关键词: 图像匹配, SURF, 积分图像, Hessian矩阵, 尺度空间, 特征点描述子

Abstract:

With the shortcomings of large data amount and long time consuming in the conventional image feature matching algorithms, a new algorithm based on SURF for image registration is presented in this paper. SURF is a new feature extraction algorithm which approximates or even outperforms other schemes with respect to distinctiveness, and robustness. And it is much faster than the other similar ones. It is fast computed based on the integral image and through the FastHessian detector, the feature points are extracted. For each feature point, the dominant orientation is assigned by computing the Haarwavelet responses, and then the descriptor is generated. Image matching is made based on the sign of the trace of the Hessian matrix and the ratios of the closest neighbor and the second closest neighbor, and an improved RANSAC technique is applied to eliminate outliers to ensure the effectiveness of the matched pairs. The experimental result shows that this algorithm can not only meet the requirement of accuracy, but also has a small data amount and fast speed for image registration.

Key words: image registration；speededup robust features；integral image；Hessian matrix, scale space；feature point descriptor