一种新的MLBP-Otsu算法及在舌裂纹分割中的应用

中医舌诊中只有精确、完整地分割出舌裂纹，才能准确地对舌裂纹信息进行定量化的特征表示与描述，而目前少有方法能达到此要求。为此，提出了一种新颖的基于改进的局部二值模式（Local Binary Pattern，LBP）算子与Otsu阈值分割技术相结合的舌裂纹分割算法（MLBP-Ostu算法）。根据舌裂纹的纹理特征对传统LBP的模式分类方法进行重定义；引入一个粗糙度度量因子R]，如某区域的R]值偏小或偏大，则把其归类为非裂纹区，不进行LBP特征值计算；利用Otsu方法对前面得到的LBP特征图进行阈值分割，从而得到舌裂纹的二值图像。实验结果表明，该算法能较精确、完整地分割出舌裂纹，并且不需要预先分割出舌体区域，为将来舌诊的定量化研究及临床应用提供了有效手段。

New MLBP-Otsu method and its application in tongue crack image segmentation

To represent and describe the tongue crack information of tongue diagnosis in Traditional Chinese Medicine （TCM）, it needs to segment tongue cracks from tongue image correctly and completely. But few methods can meet the requirements. In this paper, a novel method based on the combination of modified LBP （Local Binary Pattern） operator and Otsu’s threshold technique is proposed, i.e., MLBP-Ostu algorithm. The principle of classification of the traditional LBP local pattern has been redefined according to the texture features of tongue tracks. A roughness measurement factor R] is introduced. The region with over-high or over-low R] value will be regrouped into the meaningless subclass, and the LBP operator will not operate on the central pixel of this region. To acquire the final binary image with true cracks, the Otsu’s method is used to threshold the LBP feature image obtained by the fore step. Experiments on the several of typical tongue images show that the method proposed in this paper can achieve correct and complete segmentation of tongue cracks. In addition, it isn’t necessary to segment the tongue body part from tongue image in advance. So, it will be an effective tool for the research on the quantification of tongue diagnosis and its clinical application.