基于融合特征的SURF配准算法  

A SURF image registration algorithm based on fusion features

作　　者：秦煜 吴静静 安伟 张美娟[3] 

机构地区：[1]江南大学机械工程学院,江苏无锡214122 [2]江南大学江苏省食品先进制造装备技术重点实验室,江苏无锡214122 [3]无锡职业技术学院,江苏无锡214122

出　　处：《计算机工程与科学》2017年第10期1890-1895,共6页Computer Engineering & Science

基　　金：国家自然科学基金(61305016)

摘　　要：SURF算法是一种尺度不变、旋转不变且鲁棒性良好的配准算法,但其丢失了图像的颜色特征,因此对于彩色图像的配准效果不佳。为此,提出了一种基于融合特征的SURF配准算法。该算法首先利用彩色图像的颜色不变量和DLBP纹理特征构造融合特征灰度图,并提出了一种基于彩色图像颜色直方图的自适应方法来调节融合特征的权重;然后,利用SURF算法在融合特征灰度图上进行特征点的提取与匹配;最后,使用改进的RANSAC算法去除误匹配点。实验结果表明,对于彩色图像,此算法有效地增加了提取的特征点数,并加快了配准速率。The speed-up robust features （SURF ） algorithm is a scale-invariant, rotation-invariant and robust registration algorithm. However,since the color features of the image can be lost,its regis-tration effect of color images is poor. W e therefore propose a SURF registration algorithm based on fu-sion features. Firstly,w e utilize the color invariant and the double local binary patterfeatures of the color image to construct the grayscale of fusion features. Meanwhile,w e propose an a- daptivemethod to adjust theweights of fusion features based on the color histogramof Then,the SURF algorithm is used to extract and match the feature points on the grayscale of fusion fea-tures. Finally,the improved random sample consensus （RANSAC ） algorithm is used match points. Experimental results show that the proposed algorithm can effectively increase the featurepoints and speed up the registration rate for color images.

关 键 词：SURF算法 颜色不变量 DLBP纹理特征 RANSAC算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]