摘要
尺度空间技术由于能够很好地模拟人类的视觉机能,已经成为理解和分析图像的一种现代化工具.本文提出了一种基于L曲率的尺度空间形状分析技术.考虑了该技术在形状描述和角点检测中的应用,并具体地给出了一种检测角点的多尺度曲率积算法.L曲率的尺度空间图表明它的零点和极值点关于尺度参数是稳定的,具有较强的鲁棒性.新的角点检测算法能够增强形状特征点的信息,抑制噪声,可以得到良好的实验结果,且计算量较少.
Scale-space techniques have been considered as modern tools for image understanding and analysis because they are consistent with the concept of human beings. In this paper, a scale-space technique for shape analysis based on the L curvature is proposed, and its applications in shape representation and corner detection are also studied. Analysis of its relation to planar curvature matched very well with experimental results. So, in sometimes,L curvature takes place of real curvature. And L curvature has more advantage to real curvature. The compute of L curvature is easy and accurate. Therefore, scale space based on L curvature has more merits. Multi- scales L curvature products （MSCP） algorithms is introduced in detail. Then a new corner detector is proposed. Extract zero crossings and the local extreme points of L curvature, and draw them in two plane coordinates, form Curvature Scale-Space（CSS） maps. The CSS maps constructed with the L curvature indicate that the scale space trajectories of zero crossings and the local extreme points are stable with respect to the input parameter. Experiments are conducted which show that the new corner detector can enhance the information of shape feature and suppress noise, and therefore can achieve a good performance in corner detection. Moreover, this method does not have the undesirable effect of the Gaussian smoothing, need less calculation, have more robustness to noise. The new corner detector is robust, simple and effective.
出处
《南京大学学报（自然科学版）》
CSCD
北大核心
2012年第2期172-181,共10页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(61075040)
江苏省省属高校自然科学研究重大项目(10KJA520047)
航空科学基金(2009ZH52069)
关键词
尺度空间
形状分析
角点检测
L曲率
scale-space, shape analysis, corner detection, L-curvature
