ColorMeter和VISIA-CR图像提取肤色的方法对比

Comparing methods of measuring skin color using ColorMeter and VISIA-CR images

目的:建立基于VISIA-CR图像提取中国人群面部肤色表型的方法,比较VISIA-CR图像提取的肤色参数与ColorMeter肤色仪测量值的一致性,为利用面部图像数据提取肤色表型建立参考方法。方法:基于机器学习模型自建算法自动批量提取VISIA-CR图像的肤色参数,并比较提取面积、位置和光源模式对结果的影响;随后利用256例健康人群面部VISIA-CR图像提取的肤色参数与ColorMeter肤色仪测量值进行了Pearson相关性分析。结果:VISIA-CR标准光和交叉偏振光两种模式下提取的肤色参数强相关(L^(*),R=0.86;a^(*),R=0.90;b^(*),R=0.85;P<0.05),相邻或对称位置下不同面积的肤色参数也强相关(R>0.82,P<0.05),提示光源模式、相邻或对称位置和测量面积对VISIA-CR图像提取结果的影响不大。VISIA-CR图像提取的肤色参数与ColorMeter测量值有相关性(P<0.05),其中交叉偏振光模式下,较大提取面积的肤色参数L^(*)值与仪器测量值强相关(R>0.73)。结论:该方法可从VISIA-CR图像中准确提取肤色参数且能较好解决肤色不均的测量偏差问题,交叉偏振光模式和较大提取面积下的肤色参数L^(*)值更接近ColorMeter仪器测量结果,推荐用于肤色表型的提取。

Objective: To establish a method for producing facial skin color phenotype of Chinese population based on images captured by VISIA-CR, and to compare the consistency of skin color parameters produced by using VISIA-CR image with that produced by using ColorMeter images, in order to establish a standard for producing skin color phenotype using facial image data. Methods: Skin color parameters of VISIA-CR images were extracted automatically by machine learning algorithm, and the effects of extraction area, location and light source mode on the results were compared. Subsequently, skin color parameters were extracted from 256 healthy subjects’ facial VISIA-CR images, and Pearson correlation analysis was performed with the measured values of ColorMeter. Results: Skin color parameters extracted under VISIA-CR standard light and cross-polarized light were significantly correlated(L^(*),R=0.86;a^(*),R=0.90;b^(*),R=0.85;P<0.05), and skin color parameters of different areas under adjacent or symmetric positions were also significantly correlated(R>0.82,P<0.05), suggesting that light source mode, adjacent or symmetric measurement area had little influence on VISIA-CR image extraction results. The skin color parameters extracted from VISIA-CR images were correlated with the measured values of ColorMeter(P<0.05). In the cross-polarized light mode, the color parameter L^(*) value of the larger extraction area is significantly correlated with the instrument measured value(R>0.73). Conclusion: This method can stably extract skin color parameters from VISIA-CR images, and it can better solve the deviation problem in measuring uneven skin color. The L^(*) value of skin color parameters extracted from the larger area of cross-polarized light mode is closer to the result of instrument measurement ColorMeter, so it is recommended to be used for skin color phenotype extraction.