摘要
利用加速鲁棒特性(SURF)算法搜索影像的特征点,基于薄板样条(TPS)建立对应域之间连续和光滑形变变换,用于影像视差的估计,在此基础上提出一种新的立体影像密集匹配方法.首先,基于SURF算法进行特征匹配;然后,利用极线和TPS变换约束选择稳定可靠的同名点,计算影像之间的TPS变换矩阵,估计对应点的位置;最后,基于影像灰度进行密集匹配,并利用极线约束和TPS矩阵删除误匹配点.为保证视差估计的可靠性,利用种子点的外接凸边形对匹配区域进行限定.以人脸三维重建中的影像匹配为例,得到了稳定可靠且密集的同名点.
Speeded-up robust features （SURF） has more chance of finding corresponding feature points,which can be used as seed points for densely matching. Thin-plate spline （TPS） transformation can establish a smooth and continuous transformation relationship between corresponding regions which can be used to estimate parallax in matching. Based on the two methods,the paper proposes a novel matching solution. Firstly,a set of reliable seed points was obtained by filtering the corresponding points obtained using SURF with TPS. Then,the corresponding points of each pixel in left image were estimated by the TPS matrix. Thirdly,the dense match can be carried out between two images. Since the seed points and TPS transformation could estimate the coordinates of corresponding point more accurately,so the dense match can performed easily,reliably and quickly. At the same time,the matching region was confined which is the outer protruding polygon surrounded seed points. Finally,the errors corresponding points using TPS and epipolar restriction were deleted. This new method was clarified using some stereo images in human face reconstruction whose similarity is low very much. Using this method,a dense and reliable point cloud can be obtained.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2010年第7期91-94,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
中国博士后科学基金资助项目(20070410282)
