舌诊图像点刺和瘀点的识别与提取

Tongue spots and petechiae recognition and extraction in tongue diagnosis images

计算机舌诊系统中,点刺和瘀血点是重要的舌象。基于斑点检测、支持向量机(SVM)和K-均值聚类算法,提出了对舌诊图像中点刺和瘀点的识别及提取方法。首先利用SimpleBlobDetector斑点检测算法检测斑点,并提取出斑点数量、大小和分布等特征值生成特征向量,再使用SVM进行点刺(瘀点)舌象识别。点刺(瘀点)提取同样基于斑点检测算法,提取斑点颜色特征,使用K-均值聚类将斑点聚类为多个小类簇,定义基于加权颜色空间距离的判别函数,将聚类结果同第一次斑点检测的结果对比,得到正类和负类,最终提取出点刺和瘀点。利用该方法进行实验,识别正确率达到97.4%,提取误检率为6.0%,漏检率为10.1%,表明了本方法的有效性和应用价值。

Tongues spots and petechiae are important tongue patterns in the computer tongue diagno- sis system. We propose a method to recognize and extract spots and petechiae in tongue images based on blob detection, support vector machine （SVM） and k-means clustering. Firstly, we apply the Simple- BlobDetector algorithm to detect blobs in tongue images. Secondly, we obtain the characteristic values of blob number, size and distribution to generate the feature vector. Thirdly, we utilize the SVM classifier to recognize tongues with spots or petechiae. The detection of spots or petechiae also bases on blob detection. Blob detection result is clustered into several groups by using k-means clustering after extrac- ting color features. To extract the spots or petechiae, we define a discriminant function based on weighted color space distance, compare the clustering results with the former blob detection results, and achieve a binary classification of clustering groups. The positive class is the extraction results. Experi- mental results show that the recognition accuracy can reach 97.4%, the false alarm rate is 6.0% and the missing alarm rate is 10.1%. The results also verify the availability and application value of our method.