视觉显著性与共分割框架融合的舌象分割研究

项目名称： 视觉显著性与共分割框架融合的舌象分割研究

项目编号： No.61202318

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 李佐勇

作者单位： 闽江学院

项目金额： 24万元

中文摘要： 舌象分割是中医自动化舌诊中的关键环节，舌体纹理、形状上巨大的个体差异性以及弱边缘性给舌体分割带来了极大的困难。现有的方法采用边缘检测结合主动轮廓模型来提取舌体轮廓，对舌体弱边缘检测效果不佳，难以剔除脸部褶皱以及舌体内部纹理等引起的虚假舌体边缘，且容易发生口腔空隙、牙齿以及舌尖部分的误分割。本项目以舌象分割为基本背景，以提高分割的准确性为切入点，探索共分割启发下，融合视觉显著性的舌象分割新框架。首先，结合阈值分割、边缘检测、聚类进行舌体区域粗定位研究，自动剪切出舌体所处的大致区域；然后，研究共分割思维启发下，融合视觉显著性的舌体轮廓点提取，结合舌体形状先验知识剔除虚假轮廓点，连接剩余轮廓点形成舌体初始轮廓；最后，研究主动轮廓模型，修正初始轮廓，使之收敛于真实轮廓。此项研究将有效地提高舌象分割的性能，促进中医自动化舌诊的推广，具有良好的应用前景；同时可以促进视觉显著性、共分割理论与方法的发展。

中文关键词： 视觉显著性；稀疏表示；边缘检测；舌象分割；图像去噪

英文摘要： Tongue image segmentation is a critical step in Chinese automatic tongue diagnosis. The great difficulty involved could be attributed to large personal difference on tongue's texture and shape, as well as weak edge. Existing methods usually utilize edge detection in combination with active contour model to extract the contour of physical tongue, with insufficient effect on weak edges. The methods also fail to rule out fake edge caused by face wrinkles, textures in tongue body and so on. In addition, these approaches are more liable to misclassification of oral cavity space, teeth and tongue tip. This project aims to improve segmentation accuracy by exploring an innovative segmentation framework with the motivation of image cosegmentation under the background of tongue image segmentation. The proposed scheme first adopts image thresholding,edge detection and clustering to approximately locate tongue body region, i.e., automatically separate the rough area where tongue body positioned from its neighborhood part. Secondly, contour points of tongue body are extracted by means of visual saliency under the inspiration of image cosegmentation idea, while those fake contour points are excluded with the help of prior information about tongue body shape and remainder contour points are then connected to develop initial co

英文关键词： Visual saliency；Sparse representation；Edge detection；Tongue image segmentation；Image denoising