基于舌图像多特征融合与机器学习的裂纹舌识别算法  

Crack tongue recognition algorithm based on tongue image multi-features fusion and machine learning

作　　者：赵颖 李玉双[1] 武小荣 ZHAO Ying;LI Yushuang;WU Xiaorong(School of Science,Yanshan University,Qinhuangdao,Hebei 066004,China;Beijing Bayes Health Technology Co.Ltd.,Beijing 100080,China)

机构地区：[1]燕山大学理学院,河北秦皇岛066004 [2]北京贝叶斯健康科技有限公司,北京100080

出　　处：《燕山大学学报》2022年第6期522-528,共7页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(61807029);河北省自然科学基金资助项目(A2020203021);河北省引进留学人员资助项目(C20200365)。

摘　　要：舌裂纹是中医舌诊辨证施治的重要信息源,能够客观、准确地反映某些典型疾病和中医证候的变化。针对传统的裂纹舌诊断易受医生经验、环境变化等因素的影响,提出了基于舌图像多特征融合与机器学习的裂纹舌识别算法。首先,采用Grabcut方法对原始舌图像进行舌体分割;然后,提取图像基于灰度共生矩阵的纹理特征,基于低阶颜色矩的颜色特征,以及基于方向梯度直方图的形状特征;最后,将三类特征及其不同的组合形式分别输入四个经典的机器学习模型,完成裂纹舌识别。实验结果表明:多特征融合往往有助于提高机器学习模型的识别能力,尤其是融合三类特征的自适应提升树(AdaBoost)取得了几乎能与深度学习模型相媲美的识别效果:AUC为0.97,准确率为0.91,精确率为0.91。可见,提出的裂纹舌识别算法有助于传统中医舌诊的客观化、定量化和标准化。Tongue cracks supply important information for tongue diagnosis of Traditional Chinese Medicine(TCM),it can reflect the changes of certain typical diseases and TCM syndromes objectively and accurately.In view of the fact that the traditional tongue crack diagnosis is easily affected by doctors′experience and environmental changes,a tongue crack recognition algorithm based on tongue image multi-feature fusion and machine learning is proposed.Firstly,the Grabcut method is used to segment the original tongue image.Then,texture features based on gray co-incidence matrix,color features based on low-order color moments,and shape features based on directional gradient histogram are extracted.Finally,the three types of features and their different combinations are input into four classical machine learning models to complete the crack tongue identification.The results show that the multi-feature fusion usually improves the prediction performance of models,the Adaptive Boostingwith three types of features has achieved results that are almost comparable to deep learning modelswith an AUC of 0.97,accuracy of 0.91 and precision of 0.91.In a word,the proposed algorithm could promote the objectification,quantification and standardization of tongue diagnosis.

关 键 词：中医舌诊 裂纹舌 机器学习 识别算法 颜色特征 纹理特征 形状特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]