自适应邻域测试的图像误匹配点剔除算法
|
摘要
针对图像特征点暴力匹配与比率测试得到的匹配点对在数量与正确率不能兼顾的情况,提出了一种基于自适应邻域测试的误匹配点对剔除算法。对特征点进行暴力匹配与高阈值的比率测试得到初始匹配点集,对初始匹配点对中的每个匹配特征点进行自适应邻域测试,测试出初始匹配点集中明显的误匹配点对并将之剔除,达到只剔除误匹配而不会误剔除正确匹配的效果。实验结果表明,在保证正确率不降低的前提下,该算法获取的匹配点对数量比原算法多3成以上,并且该算法对图像旋转、尺度缩放具有较好通用性。
Aiming at the situation that the matching pair of image feature point brute-match and ratio test can't take into account both the quantity and the correct rate, an algorithm based on adaptive neighborhood test is proposed to eliminate the mismatched pair. Firstly, the ratio of violent matching and high threshold of feature points is tested to obtain the initial matching point set, then adaptive neighborhood testing is performed on each matching feature point in the initial matching point pair, it tests out the obvious mismatch points in the initial matching points and removes them, achieves the effect of removing only false matches without mistakenly removing the correct match. The experimental results show that the number of matching pairs obtained by this algorithm is more than 30% higher than the original algorithm, on the premise that the correct rate is not reduced, and the algorithm has good versatility for image rotation and scaling.
Aiming at the situation that the matching pair of image feature point brute-match and ratio test can't take into account both the quantity and the correct rate, an algorithm based on adaptive neighborhood test is proposed to eliminate the mismatched pair. Firstly, the ratio of violent matching and high threshold of feature points is tested to obtain the initial matching point set, then adaptive neighborhood testing is performed on each matching feature point in the initial matching point pair, it tests out the obvious mismatch points in the initial matching points and removes them, achieves the effect of removing only false matches without mistakenly removing the correct match. The experimental results show that the number of matching pairs obtained by this algorithm is more than 30% higher than the original algorithm, on the premise that the correct rate is not reduced, and the algorithm has good versatility for image rotation and scaling.
引文
[1]陈天华,王福龙.实时鲁棒的特征点匹配算法[J].中国图象图形学报,2016,21(9):1213-1220.
[2]Ke Y,Sukthankar R.PCA-SIFT:a more disti-nctive representation for local image descriptors[C]//CVPR,2004:506-513.
[3]吴伟交,王敏,黄心汉,等.基于向量夹角的SIFT特征点匹配算法[J].模式识别与人工智能,2013,26(1):123-127.
[4]戴雪梅,郎朗,陈孟元.基于改进ORB的图像特征点匹配研究[J].电子测量与仪器学报,2016,30(2):233-240.
[5]杨哲.基于单目摄像头的实时三维重建[D].杭州:浙江大学,2017.
[6]朱士伟.序列图像三维重建关键技术研究[D].南京:东南大学,2015.
[7]Lowe D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[8]Dong Y,Cui Y,Science S O,et al.An approach to the partial shoeprints image matching based on SIFT and RANSAC[J].Journal of Henan University of Engineering,2017.
[9]黄丽,李晓明.基于ORB特征的图像误匹配剔除方法研究[J]机电工程,2017,34(5):552-556.
[10]赵烨,蒋建国,洪日昌.基于RANSAC的SIFT匹配优化[J]光电工程,2014,41(8):58-65.
[11]Wang Yue,Zheng Jin,Xu Qizhi,et al.An improved RANSAC based on the scale variation homogeneity[J]Journal of Visual Communication and Image Representation,2016,40:751-764.
[12]贾雯晓,张贵仓,汪亮亮,等.基于SIFT和改进的RANSAC图像配准算法[J].计算机工程与应用,2018,54(2):203-207.
[13]夏克付,李鹏飞,陈小平.基于改进RANSAC算法的单应矩阵鲁棒估计方法[J].计算机工程与应用,2017,53(23):147-152.
[14]Wei C H,Li Y,Gwo C Y.The use of scale invariance feature transform approach to recognize and retrieve incomplete shoeprints[J].Journal of Forensic Sciences,2013,58(3):625.
[15]翁秀玲,王云峰,余小亮.一种图像立体匹配的误匹配点剔除与校正方法[J].软件,2016,37(10):20-24.
[16]Zhong S H,Liu Y,Chen Q C.Visual orientation inhomogeneity based scale-invariant featuree transform[J].Expert Systems with Applications,2015,42(13):5658-5667.
[2]Ke Y,Sukthankar R.PCA-SIFT:a more disti-nctive representation for local image descriptors[C]//CVPR,2004:506-513.
[3]吴伟交,王敏,黄心汉,等.基于向量夹角的SIFT特征点匹配算法[J].模式识别与人工智能,2013,26(1):123-127.
[4]戴雪梅,郎朗,陈孟元.基于改进ORB的图像特征点匹配研究[J].电子测量与仪器学报,2016,30(2):233-240.
[5]杨哲.基于单目摄像头的实时三维重建[D].杭州:浙江大学,2017.
[6]朱士伟.序列图像三维重建关键技术研究[D].南京:东南大学,2015.
[7]Lowe D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[8]Dong Y,Cui Y,Science S O,et al.An approach to the partial shoeprints image matching based on SIFT and RANSAC[J].Journal of Henan University of Engineering,2017.
[9]黄丽,李晓明.基于ORB特征的图像误匹配剔除方法研究[J]机电工程,2017,34(5):552-556.
[10]赵烨,蒋建国,洪日昌.基于RANSAC的SIFT匹配优化[J]光电工程,2014,41(8):58-65.
[11]Wang Yue,Zheng Jin,Xu Qizhi,et al.An improved RANSAC based on the scale variation homogeneity[J]Journal of Visual Communication and Image Representation,2016,40:751-764.
[12]贾雯晓,张贵仓,汪亮亮,等.基于SIFT和改进的RANSAC图像配准算法[J].计算机工程与应用,2018,54(2):203-207.
[13]夏克付,李鹏飞,陈小平.基于改进RANSAC算法的单应矩阵鲁棒估计方法[J].计算机工程与应用,2017,53(23):147-152.
[14]Wei C H,Li Y,Gwo C Y.The use of scale invariance feature transform approach to recognize and retrieve incomplete shoeprints[J].Journal of Forensic Sciences,2013,58(3):625.
[15]翁秀玲,王云峰,余小亮.一种图像立体匹配的误匹配点剔除与校正方法[J].软件,2016,37(10):20-24.
[16]Zhong S H,Liu Y,Chen Q C.Visual orientation inhomogeneity based scale-invariant featuree transform[J].Expert Systems with Applications,2015,42(13):5658-5667.