基于双层回归森林模型的头影测量图像结构特征点自动定位  

Anatomical Landmark Localization in Lateral Cephalograms by Using Two-Layer Regression Forests

作　　者：秦臻 戴修斌[2] 谢理哲[3] QIN Zhen;DAI Xiubin;XIE Lizhe(College of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;College of Geographic and Biologic Information,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;Jiangsu Province Key Laboratory of Oral Disease,Nanjing Medical University,Nanjing 210096,China)

机构地区：[1]南京邮电大学通信与信息工程学院,南京210003 [2]南京邮电大学地理与生物信息学院,南京210023 [3]南京医科大学口腔疾病江苏省重点实验室,南京210096

出　　处：《应用科学学报》2019年第4期481-489,共9页Journal of Applied Sciences

基　　金：国家自然科学基金（No.31671006,No.61671255）;江苏省青蓝工程计划基金;江苏省六大人才高峰项目基金（No.GrantJY-058）;南京邮电大学1311人才项目基金资助

摘　　要：为了实现X射线头影测量图像中结构特征点的自动定位,提出一种基于双层回归森林模型的头影测量图像结构特征点自动定位方法.首先从图像中提取外观特征训练第1层回归森林模型,通过该模型生成针对当前特征点位置的偏移距离图;然后从偏移距离图中提取上下文特征,并结合外观特征训练第2层回归森林模型;接着将双层回归森林模型用于待检测的X射线头影测量图像,预测出图像中每个像素关于目标特征点的偏移距离;最后根据回归投票方法求得结构特征点位置.实验结果表明,基于双层回归森林模型的自动定位方法能较准确地获得头影测量图像中结构特征点的位置.To automatically detect anatomical landmarks in cephalometric X-Ray images,a context-aware landmark detection method using two-layer regression forest models is proposed.First,it extracts appearance features from images to train the first-layer regression forest model,which can be used to generate a displacement map for each landmark per training image.Second,from the displacement maps,the context features are computed and combined with appearance features to train the second-layer regression forest.Then,by exerting the trained two-layer regression forest model on the new cephalometric X-Ray images to be processed,the displacement vectors of all pixels to each target landmark will be produced.Finally,the proposed method uses regression voting to acquire the landmark position in the testing image.Experimental results show that the proposed method has good performance in the detection of cephalometric landmarks in dental X-Ray images.

关 键 词：头影测量图像 结构特征点检测 上下文特征 回归森林 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]