基于三角形几何相似性的图像配准与拼接

Image registration and stitching based on triangle geometry similarity

介绍了一种基于三角形几何相似性的图像配准方法。提取两幅待拼接图像的特征点,将每幅图像各自的重叠区域内或图像内容复杂情况下的选定区域内的特征点任意组合为三角形,得到分别对应于每一幅图像的三角形集合。然后根据定义的新的三角形表示方法,包括最大角方向和最小角方向,在两组三角形集合内层层筛选任意组合的三角形对,最终找到其中的匹配三角形对,即相似三角形对,从而找到匹配的点对。最后计算图像间变换矩阵,对图像进行拼接,得到了一张具有更宽视野的无缝拼接图。该方法只与特征点间相互几何位置有关,所以对两幅图像间的灰度差异、任意的旋转、缩放等都表现了很强的鲁棒性。

An image registration method based on triangle geometry similarity was proposed. Firstly, the Harris feature points were extracted from both images that would be stitched together. In overlapped area, or in the area that we manually selected in the case of the complex content of the image, we respectively got a triangle group by arbitrarily selecting three feature points from each image. According to the new representation of triangle, including the biggest directional angle and the smallest directional angle, the two triangle groups were filtered under some rule so as to find the matching triangles, such as the pair of triangles with geometry similarity, and then we got the matching points. Calculating the 4×4 image transform matrix with these matching point pairs and stitching the two images, finally, a seamless panorama with wider field of vision was obtained. The method only has relation with the geometrical position of the feature points, so it is robust to gray difference, rotation and scaling between two images.