摘要
针对传统螺纹图像匹配方法误匹配率高、难以实现图像拼接的问题,提出了基于尺度不变特征变换(SIFT)改进的匹配算法。先将采集图片进行枕形失真校正,在此基础上构建图像尺度空间,并在标定重叠区域内搜索高斯差分(DoG)金字塔的SIFT特征点。利用快速最近邻逼近搜索函数库(FLANN)匹配特征点,结合坐标比较和随机抽样一致性算法(RANSAC)进一步剔除误匹配,最终匹配正确率达到99%以上。实验结果表明:基于标定区域内的特征提取及匹配约束条件可提高匹配速度和精度。相比传统匹配方法,本文匹配方法对相似性较高的螺纹图像匹配具有鲁棒性和优越性,适用于螺纹桶内壁图像的全景拼接。
To solve the problem that traditional thread image matching method was difficult to realize image stitching owing to high mismatching rage,an improved matching algorithm based on scale-invariant feature transform( SIFT) was proposed. Firstly,the pincushion distortion of the collected images was corrected. Then,image scale space was built and SIFT feature points between the difference of Gaussian( Do G) pyramid based on calibration overlapping area were searched. Lastly,fast library for approximate nearest neighbors( FLANN)was used to match feature points. And error matching was eliminated by coordinate comparison and random sample consensus algorithm( RANSAC). The final matching accuracy reached to 99%. The experimental results show that the matching speed and accuracy are improved by feature extraction in calibration region and matching constraints. Compared with traditional matching method,the new method has robustness and superiority to the thread images with high similarity,which can be applied to the image stitching of the inner wall of thread-barrel.
出处
《河南科技大学学报（自然科学版）》
CAS
北大核心
2017年第5期37-42,共6页
Journal of Henan University of Science And Technology:Natural Science
基金
国家自然科学基金项目(61374050
51477125)
江苏省科技研究与发展计划基金项目(BE2016155)
