基于图像处理的望诊面色自动识别研究

Study on Observation Diagnosis Automatic Complexion Recognition Based on Image Processing

目的结合图像处理与中医理论实现面色自动识别。方法根据中医望诊面色相关理论,采用基于YCbCr颜色空间的椭圆肤色模型和主动外观模型算法对面部皮肤进行感兴趣区域分割,采用RGB空间、HSV空间、Lab空间中的颜色直方图、颜色空间的统计特征以及局部二值模式(LBP)特征对各区域进行颜色与纹理特征提取,使用支持向量机、极限学习机、BP神经网络对提取到的面色特征进行识别比较。结果将面色识别区域分成8块。结合颜色特征、纹理特征和LBP特征时,面色识别率达89.08%。青、赤、黄、白、黑5种面色中,白色采用BP神经网络的分类准确率最高,达89.5%。结论本研究结合肤色检测、图像处理与中医望诊理论,可实现面部肤色自动识别。

Objective To realize automatic complexion recognition through combining image processing with TCM theory. Methods According to theories of TCM observation diagnosis complexion, elliptical skin color model based on YCbCr color space and active appearance model algorithm were used to segment the facial skin. Color and texture features were extracted for each region by using RGB space, HSV space, color histogram in Lab space, statistical features of color space and local binarization （LBP） features. Support vector machine, extreme learning machine and BP neural network were used to identify and compare the extracted facial features. Results The face recognition areas were divided into 8 blocks. Combined with color features, texture features and LBP features, the face recognition rate reached 89.08%. Among the five colors of cyan, red, yellow, white and black, the BP neural network on white complexion had the highest classification accuracy rate of 89.5%. Conclusion This study combines skin color detection, image processing and TCM complexion recognition theory to achieve automatic facial complexion recognition.