结合特征和时空关联的稳健图像匹配方法  

THE ROBUST IMAGE REGIRSTRATION METHOD BY COMBINING FEATURE AND SPATIAL-TEMPORAL CORRELATION

作　　者：邓志鹏[1] 杨杰[1] 全勇[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200030

出　　处：《红外与毫米波学报》2003年第6期447-450,共4页Journal of Infrared and Millimeter Waves

摘　　要：提出了基于特征和时空关联的积相关图像匹配算法 .首先 ,采用Bubble小波边缘检测方法提取实时图与基准图特征 ;其次 ,利用惯导位置信息平移基准图 ,实现序列实时图在其对应基准图中的相对位置相同 ;最后 ,基于匹配概率与匹配面积成正比原理叠加序列积相关矩阵 ,并确定匹配位置 .该方法克服了因面积增大由几何失真导致的匹配概率下降的缺点 ,使匹配概率已精度得到提高 ,具有较好的匹配稳健性 .A product correlation algorithm of image registration based on feature and spatial-temporal correlation was presented. At first edge features of real-time images and base image are extracted by using Bubble wavelet. Then, the base image is shifted according to position guided by inertia system, in order to make the position of real-time image in base image to be same. Finally, owing to the direct ratio between probability of registration and area of registration, the sequential product correlation matrix is added I and the registration position is searched in the added matrix. This method has overcome the shortcoming that the probability of registration decreasing with the increasing of the area owing to geometric distortion, and has improved the probability, precision and robustness of registration.

关 键 词：景象匹配 特征提取 时空关联 小波变换 匹配概率 匹配面积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] O241[自动化与计算机技术—计算机科学与技术]