Research on Dimensional Reduction for SURF Algorithm
SURF（Speed-up robust features）算法进行图像特征点匹配时需要循环遍历待匹配图像所有特征点,计算特征点之间的SURF64描述距离,耗时大。本文对SURF算法进行了16维与4维的降维研究。实验结果表明,16维SURF算法性能与64维SURF算法基本相当,但能大幅度降低运算时间;4维运算性能降低较大,不能用于特征点匹配,但4维SUFR描述算法可以扩展到图像的各个像素点,用于ICP算法及图像的稠密匹配。Matching image feature point by SURF（Speed-up robust features） algorithm needs to loop through all feature points on the image to be matched, and compute the sixty-four dimensional distance of SURF between fea-ture points, which will take a long computation time. The sixteen dimensional SURF and four SURF are imple-mented. The experimental results show that the performance of the sixteen dimensional SURF is almost the same as that of the sixty-four dimensional SURF, moreover the sixteen dimensional SURF takes less time than sixty-four dimensional SURF. The performance of four dimensional SURF is much inferior to that of sixty-four dimensional SURF, so it cannot used to match feature point. However the four dimensional description of SURF can be ex-panded to all feature points, which can be applied in ICP algorithm and dense matching algorithm.