中医舌象分割技术研究进展: 方法、性能与展望

中医舌诊的客观化、定量化研究是中医现代化发展中的重要课题. 数字化采集到的舌图像包括舌体及部分面部区域, 为了便于后续舌象自动分析, 需要首先将舌体部分从图像中分割出来, 分割效果将直接影响后续舌象特征分析的准确性. 基于传统方法的舌象分割技术虽然取得了很大进展, 但其性能仅能达到半自动分割, 对较难分割的图像往往需要借助人机交互来完成. 近年来, 深度学习技术在图像处理及计算机视觉等多个领域取得了突破, 其在图像语义分割任务中也取得了远超传统方法的进展. 基于深度学习的舌象分割技术已经基本实现了全自动的鲁棒分割. 本文首先从传统分割方法和基于深度学习的分割方法两方面对中医舌象分割技术发展中的主要方法进行综述; 其次, 采用我们收集的舌象数据库对典型的方法进行性能评估, 并对不同舌象分割方法的特点进行分析与讨论. 最后, 对中医舌图像分割方法潜在的发展方向进行了展望.

Review on Tongue Image Segmentation Technologies for Traditional Chinese Medicine: Methodologies,Performances and Prospects

The objectification and quantitative analysis of tongue diagnosis is an important topic in the development of traditional Chinese medicine (TCM) modernization. The digitally acquired tongue images include the tongue and part of the face region. In order to facilitate the automatic analysis of the tongue image, the tongue needs to be segmented from the whole image, and the segmentation results will directly affect the accuracy of the tongue image feature analysis. Although traditional methods of tongue-image segmentation has made great progress, their performance can only achieve semi-automatic segmentation. There are images that difficult to segment perfectly without human-computer interaction. In recent years, with the breakthrough of deep learning technology in the field of image processing and computer vision, it has achieved far more performance than traditional methods in the semantic segmentation tasks. The deep-learning based tongue-image segmentation technologies have achieved fully automatic robust segmentation. This survey gives a detailed overview of the history, state of the art, and typical methods in this domain. Firstly, the typical segmentation methods are presented. Then, they are used for migration learning and network testing based on our self-built tongue image database. In addition, this paper analyzes the characteristics of these segmentation methods and obtains the advantages and disadvantages of them. Finally, this paper summarizes the methods of Chinese medicine tongue image segmentation, and discussed to the development direction.