图像拼接--Construction and Refinement of Panoramic Mosaics with Global and Local Alignment

Construction and Refinement of Panoramic Mosaics with Global and Local Alignment
International Conference on Computer Vision , 1998 , 48 (2) :953

本文针对图像序列做全景拼接

本文的三个改进的地方：
1） 这里我们没有使用 cylindrical or spherical coordinates，因为 these representations introduce singularities near the poles of the viewing sphere

我们采取的策略是 associating a rotation matrix (and optionally focal length) with each input image, and performing registration in the input image’s coordinate system (we call such mosaics rotational mosaics)

2）对于多图像拼接问题，我们需要处理一个问题： accumulated misregistration errors，这里我们使用一个 global optimization technique, derived from simultaneous bundle block adjustment in photogrammetry [10], to find the optimal overall registration. 对所有图像序列进行一次整体的对齐，以便解决累积对齐误差

3） any deviations from the pure parallax-free motion model or ideal pinhole (projective) camera model may result in local misregistrations，which are visible as a loss of detail or multiple images (ghosting)
解决方法： we compute local motion estimates (block-based optical flow) between pairs of overlapping images, and use
these estimates to warp each input image so as to reduce the misregistration

整个算法的流程如下：
1）使用 rotational motion model 我们得到一个初步的整体拼接图
2）使用 global alignment (block adjustment) 对所有图像的整体拼接误差优化调整
3） local alignment (deghosting) algorithm 用于降低局部对齐误差

2 Alignment framework 对齐框架
我们采用一个 hierarchical motion estimation framework，包括四个部分： (i) pyramid construction, (ii) motion estimation, (iii) image warping, and (iv) coarse-to-fine refinement.

2.1 8-parameter homographies
两个图像重叠区域，我们可以使用 a planar perspective motion model 来描述其关系
The 8-parameter algorithm works well provided that initial estimates of the correct transformation are close enough.
However, since the motion model contains more free parameters than necessary, it suffers from slow convergence
and sometimes gets stuck in local minima. For this reason, we prefer to use the 3-parameter rotational model described next.

2.2 3D rotations and zooms
一个简化版本模型用于建模图像重叠区域的对应关系

3 Global alignment (block adjustment)
主要解决这个问题： accumulated misregistration errors

1）Corresponding points between pairs of images are automatically obtained using patch-based alignment.
2） Our objective function minimizes the difference between ray directions going through corresponding points,andusesarotationalpanoramicrepresentation.
3） The minimization is formulated as a constrained least-squares problem with hard linear constraints for identical focal lengths and repeated frames.

4 Local alignment (deghosting)
to compute the flow between all pairs of images, and to then infer the desired local warps from these computations

11