基于在线学习的亮度自适应皮肤区域检测方法

Luminance Adaptive Method for Skin Area Detection Based on Online Learning

提出了一种基于YCbCr色彩空间的亮度自适应的在线学习型肤色检测方法。在亮度自适应性方面,该方法通过考察肤色区域的颜色分量随亮度分量的变化规律,对传统的高斯模型进行了改进,同时提出了一种亮度自适应的阈值模型。在学习样本的选取方面,该方法利用人眼检测和基于结构的方法找出面部区域并以此作为肤色样本来计算模型参数。测试结果表明,该方法对光照变化及受试者肤色差异均有很好的鲁棒性,且在检测的正确率(Accuracy)和F1分数(F1-score)方面优于传统方法。

A luminance adaptive method for skin area detection in YCbCr color space was proposed based on online learning. In terms of luminance adaption, the traditional Gaussian model was improved by studying the relationship between color channels and luminance channel of skin area, and a luminance adaptive threshold model was designed. In terms of samples selection, human eyes detection and structure-based approach was utilized to find facial area, and hence the model parameters were obtained based on these skin samples. Detection results have shown that this method is not only very robust on the variation of skin color as well as on illumination changes, but also better on detection accuracy and Fl-score compared to traditional methods.