基于稀疏特征匹配和形变传播的无缝图像拼接

Sparse Feature Matching and Deformation Propagation for Seamless Image Stitching

该文给出了一种基于稀疏特征匹配和形变传播的无缝图像拼接方法。首先,在配准图像的重叠区域中寻找一条结构误差最小的最佳接缝,从一边的图像中选取目标区域;接着,沿着接缝在两边的图像区域中检测出显著的结构特征,并进行特征匹配,获得目标区域中接缝上的匹配特征点及与之关联的边缘特征点的结构形变矢量;然后,通过求解泊松方程,将这些稀疏的形变矢量稳定和平滑地传播到目标区域内部,得到目标区域中各点的形变矢量;最后,由形变矢量通过内插获得目标区域的梯度场,并由梯度场重构出最终结果。该方法执行方便快速,不需要复杂的特征检测,能够统一地纠正图像拼接中较大的结构错位和颜色过渡不自然,在全局上消除结构接缝和颜色接缝。与其它方法比较,该方法获得较明显的改进。

This paper presents a novel approach for seamless image stitching which is based on sparse feature matching and deformation propagation. First, an optimal partitioning which minimizes the structure error is found in the overlap region between the registered images, and the target region is selected from one side of partition boundary. Then, the salient structure feature is detected and matched along the partition boundary, which gets some sparse deformation vectors corresponding to the matched feature points and their associated edge points. By solving Poisson equations, these sparse deformation cues will then be propagated robustly and smoothly into the interior of the target region in which the deformation vectors of all points are derived. Finally, the gradient map of the target region is derived by interpolating the deformation vectors, from which the result is reconstructed. The implement is convenient and fast, and complex feature detection is needless. The proposed approach can handle significant structure and intensity misalignment in image stitching simultaneously, and eliminate both structure seam and intensity seam globally. Compared to several other methods, obvious improvement is achieved.