基于直线特征约束网格变形的航拍图像拼接

针对常用航拍图像拼接算法在处理存在视差的复杂图像时会出现鬼影、失真、不自然的问题,在APAP(as-projective-as-possible)算法的基础上,添加全局相似性和直线结构保护约束,提出一种基于直线特征约束网格变形的航拍图像拼接方法.首先,在图像重叠区域检测点、直线特征,并估计单应性扭曲;然后,使用网格将图像划分成多个局部图像块,建立网格优化模型,根据网格顶点坐标集定义一个包含图像扭曲不同特征的总能量函数,在APAP算法提供的局部单应性保证重叠区域对齐的前提下,对非重叠区域借鉴NISwGSP算法的全局相似性约束思想,并辅以直线结构约束提高拼接自然度;最后,在稀疏线性系统中求解能量误差最小化问题,得到最优网格顶点集,指导网格变形并完成航拍图像拼接.实验结果表明,所提出的变形方法能够减少重叠区域鬼影和非重叠区域失真问题,相比AutoStitch、APAP、AANAP、SPHP等经典算法拼接效果更自然,更能适应大视角的航拍图像拼接任务.

Linear-preserve-mesh warps in aerial image stitching

Aiming at the problems of ghosting, distortion and unnaturalness in processing complex images with large parallax using the common aerial image mosaic algorithm, this paper proposes an aerial image mosaic algorithm based on linear feature constraint mesh deformation. The algorithm adds global similarity and linear structure protection constraints on the basis of the APAP(as-projective-as-possible) algorithm. Firstly, the features of points and lines are detected to estimate the homography distortion in the overlap region of the image. Then, the image is divided into several local image blocks to establish grid optimization models. According to the grid vertex coordinate set, a total energy function containing different features of image distortion is defined. On the premise that the local homography ensures the alignment of the overlap region provided by the APAP algorithm, the global similarity constraint of the NISwGSP algorithm as well as the linear structure constraint are used to improve the naturalness of stitching for the non-overlapping region. Finally, the optimal grid vertex set can be solved to guide the grid deformation and complete the aerial image mosaic in the sparse linear system. The experimental results show that the proposed algorithm can reduce the problems of ghosting and non-overlapping area distortion. Compared with AutoStitch, APAP, AANAP, SPHP and other advanced methods, the proposed algorithm is more natural in stitching effect and is more suitable for the aerial image mosaic task with a certain angle of view.