一种适合于大尺寸航拍图像的特征点匹配方法

A feature point matching method for large-size aerial image

针对一些航拍图像尺寸较大的特点,提出一种快速有效的特征点匹配方法。给出了拉普拉斯金字塔与非均匀多方向滤波器组结合的特征点提取方法,实现图像滤波并提取局部极值点,在快速提取特征点的同时保留了特征点的尺度及方向信息。采用二值描述符对不同尺度下的特征点进行描述,描述符间距离通过异或操作快速求解。结合最小距离及次小距离确定匹配特征点对,使用误匹配阈值尽可能避免误匹配,同时使用RANSAC算法消除误匹配。试验结果表明,本文算法在保证特征点匹配率及正确率的前提下,有较高的效率。

Aimed at some large-size aerial images, a fast and effective feature point matching method is proposed In this paper, we use the feature point extracting method with the combination of Laplacian pyramid and a specified non-uniform N-dimensional directional filter to realize image filtering and extract the local extreme points. The direction and the scale information can be kept while making quick extraction of the fea- ture points. Then we apply a binary descriptor to depict feature points in different scales, and the distance of descriptors can be quickly computed by XOR operation. Using a specified threshold to avoid mismatching, applying RANSAC to eliminate mismatching points, the paper presents experimental results to show that the proposed method is efficient for large-size aerial image with the assurance of the matching rate and accuracy.