图像特征匹配中一种高效的鲁棒估计算法

An effective robust estimation algorithm for image correspondence

在图像特征匹配过程中,误匹配不可避免。提出一种新的基于拓扑约束(顺序约束和仿射不变约束)的外点去除算法,用于快速地去除图像粗匹配结果中的误配点。该算法对随机采样集进行拓扑过滤,只对满足拓扑约束的采样集进行计算。实验表明,该算法相比于传统的鲁棒估计算法RANSAC和改进的PROSAC算法,大大提高了计算效率并保持很高的计算精度,有助于提升图像匹配性能及3维重建的精度和鲁棒性。

Outliers are inevitable in image matching process. To address this issue, a novel topology constraint based outlier rejection algorithm is proposed to efficiently remove the mismatches between images after coarse matching. By using the topology constraint to filter the sample sets, the proposed algorithm calculates the transformation between images based on the sample set which fully satisfies the topology constraints. Experimental results demonstrate that the proposed algorithm can significantly reduce the computational complexity, while keeping the accuracy compared to the traditional RANSAC and improved PROSAC algorithms. Therefore, the proposed method can effectively and efficiently improve the performance of image matching, and furthermore benefits the application of 3D scene reconstruction in both accuracy and robustness.