改进的SURF特征快速匹配算法  

Improved SURF feature fast matching algorithm

作　　者：张文卿 李为相 李为 陈鑫 倪再玥 ZHANG Wen-qing;LI Wei-xiang;LI Wei;CHEN Xin;NI Zai-yue(College of Electrical Engineering and Control Science,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学电气工程与控制科学学院

出　　处：《计算机工程与设计》2019年第12期3526-3532,共7页Computer Engineering and Design

基　　金：江苏省“六大人才高峰”基金项目(XXR-012)

摘　　要：针对传统的SURF匹配算法计算复杂度高、配准精度低的问题,提出一种基于Radon变换降维SURF特征的彩色图像配准算法。改进传统的SURF描述符,利用投影变换对SURF描述符降维,并叠加图像的色彩信息;在特征匹配过程中将粗匹配和精匹配相结合,剔除误匹配点;采用加权平均法实现图像融合。实验结果表明,改进算法减少了SURF算法中的重复搜索,改善了彩色信息特征的缺失,提高图像配准精度的同时,减少算法运行时间,具有较好的图像拼接效果。Aiming at the problem of high computational complexity and low registration accuracy in the traditional fast robust feature transform(SURF)matching algorithm,a color image registration algorithm based on Radon transform dimension reduction SURF feature was proposed.The traditional SURF descriptor was improved,the SURF descriptor was reduced by the projection transformation,and the color information of the image was added.The coarse matching and the fine matching were combined to eliminate the mismatched points.The image fusion was implemented using the weighted average method.Experimental results show that the improved algorithm reduces the repeated search in the SURF algorithm,improves color information features,and improves the image registration accuracy while reduces the running time of the algorithm.It shows better image mosaic effects.

关 键 词：图像匹配 加速鲁棒特征 RADON变换 色彩模式分量 误匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]