基于改进SIFT算法的目标识别
|
摘要
针对目标识别过程中识别精度不高、实时性不好的问题,提出基于尺度不变特征转换(SIFT)算法的改进算法,该算法通过研究传统的SIFT算法特征匹配正确率不高、匹配耗时过长的问题,结合Harris算子角点检测特性提出改进,在高斯差分尺度空间内直接检测角点,使得提取的特征点数目减少,计算量降低,特征点提取的显著性提高;同时使用RANSANC方法进行特征匹配约束,减少误匹配,进一步提升目标识别的正确率。为了验证提出算法的有效性,通过MATLAB对算法在尺度变化和噪声等复杂情况下的匹配效果进行实验验证,结果表明,改进的SIFT算法匹配用时大大降低、误匹配较少,匹配正确率提高,具有较强的鲁棒性,可以准确识别目标,具有良好的目标识别能力。
For the problem of low recognition accuracy and poor real-time performance in the process of target recognition, this paper proposes an improved algorithm based on the scale invariant feature transform(SIFT) algorithm.There're several problems in traditional SIFT algorithm, such as low matching accuracy, too much time in matching.Therefore, an improved SIFT algorithm is proposed to solve these problems, by combining the corner detection characteristics of Harris operator. It directly detects the corner points in the Gaussian differential scale space, which reduces the number of extracted feature points and the amount of calculation, improves the significance of feature point extraction; Simultaneously, it uses the RANSANC method for feature matching constraints to reduce false matching and further improve the accuracy of target recognition. In order to verify validity of the proposed algorithm, the matching effect of the algorithm under complex conditions such as scale change and noise is experimentally verified by MATLAB. The results show that the improved SIFT algorithm can greatly reduce the matching time, have less false matching, and increase the matching accuracy, having strong robustness, so it can accurately identify targets and have good target recognition capability.
For the problem of low recognition accuracy and poor real-time performance in the process of target recognition, this paper proposes an improved algorithm based on the scale invariant feature transform(SIFT) algorithm.There're several problems in traditional SIFT algorithm, such as low matching accuracy, too much time in matching.Therefore, an improved SIFT algorithm is proposed to solve these problems, by combining the corner detection characteristics of Harris operator. It directly detects the corner points in the Gaussian differential scale space, which reduces the number of extracted feature points and the amount of calculation, improves the significance of feature point extraction; Simultaneously, it uses the RANSANC method for feature matching constraints to reduce false matching and further improve the accuracy of target recognition. In order to verify validity of the proposed algorithm, the matching effect of the algorithm under complex conditions such as scale change and noise is experimentally verified by MATLAB. The results show that the improved SIFT algorithm can greatly reduce the matching time, have less false matching, and increase the matching accuracy, having strong robustness, so it can accurately identify targets and have good target recognition capability.
引文
[1]常德宽,雍学善,杨午阳,等.计算机视觉成像方法在地震勘探中应用研究与探索[C]//中国石油学会2017年物探技术研讨会论文集.天津,中国,2017:4.
[2]完文韬,杨成禹.改进的SIFT算法在图像特征点匹配中的应用[J].长春理工大学学报:自然科学版,2018,41(1):44-47,52.
[3]贾平,徐宁,张叶.基于局部特征提取的目标自动识别[J].光学精密工程,2013,21(7):1898-1905.
[4]白廷柱,侯喜报.基于SIFT算子的图像匹配算法研究[J].北京理工大学学报,2013,33(6):622-627.
[5]曾峦,王元钦,谭久彬.改进的SIFT特征提取和匹配算法[J].光学精密工程,2011,19(6):1391-1397.
[6]章菲菲.基于改进SIFT算法的目标识别与跟踪技术研究[D].北京:北京理工大学,2015.
[7]王德海.基于双目立体视觉的目标识别与抓取定位[D].长春:吉林大学,2016.
[8]汪淑梦.基于改进的SIFT算法的图像配准技术的研究与实现[D].北京:中国地质大学(北京),2013.
[9]HARRIS C,STEPHENS M.A combined corner and edge detector[C]//Proceedings of the 4th Alvey Vision Conference.Manchester,Britain,1988:147-151.
[10]张海燕,李元媛,储晨昀.基于图像分块的多尺度Harris角点检测方法[J].计算机应用,2011,31(2):356-357.
[11]DONG Yinwen,WAN Luan,SHI Zhaoming,et al.An image registration algorithm based on improved sift feature[J].Applied mechanics and materials,2013,347-350:3411-3415.
[12]崔哲.基于SIFT算法的图像特征点提取与匹配[D].成都:电子科技大学,2016.
[13]宰小涛,赵宇明.基于SIFT特征描述子的立体匹配算法[J].微计算机信息,2007(24):285-287.
[14]罗钟铉,刘成明.灰度图像匹配的快速算法[J].计算机辅助设计与图形学学报,2005,17(5):966-970.
[15]孔祥思.改进的RANSAC立体匹配算法的研究[J].北京建筑大学学报,2017,33(4):39-44.
[16]刘正东.计算机视觉中立体匹配技术的研究[D].南京:南京理工大学,2005.
[2]完文韬,杨成禹.改进的SIFT算法在图像特征点匹配中的应用[J].长春理工大学学报:自然科学版,2018,41(1):44-47,52.
[3]贾平,徐宁,张叶.基于局部特征提取的目标自动识别[J].光学精密工程,2013,21(7):1898-1905.
[4]白廷柱,侯喜报.基于SIFT算子的图像匹配算法研究[J].北京理工大学学报,2013,33(6):622-627.
[5]曾峦,王元钦,谭久彬.改进的SIFT特征提取和匹配算法[J].光学精密工程,2011,19(6):1391-1397.
[6]章菲菲.基于改进SIFT算法的目标识别与跟踪技术研究[D].北京:北京理工大学,2015.
[7]王德海.基于双目立体视觉的目标识别与抓取定位[D].长春:吉林大学,2016.
[8]汪淑梦.基于改进的SIFT算法的图像配准技术的研究与实现[D].北京:中国地质大学(北京),2013.
[9]HARRIS C,STEPHENS M.A combined corner and edge detector[C]//Proceedings of the 4th Alvey Vision Conference.Manchester,Britain,1988:147-151.
[10]张海燕,李元媛,储晨昀.基于图像分块的多尺度Harris角点检测方法[J].计算机应用,2011,31(2):356-357.
[11]DONG Yinwen,WAN Luan,SHI Zhaoming,et al.An image registration algorithm based on improved sift feature[J].Applied mechanics and materials,2013,347-350:3411-3415.
[12]崔哲.基于SIFT算法的图像特征点提取与匹配[D].成都:电子科技大学,2016.
[13]宰小涛,赵宇明.基于SIFT特征描述子的立体匹配算法[J].微计算机信息,2007(24):285-287.
[14]罗钟铉,刘成明.灰度图像匹配的快速算法[J].计算机辅助设计与图形学学报,2005,17(5):966-970.
[15]孔祥思.改进的RANSAC立体匹配算法的研究[J].北京建筑大学学报,2017,33(4):39-44.
[16]刘正东.计算机视觉中立体匹配技术的研究[D].南京:南京理工大学,2005.