【作者】 肖洪涛；

【导师】 王宽全；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2007， 硕士

【摘要】 舌诊是中医学中最有临床应用价值的中医诊断方法之一。近年来,随着计算机科学与技术的迅速发展,传统中医舌诊学朝着计算机化方向发展已经成为必然趋势。本文正是试图推广计算机化中医舌诊的发展,着重进行了舌图像颜色训练和分类问题的方法研究。本文的主要贡献在于:分析了现有舌象颜色分析方法的不足之处;设计了基于像素的舌颜色分类系统结构;提出了一个基于半监督学习的像素分类算法,解决了基于像素的苔色质色分布模型的建立问题;提出了RKNN算法,将一个全局优化问题转化成一个动态局部问题,并将其应用于苔色质色分类中,解决了苔色质色海量像素分类的时间复杂度问题;并计算舌图像的“颜色比例特征向量”,将其应用在对舌图像的总体颜色分类和胰腺炎的自动诊断。首先,本文系统的总结了现有舌象舌色分析方法的不足之处,并根据这些不足产生的原因,正确地选择了舌象图像中的像素作为分类算法的研究对象。接着提出了一种基于半监督学习的医学生物特征识别的新算法,该算法的性能比监督学习与非监督学习的性能更好。其次,本文结合前向选择和后向选择的苔色质色分布模型的训练方法,建立了12种苔色质色的分布模型数据集,较好地解决了基于像素的舌颜色分类系统的颜色模型的建立问题,大大提高了训练样本的质量。本文进一步提出RKNN算法,应用其实现对苔色质色的分类,将一个全局优化问题转化成一个动态局部算法,使之适用于舌象图像中大规模像素的分类计算。最后根据舌图像的12维颜色比例特征向量,实现了对舌图像的整体颜色分类。最后,本文对训练样本和实验结果进行分析,并在中医专家的指导下,对胰腺炎的自动诊断的可行性及其自动诊断效果进行研究,取得了令人满意的成绩。更多还原

【Abstract】 Tongue Diagnosis is one of the most valuable methods in Traditional Chinese Medicine practice, and its compuerization is inevitable with the rapid development of computer science and technology these years. This dissertation is thus focused on advocating the development of computerized Tongue Diagnosis and researching on the methodology of tongue color training and classification.Major contributions of this dissertation include: analyzing the weaknesses of the current tongue color analysis methods; designing the algorithm of pixel-based tongue color classification system; proposing a semi-supervised method for tongue pixels classification; converting a global optimization problem to a dynamically local one, therefore significantly improving the processing speed; proposing an eigenvalue of color ratios and applying it to tongue color classification and the automatic diagnosis of pancreatitis.First of all, the weaknesses of the the current tongue color analysis methods are summarized systemically, based on the reason of which, pixels in the tongue images are selected as the research subjects of the classification. Later, a new medical biometrics algorithm is suggested based on semi-supervised learning with better performance than that of supervised and unsupervised learning methods.Secondly, by combining forward selection and backward selection, distribution model datasets of 12 tongue colors and pixel-based tongue color model are set up with higher quality training samples. Furthurmore, RKNN algorithm is proposed to classify tongue substance and coating color, converting a global optimization problem into a dynamically local one, suitable for the huge number of the pixels. Finally, a 12-dimension color ratio eigenvalue is applied for the whole tongue color classification.Last but not least, training samples and experiment result are analyzed. With the guidance and assistance of TCM specialists, the feasibility and the accuracy of the automatic pancreatitis diagnosis are also justified with a satisfying experiment result.更多还原