基于改进的SURF图像配准算法研究

Research on image registration based on Improved SURF algorithm

摘要:

图像特征点匹配算法是实现目标识别的一种有效算法,目前图像特征点匹配算法耗时大,而且在匹配过程中存在伪匹配点。提出了一种改进算法：在初始特征点检测阶段,根据图像大小动态构造高斯金字塔图层,提高了算法的实时性和准确性；采用设置阈值的方法对初始特征点进行优化,减少匹配时间。在特征点匹配阶段,利用提取特征点中正确匹配点与伪匹配点坐标值差异较大这种特性,对伪匹配点进行去除,最后进行目标识别。实验结果表明,在尺寸大小为800×600的图像中,SURF算法提取特征点数225个,耗时92.499 ms, Octave 3；特征点匹配率97.50% ,耗时349.716 ms。提出的改进方法更为简单有效,减少了特征点匹配的误差,能够有效缩短图像配准时间。

Abstract:

Algorithm for feature point matching is an effective algorithm to realize the target recognition. At present, algorithm for the image feature points matching is time-consuming, and there are wrong matching points in the matching process. Therefore, a kind of improved algorithm based on SURF algorithm was proposed. In process of detecting initial feature points, Gaussian Pyramid Layers was built dynamically according to the image size to improve real-time and accuracy. And initial feature points were optimized by setting the threshold, which reduced the matching time. In the stage of feature points’ matching, false matching points can be removed by the difference of coordinate value between false matching points and right matching points. Experimental results show that in the image size of 800×600, the SURF algorithm extracts 225 feature points, which takes 92.499 ms, Octave 3; the feature point matching rate is 97.50%, which takes 349.716 ms. The improved method is more simple and effective, reduces the error of feature point matching, and can effectively shorten the image registration time.