflectance reconstruction.Using four different correction algorithms to correct the camera RGB to s RGB,this paper compares the effects of various correction algorithms on the reconstructed skin reflectivity and compares the use of 140 color cards and self-made skin color cards as the training samples for reflection during the correction process reconstruction of the reflectivity,finally it compares the results of the four reflectivity reconstruction algorithms.The results show that using a self-made skin color card as the training sample in the process of camera RGB calibration to s RGB can improve the accuracy of skin reflectance reconstruction;among the four skin reflectance reconstruction algorithms,using the second-order polynomial regression method based on LASSO to correct the camera RGB to s RGB in the minimum chromatic aberration group algorithm has the best effect,the average color difference is 3.0856(35)E_{0 0},and the average root mean square error is 0.0296.In addition,this paper also proposes a model based on division of brightness,which can effectively reduce the number of training samples used in the RGB correction process of the camera without greatly affecting the reconstruction accuracy of the algorithm and increase the calculation speed of the algorithm.