摘要
针对现有的较优立体匹配算法多适用正直摄影条件,而对于存在较大几何变形的倾斜立体影像其匹配可靠性大为降低的问题,提出一种基于SIFT和几何约束的倾斜立体影像匹配传播算法。算法在初始种子匹配的基础上考虑影像场景的深度变化,针对影像的局部平面场景,采用顾及影像几何变形的归一化互相关(Normalized Cross Correlation,NCC)测度和局部单应约束进行匹配传播;而对于局部非平面场景,则采用基于核线几何约束及几何核线倾斜角改进的SIFT特征描述符完成匹配传播。实际数据验证了算法的有效性。
Existing excellent matching methods are mainly designed for vertical photogrammetry. However, for oblique stereo image pairs with significant geometrical deformations, these algorithms will fail. In order to improve matching reliability, this paper proposes a robust matching propagation algorithm based on scale invariant feature transformation algorithm （SIFT） and geometrical constraints. By considering the 3D image scenes and starting from initial seed matches, the proposed algorithm uses the normalized cross correlation （NCC） for local planar part of image scenes under homograpy mapping constraint. For local non-planar image parts of image scenes, the algorithm uses an improved SIFT descriptor under constraint of epipolar geometry only. Experiments on images of real-world scenes with significant projective deformations demonstrate the feasibility of the proposed algorithm.
出处
《遥感信息》
CSCD
北大核心
2016年第1期43-47,共5页
Remote Sensing Information
基金
国家自然科学基金(41371438)
关键词
倾斜立体影像
匹配传播
尺度不变特征变换
核线几何
单应映射
归一化互相关
oblique stereo image
matching propagation
scale invariant feature transformation
epipolar geometry
homography mapping
normalized cross correlation
