滤光片多光谱成像中运动目标场景误配准修正  

Misregistration Correction for Moving Object Scene in Filter-type Multispectral Imaging

作　　者：章文洁[1,2] 黄旻[1] 张桂峰[1] 

机构地区：[1]中国科学院光电研究院,北京100094 [2]中国科学院大学,北京100049

出　　处：《计算机技术与发展》2016年第1期18-21,共4页Computer Technology and Development

基　　金：国家杰出青年基金资助项目(61225024)

摘　　要：图像配准是多光谱相机获得精确的多光谱数据立方体的重要步骤。文中分析了运动目标对滤光片阵列型多光谱相机所获取光谱图像序列的配准结果的影响。针对此类误配准问题,文中提出一种基于运动目标检测的误配准修正方法。该方法首先通过变化检测粗提取误配准点区域,然后利用运动目标形状特征进一步提取误配准点,利用多图像序列的空间和灰度关系对误配准点进行修正。用该方法对8通道的滤光片阵列型多光谱相机的多光谱图像进行处理,实验结果表明,该方法能有效地校正多光谱成像中运动目标所造成误配准的影响,提高多光谱图像配准的准确度。处理后的图像数据更有利于图像判读、多光谱图像分类、目标识别等的后续应用。Image registration is a key step for multispectral camera to obtain multispectral data cube. The impact of moving objects on image registration for filter array type multispectral camera was analyzed. To solve the problem,a method based on the detection of moving vehicles to correct the misregistration is presented. Firstly,the potential regions of misregistration pixels are roughly extracted by change detection,and then the misregistration pixels are screened through the features of moving objects,and these pixels are corrected based on the relationships of space and gray level among the multispectral image sequences. The method is applied to process the multispectral image obtained with an 8- channel filter array multispectral camera. The experimental result indicates that the approach could fix the wrong matches effectively and improve the accuracy of multispectral imaging registration. The processed images would be more favorable to image interpretation,the classification of multispectral image,object recognition and other subsequent applications.

关 键 词：多光谱图像 配准 运动目标 目标检测 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]