Canny算法和中值滤波法的红外全景图像拼接  

Canny algorithm and median filter for infrared panoramic image mosaic

作　　者：车畅 兰文宝 CHE Chang;LAN Wenbao(Harbin University,School of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨学院工学院,哈尔滨150001

出　　处：《激光杂志》2020年第5期109-113,共5页Laser Journal

基　　金：哈尔滨学院青年博士科研项目(No.HUDF2017207);黑龙江省教育科学“十三五”规划备案课题(No.GJC1319058)。

摘　　要：因传统拼接手段不能获取有效目标向量,导致拼接后图像模糊、不完整,并且存在图像匹配准确率不高、匹配失误率过高和耗时较长的问题,因此,提出Canny算法和中值滤波法的红外全景图像拼接方法。运用Canny算法提取具有显著性特征的图像边缘向量,同时采用中值滤波法,将图像差异过大的数据做对应的过滤处理。为了降低拼接图像结构层次混乱、不齐的情况,根据过滤结果对待拼接的图像做特征定位处理,并实现目标特征匹配。最终在线性融合的基础上提出基于聚类预筛选的多分辨率融合方法,加强图像整体融合效果,从而得出拼接后的完整红外全景图像。分析实验结果可知,所提方法的匹配失误率最低值仅为8%,说明所提出方法可以在根本上降低图像匹配的失误率,增强匹配的精准度,使拼接后的图像更加清晰,并且耗时较短,充分说明该方法应用价值高。Because the traditional splicing method can't get the effective target vector,resulting in the blurred and incomplete image after splicing,and there are some problems,such as the low accuracy of image matching,the high error rate of image matching and the long time consuming,so the Canny algorithm and the median filter method are proposed for infrared panoramic image splicing.Canny algorithm is used to extract the image edge vector with significant characteristics,and median filter method is used to filter the data with large difference.In order to reduce the disorder and irregularity of the structure of the mosaic image,according to the filtering results,the mosaic image is processed for feature location,and the target feature matching is realized.Finally,on the basis of linear fusion,a multiresolution fusion method based on clustering pre screening is proposed to enhance the overall fusion effect of the image,so as to obtain the complete infrared panoramic image after splicing.Analysis of the experimental results shows that the minimum matching error rate of this method is only 8%,which shows that the proposed method can fundamentally reduce the error rate of image matching,enhance the accuracy of matching,make the stitched image clearer,and take less time,which fully shows that the method has high application value.

关 键 词：显著特征 图像拼接 计算机视觉 聚类预筛选 

分 类 号：TN911.4[电子电信—通信与信息系统]