基于改进SIFT算法的胃部图像拼接方法研究  

Research on Stomach Image Stitching Method Based on Improved SIFT Algorithm

作　　者：夏振港 华德正 张旭 王勇[2] 刘晓帆 刘新华[1] XIA Zhengang;HUA Dezheng;ZHANG Xu;WANG Yong;LIU Xiaofan;LIU Xinhua(School of Mechatronic Engineering,China University of Mining and Technology,Xuzhou 221116,China;School of Information Engineering,Xuzhou College of Industrial Technical,Xuzhou 221140,China;China Coal Education Association,Beijing 100713,China)

机构地区：[1]中国矿业大学机电工程学院,江苏徐州221116 [2]徐州工业职业技术学院信息工程学院,江苏徐州221140 [3]中国煤炭教育协会,北京100713

出　　处：《自动化仪表》2022年第9期56-61,共6页Process Automation Instrumentation

基　　金：国家自然科学基金资助项目(51975568)、江苏省自然科学基金资助项目(BK20191341)。

摘　　要：针对医疗胶囊机器人采集的图片倾角和旋角变化差异大、视野范围窄等问题,提出了一种基于改进尺度不变特征变换(SIFT)算法的胃部图像拼接方法。首先,设计了算法的总体框架,研究了基于双边滤波算法的胃部图像相似性特征增强方法。其次,改进了基于改进图像金字塔的尺度空间构建方法,提出了一种图像特征点描述符生成的方法,降低了描述符生成的计算量,完成了胃部图像的配准和融合操作。最后,对拼接图像进行了客观指标评价,并进行了仿真试验。试验结果表明:改进后的算法配准精度提高了8.88%,运行效率提高了16.8%,且改进后的算法图像拼接质量优于原算法的拼接质量。The method of stomach image stitching based on the improved scale-invariant feature transform(SIFT) algorithm is proposed for the problems of large differences in tilt and rotation angle variations and narrow field of view of images captured by the medical capsule robot. Firstly, the general framework of the algorithm is designed, and the similarity feature enhancement method of stomach images based on bilateral filtering algorithm is studied. Secondly, the scale space construction method based on the improved image pyramid is improved, and a method of image feature point descriptor generation is proposed, which reduces the computational effort of descriptor generation and completes the alignment and fusion operation of the stomach image. Finally, the stitched images are evaluated by objective indexes and simulation tests are conducted. The test results showed that the improved algorithm improved the alignment accuracy by 8.88% and the operation efficiency by 16.8%, and the image stitching quality of the improved algorithm is better than the stitching quality of the original algorithm.

关 键 词：胶囊机器人 胃部图像拼接 改进尺度不变特征变换算法 双边滤波 尺度空间 特征点匹配 描述符生成 图像配准 

分 类 号：TH-39[机械工程]