用于面诊的嘴巴定位算法研究

Mouth segmentation algorithm for facial diagnosis in traditional Chinese medicine

针对嘴巴区域特征的精确定位对于中医面诊客观化的研究具有重要意义,提出一种基于双肤色模型、椭圆拟合与改进C-V水平集模型相结合的分割嘴巴的方法。考虑到肤色在像素空间邻域与灰度值域的平滑相似性,首先给出一种二维伽马函数的自适应光照补偿法,提升非均匀光照下肤色聚类的稳定性,进而利用根据实验确定的双肤色模型,进行肤色检测,并采用数学形态学方法去除噪声等影响,Sobel方法提取轮廓;然后根据所得的边缘和初步轮廓利用直接最小二乘椭圆拟合法提取人脸区域;最后采用改进C-V水平集模型对嘴部区域进行分割。实验结果表明,采用该算法能够得到更好的分割效果,满足中医面诊图像分割要求,为面部五官的进一步分割和检测奠定基础。

The precise location according to the characteristics of the mouth area is of great significance for the study of the objectification of traditional Chinese medicine. Therefore, a method based on double skin color model, ellipse fitting and improved C-V level set model is proposed in this study. With the consideration of the smooth similarity of skin color in spatial neighborhood and gray-level domain, an adaptive illumination compensation method based on two-dimensional gamma function is firstly proposed to improve the stability of skin color clustering under non-uniform illumination. Subsequently, the experimentally determined double skin color model is used to perform skin color detection;and mathematical morphology is used to remove noise and other effects;and Sobel method was used to extract the contours. Based on the obtained edge and preliminary contours,the face region was extracted with direct least squares ellipse fitting method. Finally, the mouth area was segmented with the improved C-V level set model. The experimental results reveal that the proposed algorithm can obtain better segmentation results and meet the requirements of facial diagnosis in traditional Chinese medicine, laying a foundation for further segmentation and detection of facial features.