In this study, an image-based measurement system was developed for human facial skin colour, involving the development of a digital imaging system, collection of facial skin colour from 60 human subjects, generation of different colour characterization models, and performance evaluation. The factors that affect facial skin colour characterization, including different training datasets (two colour charts and the collected facial skin colour dataset), mathematical mapping methods (linear transformation, polynomial regression, root-polynomial regression and neural network) and camera image formats (JPG and RAW), were investigated and quantified not only by the conventional method of CIELAB colour difference, but also two newly introduced measures, facial colour contrast and skin colour gamut. The results indicate that the RAW image format for camera digital signals gave a more stable performance than the JPG format images, and the higher order polynomial regression with good predictive accuracy in terms of CIELAB colour difference did not perform well for the whole facial image. It is suggested to evaluate the model performance using the colour of both specific facial positions and the overall facial skin colour. Our comparative analysis in this study provides useful guidance for determining the colour characterization model for facial skin.

中文翻译：

---

基于图像的人脸肤色测量系统的开发

在这项研究中，开发了一种基于图像的人类面部肤色测量系统，包括开发数字成像系统、收集 60 名人类受试者的面部肤色、生成不同的颜色表征模型和性能评估。影响面部肤色表征的因素，包括不同的训练数据集（两个颜色图表和收集的面部肤色数据集）、数学映射方法（线性变换、多项式回归、根多项式回归和神经网络）和相机图像格式（JPG和 RAW），不仅通过 CIELAB 色差的常规方法进行调查和量化，而且还通过两个新引入的措施，面部颜色对比度和肤色色域进行了调查和量化。结果表明，相机数字信号的RAW图像格式比JPG格式的图像表现更稳定，CIELAB色差预测精度好的高阶多项式回归对整个人脸图像表现不佳。建议使用特定面部位置的颜色和整体面部肤色来评估模型性能。我们在本研究中的比较分析为确定面部皮肤的颜色表征模型提供了有用的指导。建议使用特定面部位置的颜色和整体面部肤色来评估模型性能。我们在本研究中的比较分析为确定面部皮肤的颜色表征模型提供了有用的指导。建议使用特定面部位置的颜色和整体面部肤色来评估模型性能。我们在本研究中的比较分析为确定面部皮肤的颜色表征模型提供了有用的指导。