基于图的半监督学习的遮挡边界检测方法  

Occlusion Boundary Detection Using Graph-based Semi-supervised Learning

作　　者：张世辉[1,2] 张钰程 张红桥[1] 李鑫[1] ZHANG Shi-hui ZHANG Yu-cheng ZHANG Hong-qiao LI Xin(School of Information Science and Engineering, Yanshan University, Qinhuangdao, Hebei 066004, China The Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province, Qinhuangdao, Hebei 066004, China)

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省计算机虚拟技术与系统集成重点实验室,河北秦皇岛066004

出　　处：《计量学报》2016年第6期576-581,共6页Acta Metrologica Sinica

基　　金：国家自然科学基金(61379065);河北省自然科学基金(F2014203119)

摘　　要：提出了一种基于图的半监督学习检测深度图像中遮挡边界的方法。该方法首先获取已标记的像素点和待检测深度图像中的像素点作为顶点构建连通无向图，其次提取无向图中各像素点的最大深度差特征和八邻域有效深度差之和特征组成特征向量，根据像素点的特征向量计算无向图中顶点之间的相似性并将该相似性作为无向图中对应边的权值，然后根据图的半监督学习思想判断无向图中待检测像素点是否为遮挡边界点，最后可视化遮挡边界点得到深度图像中的遮挡边界。实验结果表明，所提方法尽管只需少量的标记样本，但在准确性上却同已有基于监督学习的方法相当。A novel occlusion boundary detection approach is proposed for depth image by using graph-based semi- supervised learning. Firstly, the connected undirected graph is constructed with the labeled and unlabeled pixels as vertexes. Secondly, the feature vector of each pixel is gained by extracting its maximal depth difference and the sum of eight neighborhood effective depth differences, and the similarity between the pixels are computed as the weight of the corresponding edge in the undirected graph. Thirdly, the pixels to be detected in the undirected graph are labeled as occlusion or nonoeclusion boundary point according to graph-based semi-supervised learning idea. Finally, the occlusion boundary points in depth image are visualized and the occlusion boundary can be obtained. The experimental results show that, although only a small amount of labeled samples, the proposed approach is equivalent to the existing supervised-based learning method in accuracy.

关 键 词：计量学 图像识别 遮挡边界 图的半监督学习 深度图像 无向连通图 八邻域有效深度差 

分 类 号：TB96[机械工程—光学工程]