基于Mask R-CNN的舌体分割方法

Tongue Segmentation Method Based on Mask R-CNN Classifier

舌体分割是计算机辅助中医舌象诊断的第一步,但其易受舌附近人体组织的影响,从而使分割难度增大。如何使用机器学习优化舌体图像的分析流程是当前研究的热点之一。针对此问题,应用卷积神经网络MaskR-CNN进行分割获得的舌体边缘比较准确,能提高舌体图像分割的准确性。基于Mask R-CNN的舌体分割首先收集舌体图像数据,对信息进行数据清洗,构建出舌体图像数据平台。然后,采用Mask R-CNN训练得到舌体分割模型,最后分别使用训练好的模型和GrabCut算法分割4种特征舌体图像,对两种方法的分割效果进行对比及分析。实验结果表明,该方法在分割效果等方面均优于GrabCut算法,在评价指标中评估结果准确率均高于90%,可以为舌体图像分割、中医舌象诊断提供技术支持。

Tongue segmentation is the first step of computer-aided tongue diagnosis in traditional Chinese medicine,but it is easily affected by the lips and skin around the tongue,which makes segmentation become more difficult.How to optimize the analysis process of tongue image by machine learning is one of the hot-spots in current research.In the background of this problem,the edge of tongue obtained by using convolutional neural network Mask R-CNN is more accurate,which can improve the accuracy of tongue image segmentation.The first step of the method was collected tongue image data,and then cleaned the tongue image information to construct a tongue image data platform.Then,the tongue segmentation model was obtained by Mask R-CNN training.Finally,four kinds of characteristic tongue images were segmented by using the trained model and GrabCut algorithm respectively.The segmentation effects of the two methods were compared and analyzed.The experimental result showed that this method was superior to other classical methods in segmentation effect,and the accuracy of evaluation results is above 90%,thus it can provide technical support for tongue image segmentation and tongue image diagnosis.