摘要
针对图像同时存在仿射变形和成像特性差异时传统方法难以快速可靠匹配图像的问题,提出仿射变形下的异源图像匹配方法.首先利用二维Moran信息准则从小图中筛选若干有效子区;然后基于梯度方向相关方法在平移空间内计算每个有效子区与大图的相似度;最后对多个子区的平移空间相似度进行变换与组合得到仿射空间图像相似度的近似描述,求解仿射空间相似度最大值得到图像匹配结果.该方法避免了多次对全图进行仿射变换和比较相似度的运算,其计算复杂度仅与传统平移空间图像匹配方法相当,远低于仿射空间图像匹配方法.实验结果表明,文中方法不仅计算速度快,同时还具有很高的匹配成功率.
A rapid and robust approach is proposed to match multi-sensor images in affine space. Firstly, valid sub-regions are automatically selected from the small image based on 2-dimensional Moran information. Then, similarity map between each valid sub-region and the large image is calculated by gradient orientation correlation in translation space. The translation similarity maps are transformed and combined to appropriate the affine similarity. Finally, matching result is obtained by finding the maximum value of combined map in affine space. The proposed approach avoids affine transformation and image comparison which always execute many times in traditional affine matching approach, and has a comparatively low computing complexity as traditional translation matching method. Experiments show that our approach not only runs fast, but also has higher matching rate than other methods.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2015年第8期1512-1517,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家"九七三"重点基础研究发展计划项目(2013CB733100)
国家自然科学基金(61402489)
关键词
图像匹配
异源图像
仿射变换
子区相似度
二维Moran信息
梯度方向相关
image matching
multi-sensor image
affine transformation
sub-region similarity
2-dimensional moran information
orientation correlation
作者简介
涂国勇(1970-),男,硕士,高级工程师,主要研究方向为飞行器试验评估;
(lizhuang2007@hotmail.com)李壮(1982-),男,博士.工程师,主要研究方向为图像匹配与目标识别;
李伟建(1979-),男,博士,工程师,主要研究方向为飞行器制导与控制;
朱宪伟(1979-),男,博士,讲师,主要研究方向为图像特征提取与匹配.