【作者】 杜建强；

【导师】 卢炎生；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2008， 博士

【摘要】 舌诊是我国传统医学四诊“望、闻、问、切”中的重要内容,是中医诊法最重要的特色之一,在中华民族几千年的繁衍生息中发挥了重大作用。但是传统的舌诊方法主要依靠医生目视观察进行判断分析,缺乏客观评价依据,制约了舌诊的进一步应用和发展。以现代科学技术手段研究舌诊原理,使其更加科学化、客观化、定量化,已成为舌诊研究的必然方向。用数码相机拍摄的原始舌象通常由舌体、部分脸颊和上下嘴唇组成。由于舌体同某些背景颜色较为相近,而且舌体同背景之间不存在明显的分界线,因此自动分割的难度较大。将舌象图像转换到HSI空间,分析Hue分量和Intensity分量的分布情况,确定分量的阈值,然后对图像进行分割并二值化,对二值化图像采用序贯算法快速标记连通区域,获取舌体区域,再利用数学形态学算法填补舌体区域的细小空洞,最后通过图像运算获得舌体图像,可以快速实现舌体的自动分割。对于白苔覆盖整个舌面或程度较深的腐腻苔情况,融合颜色和空间信息构建相似性准则,较好地解决了舌体自动分割问题。临床舌诊中,舌质的颜色、舌苔的颜色及其分布是舌诊辨证论治的主要依据。采用图像分析技术将舌象的舌质和舌苔分离有助于舌诊客观化的研究。通过对舌象色度直方图的分析处理,自动地确定舌象上颜色的类数和各种颜色的大致中心位置。在此基础上,对于标准模糊聚类算法进行了改进并应用于舌象的舌质舌苔分离。实验结果表明,同标准FCM算法相比,该算法加快了聚类迭代速度,减少了系统运算时间。相对于阈值分割法,能够获得更加符合中医临床要求的舌质舌苔分离效果。在上述成果的基础上,研究了一种基于HSI和FCM的彩色图像快速分割算法。首先将彩色图像从RGB色彩空间转换到HSI空间,然后联合利用S(饱和度)分量和I(亮度)分量进行阈值分割,消除图像中的色彩噪声,最后针对H(色调)分量进行模糊聚类。根据色调数据的特点,修正了样本数据到聚类中心的距离计算公式,给出了统计有效样本权重的算法,对于有效色调值进行样本加权聚类,加快了聚类速度。实验表明,该算法的运算性能高于标准FCM算法,获得了较好的彩色图像分割效果。在舌质舌苔分离的基础上,根据舌质舌苔在舌面上的分布特点,提取舌象的空间特征作为识别准则的基础,可以准确地识别舌质和舌苔。更多还原

【Abstract】 Tongue inspection is not only main content of four diagnostic methods "observation, auscultation and olfaction, interrogation, palpation" for TCM(Traditional Chinese Medicine), but also one of most important characteristics of TCM diagnostic methods. It has played the significant role during thousand years for Chinese nation. Traditional tongue inspection mainly depends on doctor’s eyesight to diagnose a disease. The diagnostic result is not only restricted by doctor’s knowledge level and diagnostic experience, but also influenced by external environment such as light and temperature. It lacks of objectivity to be limited for further development. It is the only way to combine TCM expert’s clinical experiences with modern information technology to realize quantitative, objective and standard tongue inspection.Original tongue image captured by digital camera under standard light source situation usually contains tongue body, upper lip, partial lower lip and face. Because the color of tongue body and background is similar and no obvious boundary is existed beween tongue body and background, automatic segmentation is difficult. The Original image was converted to HSI space. It was segmented by threshold value of hue and intensity component before it was converted to binary image. Then the sequential algorithm was used to quickly mark connected area and morphologic method was applied to fill little holes in tongue area. Experimental result shows it is a fast algorithm with good segmentation effect. For cases that the white coating covered entire surface of tongue or tongue coating texture was putrid or greasy, spatial information was considered to construct similar criterion and the automatic segmentation of tongue image with tongue coating performed well.Among tongue image features, the color of tongue body and tongue coating and the distribution are main evidences for tongue inspection. According to characteristic of tongue image, it was transformed to HSI space from RGB space and smooth processing was carried for the hue histogram. Then the number of color and initial values of color centers were automatically determined. So standard FCM algorithm was modified and applied to separate the tongue body and tongue coating. Experiments indicate that this algorithm speeds up the clustering iteration, reduces the system operation expenses and enhances the algorithm usability. Compared with method of threshold value segmentation, it can obtain better separation effect of tongue body and tongue coating.On the basis of above result, a fast approach named as CISHF is presented to segment color image. Color image was transformed from RGB space to HSI space firstly. Then rough segmentation was done by threshold value of saturation and intensity to eliminate the noise. Finally hue data was clustered by fuzzy c-means. The formula was revised to calculate the distance from sample data to the cluster center according to characteristics of hue data. The weight of effective hue value was calculated to speed up the cluster process. Experiments show that the performance of the presented algorithm is higher than standard FCM method and better segmentation effect can be obtained.On the basis of separation of tongue body and tongue coating, recognition expression was constructed according to distribution regulation of tongue body and tongue coating and good recognition rate was obtained.更多还原