摘要： 针对两幅视差图像的拼接问题，提出了一种新算法，即利用Hessian仿射不变检测算子检测出特征区域，利用SIFT特征描述算子提取特征区域特征矢量，根据特征矢量的欧几里德距离来建立图像间的稀疏对应关系；由这些对应点稀疏地确定场景中的一些点，以这些点为顶点建立场景的三角面片近似，再据此将重叠区域重投影生成推扫式成像的中间部分图像。将中间部分推扫式成像图像和原左图像的左半部分以及原右图像的右半部分一起拼接生成大图像。利用实际图像进行的拼接实验表明该算法是一个有效的视差图像拼接算法。

关键词: 图像拼接, 视差, 宽基线

Abstract: A parallax image mosaic algorithm is presented. For two wide baseline parallax images whose epipolar lines are parallel with each other, the sparse correspondences can be got by detecting the local feature region by Hessian-Affine region detector, by describing the local feature region by SIFT descriptor, and by computing the Euclidean distance of the feature vectors. The sparse scene points can be got by those sparse correspondences, and the scene can be approximated as piecewise planar by those scene points. Then the scene is re-projected to create a push-broom image. The large image is got by mosaicking the push-broom image with the two original images. Example of constructing successful mosaics from real images is shown. The experiment demonstrates the feasibility of the algorithm for mosaic images with parallax.

Key words: Image mosaic, Parallax, Wide baseline

中图分类号: 

• TP391.4