一种有效的舌体自动化分割算法*

舌体轮廓正确分割是实现中医舌诊信息化的重要前提.目前主流方法是用阈值方法或先验知识得到舌体的初始轮廓线,然后再用Snake模型使得曲线收敛到舌体边缘.但它们都存在两个问题:对比度较小的舌像,如舌面和脸部有相似的对比度的舌象,基本上没法处理;得到的初始化轮廓线无法克服嘴唇的影响,因此用Snake方法曲线常收敛于嘴唇边缘.通过对舌象的低层信息的研究,发现使用简单而有效的直方图均衡化能扩大图像的颜色分辨力,使得现有的阈值处理方法能在原有基础上很好地处理那些原来无法处理的舌体图像;然后根据舌体形状、位置等先验信息,用几何方法修正受嘴角和嘴唇影响得到的边缘.实验结果表明,该方法取得了很好的分割效果.

Effective Algorithm on Automatic Segmentation of Body of Tongue

The correct segmentation of the body of tongue is premise to the information on diagnosis by the feature of tongue in Traditional Chinese Medicine. The main method at present is to get initial contour of the body of tongue by threshold algorithm or prior knowledge, and then makes the initial contour converge into the edge of the body of tongue by Snake model. However, there exist two flaws in these algorithms: Firstly, it is difficult to deal with some tongue images with low contrast, which behaves in similar contrast between surface of tongue and face; secondly, the initial contour can not overcome the influence of lip, and the final curve often converges into the edge of lip by Snake. Researching on the low level information of tongue images,a simple and effective strategy to solve these two limitations is proposed. At first,it used histogram equilibrium before the threshold algorithm, with which the images which were failed to be dealt with can now be done well. And then according to the prior knowledge about shape and position of the tongue,a geometric method to correct the edge influence on lip or corners of the mouth is puts forward.According to the results of experiments, this algorithm is effective.