基于改进SIFT算法的超声图像拼接方法  

An ultrasonic image mosaic method based on improved SIFT algorithm

作　　者：迟大钊[1] 徐智贤 刘海春 李庆生 郭强 苏维刚 贾涛 CHI Dazhao;XU Zhixian;LIU Haichun;LI Qingsheng;GUO Qiang;SU Weigang;JIA Tao(National Key Laboratory of Precision Welding and Joining of Materials and Structures,Harbin Institute of Technology,Harbin,150001,China;PipeChina Engineering Quality Supervision and Inspection Company,Beijing,100013,China)

机构地区：[1]哈尔滨工业大学,材料结构精密焊接与连接全国重点实验室,哈尔滨150001 [2]国家管网集团工程质量监督检验有限公司,北京100013

出　　处：《焊接学报》2024年第10期1-7,共7页Transactions of The China Welding Institution

基　　金：国家自然科学基金资助项目(52375328)。

摘　　要：针对大型结构超声C扫描获得的多幅子图像,为了获取被测结构的全景图像,研究图像的拼接方法.根据超声波法成像检测的动态过程并结合数字化图像处理技术,在传统的基于尺度不变特征变换(scale invariant feature transform,SIFT)算法基础上,提出了一种改进的超声C扫描检测图像拼接方法.首先,针对传统SIFT算法存在超声图像配准成功率低的问题,利用超声扫查过程中探头起始位置的矢量差对SIFT算法得到的特征点粗匹配结果进行筛选;其次,采用动态规划的方法寻找最佳缝合路径;最后,沿最佳缝合路径进行渐入渐出融合以改善融合区域的视觉效果.人工缺陷试块及焊接结构试件超声图像的拼接结果表明,基于改进的SIFT图像拼接算法可以将多幅超声C扫描子图像有效拼接为全景图像;所提方法特征点匹配准确率高、图像融合失真小,优于常规SIFT图像拼接算法;拼接图像中,目标位置关系对应良好,可实现结构加工质量的整体无损评价.A comprehensive non-destructive testing of large structures usually needs a series of C-scans.In order to obtain a panoramic image of the structure under test,the method of sub-image mosaic is studied.According to the dynamic process of ultrasonic imaging and combined with digital image processing technology,an improved image mosaic method for ultrasonic Cscan detection is proposed based on the traditional scale invariant feature transform(SIFT)algorithm.Firstly,in view of the low success rate of ultrasound image registration using the traditional SIFT algorithm,the obtained matching feature points are screened through the vector difference of the starting positions of ultrasonic probe.Secondly,a dynamic programming method is used to find the best stitching path.Finally,a gradual in and out fusion is carried out along the best path for stitching to improve the visual effect of the fused area.Artificial defect contained block and welded piece are prepared and tested.The results of ultrasonic image mosaic show that the improved SIFT algorithm can effectively stitch multiple ultrasonic C-scan sub-images into panoramic images,and the proposed method has high accuracy of feature point matching and small image fusion distortion,which is better than the conventional SIFT image mosaic algorithm.In the mosaic image,the positions of targets match well,which can achieve overall non-destructive evaluation of structural processing quality.

关 键 词：无损检测 超声成像 图像拼接 SIFT算法 

分 类 号：TG409[金属学及工艺—焊接]