摘要
针对激光三维成像传感器与可见光传感器图像分辨率差异较大,配准过程中特征点误匹配情况严重的问题,提出了一种基于改进SURF算子的高低分辨率图像配准方法。首先,采用双线性插值算法对低分辨率图像进行预处理,然后在经典SURF算子的基础上,采用最近邻向量匹配法完成SURF特征的粗匹配,并基于特征偏移一致性原则对匹配情况做进一步优化,最后结合RANSAC和最小二乘法求出图像之间的仿射关系,利用所求的变换参数插值得到配准后的图像。实验结果表明,该配准方法在保持配准速度的同时,结构相似性测量指数提高了约11%,进一步提高了配准的精度。
Due to the large resolution difference between the three-dimensional laser imaging sensor and visible ima-ging sensor,mismatching feature points are numerous.In order to solve this problem,a registration method of high-lowresolution images based on improved SURF is proposed.Firstly,low-resolution image is processed through bilinear in-terpolation.Subsequently,SURF feature′s coarse matching is completed by using the nearest neighbor vector based onthe classical SURF.The matching is further optimized according to feature shift coherence rule.Finally,affine relation-ship between the images is obtained by using SANSAC and least square method.The registration image is obtained byinterpolation based on affine transform parameters.The experiment shows that the proposed registration method cankeep the registration speed,structural similarity measure index is increased by about 11%,and meanwhile the accura-cy of registration is further improved.
出处
《激光与红外》
CAS
CSCD
北大核心
2014年第2期207-212,共6页
Laser & Infrared
基金
国家高技术研究发展计划资助项目
关键词
图像处理
高低分辨率配准
特征偏移一致性
SURF
image process
high-low resolution images registration
SURF
feature shift coherence
