改进BRISK特征的快速图像配准算法

Fast image registration approach based on improved BRISK

为了实现超分辨率图像重建中高精度快速图像配准,提出一种改进BRISK特征的快速图像配准算法。原有BRISK算法在特征提取和匹配过程中,忽视了角点分布信息,其匹配策略单一,导致误匹配率高。针对该问题,首先利用BRISK算法构建连续尺度空间,在此基础上对图像进行分块,然后利用图像区域显著性自适应选择角点检测阈值,获得均匀分布的角点,最后利用快速最近邻FLANN算法结合RANSAC的方法进行二值特征快速匹配。实验结果表明:改进的BRISK算法相比原算法在保持速度的基础上达到亚像素级配准精度,并具有优越的场景适应性能。

In order to realize accurate and fast image registration for super-resolution image reconstruction, a fast image registration approach based on improved BRISK feature was proposed. The BRISK algorithm neglected comer distribution information and the single matching strategy led to high false matching rate. To solve this problem, firstly, based on the original BRISK algorithm, a continuous scale space was built, and then the original image was divided into several blocks. After that, adaptive comer extraction thresholds were selected by the image saliency map to obtain uniform distribution of the comer points. Finally, the matching of binary features was carried out via FLANN algorithm and the RANSAC. The experimental results indicate that the improved BRISK algorithm enhance the accuracy of registration process up to sub-pixel compared with the original BRISK algorithm, with an advantage of scene adaptation, while nearly retaining its speed performance.