视差约束的改进Hough变换多光谱影像匹配方法  

Matching method for multispectral images based on improved hough transformation with disparity constraint

作　　者：赵宝玮[1,2] 相里斌[1,2] 黄旻[2] 吕群波[2] 张桂峰[2] 

机构地区：[1]西安电子科技大学物理与光电工程学院,陕西西安710071 [2]中国科学院光电研究院,北京100094

出　　处：《西安电子科技大学学报》2015年第6期94-98,共5页Journal of Xidian University

基　　金：国家杰出青年科学基金资助项目(61225024)

摘　　要：针对机载滤光片型多光谱相机影像配准中同一目标在不同谱段表现的特征各不相同以及不同谱段图像存在较大灰度差异等问题,提出了视差约束的改进Hough变换多光谱影像匹配方法.该方法使用金子塔影像匹配策略,顶层金字塔影像通过尺度不变特征转换算子进行匹配来提供初始的视差约束条件,其他各层金字塔左右影像利用改进Hough变换影像方法进行匹配,匹配过程中利用上层匹配结果为当前层影像匹配提供视差约束条件,提高匹配速度及可靠性.通过对两幅1 415×1 415像素的机载多光谱飞行影像数据进行匹配,该方法仅耗时7.8s,配准精度能够达到0.65个像素.To overcome the problems of image matching caused by different charactors and the grey-level in every spectrum of the airborne filter multispectral imager for the same target, a matching method for multispectral images based on improved Hough transformation with disparity constraint is presented. Firstly, we get the N layer pyramid images of the two multispectral images for registration by down- sampling, and then calculate the transformation parameters of each layer between the left and right images progressively. In this progress, the transformation parameters of the two adjacent top Pyramid images can be calculated by using the feature points extracted by the SIFT operator, and the transformation parameters of the other adjacent Pyramid images can be calculated by using the transformation parameters of their upper images and the improved Hough transformation. At the same time, the attitude parameters and the geometric transform relation calculated from the upper layer between the images can be used to improve the matching speed and reliability. By processing two airborne flight multispectral images with 1 415× 1415 pixels, this approach only takes 7.8 secends, and its registration accuracy can reach 0.65 pixels.

关 键 词：金字塔方法 多光谱 改进的Hough变换 递进影像匹配 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]