基于视觉注意和支持向量机的舌体自动分割方法的探讨  

Automatic tongue image segmentation based on visual attention and support vector machine

作　　者：刘哲[1] 陈家旭[2] 赵宇明[1] 苏群[1] 徐晓新[1] 周玉华[1] 隋芯[1] 

机构地区：[1]北京中医药大学信息中心,北京100029 [2]北京中医药大学基础医学院

出　　处：《北京中医药大学学报》2013年第1期18-20,共3页Journal of Beijing University of Traditional Chinese Medicine

基　　金：北京中医药大学创新团队项目(No.2011CXTD-07)

摘　　要：目的探讨基于视觉注意机制和支持向量机(support vector machine,SVM)相结合的舌体自动提取方法,为模式识别方法应用到舌体图像分割提供新思路。方法将舌图像经过视觉特征提取、高斯金字塔多尺度变换,依据多特征图合并策略生成显著图并进行二值化;在不需要人工干预的情况下,从显著区和非显著区分别随机选取正类训练样本和负类训练样本;机器自动学习样本创建SVM分类器,最后用训练好的SVM分类器对完整舌图像进行分割。结果获得的正常舌、裂纹舌、齿痕舌等多种舌象的分割效果良好,没有特征信息丢失的情况,并具有一定的抗噪能力。结论基于视觉注意和SVM舌体自动分割方法在无需任何先验知识的条件下,具有较稳定的分割效果,为模式识别应用到舌体图像分割中作了初步探索。Objective Objective To probe into automatic extractive method of tongue images based on the combination of visual attention mechanism and support vector machine （ SVM）, and offer a new thinking train for applying pattern recognition method in tongue image segmentation. Methods After visual feature extraction and Gaussian pyramid muhi-scale transformation, tongue images were created to saliency maps and then given binarization according to multi-feature merge strategy. The positive training samples and negative training samples were randomly selected from significant and non-significant areas without manual intervention, and SVM classifier was generated through sample automatic learning. Finally, the trainined SVM classifier was used to segment complete tongue images. Results This method had good segmentation effect on normal tongue, fissured tongue and teeth-marked tongue and anti-noise ability, and feature data was not lost. Conclusion The automatic segmentation method of tongue images based on visual attention and SVM has a stable segmentation effect in the condition without any priori knowledge, which is an initial exploration to pattern recognition application in tongue image segmentation.

关 键 词：舌图像分割 视觉注意 支持向量机 模式识别 

分 类 号：R241.25[医药卫生—中医诊断学]