摘要
经典的SURF(speeded up robust features)算法在匹配的过程中,对一幅图像上每一个选定的特征点都要与另一幅图像上所有的特征点一一进行匹配,耗时较高,并且由于误匹配导致匹配的准确率下降.基于此,结合特征点的分类思想对SURF算法进行改进.根据特征点邻域内像素之间的差值形成一个4维的特征向量,与SURF的特征描述子相结合形成68维的特征向量,以达到提高匹配速度和准确率的目的.在哥伦比亚大学Coil-100图像库中对改进SURF算法进行试验.结果表明,相对于经典SURF算法,改进SURF算法在速度上有很大的提高.
Classic SURF(speeded up robust features)algorithm needs to match feature points with another image's feature points one by one which wastes lots of time and leads to low match accuracy.This paper proposes a method of feature points classification.Firstly,a four-dimensional feature vector is formed based on the difference of pixels in feature point neighborhood.Secondly,the final vector is obtained by combining the four-dimensional vector with the SURF descriptor.Our experiments are done in Coil-100 image database.The experimental results prove that the method has great improvement in speed.
出处
《江苏师范大学学报(自然科学版)》
CAS
2014年第3期41-46,共6页
Journal of Jiangsu Normal University:Natural Science Edition
基金
山东大学自主创新基金资助项目(2012ZD039)