目的 舌诊客观化是中医现代化的重要内容之一,其中舌苔的量化描述是其重要指标.传统的舌苔薄厚分类只是定性地将舌苔分为薄厚两类,缺乏定量描述.本文提出一种灰度共生矩阵和小波纹理特征提取结合的方法.方法 首先采用Daubechies2正交小波对舌图像进行一级小波分解,并求3个细节子图的平均值和方差作为特征,然后结合灰度共生矩阵在一级小波分解后的近似子图中提取三个方向的对比度、逆差距、能量和相关性作为补充特征,在此基础上训练支持向量机(support vector machine,SVM)分类器,对舌苔薄厚进行定性判断,同时建立支持向量回归(support vector regression, SVR)模型对舌苔薄厚定量分析.结果 实验采取200例舌象样本,其中100例薄苔,100例厚苔,选取其中60幅薄苔和60幅厚苔样本进行分类器训练,其余80例样本作为测试.结论 实验结果表明,与传统的小波纹理特征提取方法相比,本文提出的方法能够提升舌苔薄厚的分类效果,并在定量分析中取得较好的结果.%Objective The objectification of tongue diagnosis is one of the important contents of modernization of Chinese medicine,and the quantitative description of tongue coating is its important indexes.The thickness of tongue coating in traditional classification is qualitatively divided into two categories:thick coating and thin coating,yet there is a lack of quantitative description.A method combining gray-level co-occurrence matrix(GLCM)and wavelet texture feature extraction is proposed in this paper.Methods Firstly,the orthogonal wavelet of Daubechies2 is adopted to decompose the tongue image into primary wavelet,and we calculate the average and variance of the three detail subgraphs as features.Secondly,GLCM is used to extract the contrast,inverse gap,energy and correlation of 0°,45°,90° the three directions in approximate subgraph.On this basis,classifier of support vector machine(SVM)is trained to analyze the tongue images qualitatively based on that features.At the same time,the support vector regression(SVR)model for quantitative analysis of tongue coating thickness is established. Results A total of 200 cases of tongue samples are taken,of which 100 cases are thin fur and 100 cases are thick fur,60 cases of thin fur and 60 cases of thick fur samples are selected to train the classifier, and 80 samples are test samples.Conclusions The experimental results show that compared with the traditional wavelet texture feature extraction method,the method proposed in this paper can improve the effect of classification of tongue coating thickness.
展开▼
