摘要
现有的多尺度曲率乘积算法MSCP(Multi Scale Curvature Product)在检测过程中存在漏检和误检的现象,并且由于阈值设置不合理容易导致算法不稳定。针对以上不足,提出一种基于曲率多尺度的自适应角点检测算法AMCP(Adaptive Multi-scale Curvature Polynomial)。首先,结合尺度多项式的方法,不仅显著增强了曲率极值点的峰值,而且避免了曲率积对一些角点平滑;然后,构造局部曲率显著度LCCS(Local Corners Curvature Saliency),从而用自适应的阈值代替全局阈值,增强算法应对尺度,旋转等变化的鲁棒性;最后,提出曲率增长度的方法,通过该方法有效地区分圆角点和钝形角点。通过实验表明,AMCP算法提高了角点检测的正确率以及稳定性,相较于MSCP算法以及改进的CSS(Corner Detection Through Curvature Scale Space)算法具有更加优越的检测性能。
Current multi-scale curvature product （MSCP） algorithm has the phenomenon of missed detection and false detection in detecting process, moreover, the improper threshold setting may lead to the instability of the algorithm. In light of the above deficiencies, we proposed an adaptive muhi-scale-based corner detection algorithm （AMCP）. First, we combined the multi-scale polynomial method, this not only significantly enhanced the peak of curvature extreme points ,but also prevented the smoothing on some comers by curvature product. Then we constructed the local corner＇s curvature saliency （LCCS） so that replaced the global threshold with adaptive threshold and thus enhanced the robustness of the algorithm when dealing with scale and rotation changes. Finally, we presented the curvature growth degree method, and effectively distinguished the round corner and obtuse corner with the method. It was demonstrated through experiment that the AMCP algorithm improved the accuracy and stability of corner detection, and had better detection performance than MSCP and the improved CSS algorithm.
出处
《计算机应用与软件》
CSCD
2016年第1期185-189,共5页
Computer Applications and Software
基金
国家自然科学基金项目(61272195)
重庆市自然科学基金项目(cstc2012jj A1699)
