摘要
提出了用于目标物识别和分类的极半径不变矩 .对目标物进行分割后 ,先求出目标物的形心 ,进而求出极半径矩、归一化矩和归一化的中心矩 .在此基础上 ,给出了 5个具有平移、旋转和尺度变换不变性的特征量用于物体形状的识别 .文中给出了这些不变矩的特性 ,并给出了极半径不变矩和边界序列矩以及Hu提出的不变矩的实验比较结果 .该文提出的极半径不变矩 ,既可用于区域目标的识别 ,也可用于边界形状的识别 .
A novel moment, called polar-radius-invariant-moment, is proposed for the object recognition and classification. Following the process of segmentation, we get binaried image. The centroid denoted by (x c ,y c ) of the object is calculated firstly. Then the polar radius r, that is, the distance from an arbitrary point of the object to the centroid,is computed.We define the p - order polar radius moment m p and central moment m c p as m p = ∫∫ Dr p ds, m c p = ∫∫ D(r- r- ) p ds, respectively. Normalized moment and normalized central moment are defined as m np =1A ∫∫ Dr r- p ds, m ncp = 1A ∫∫ Dr- r- r- p ds, respectively, where A is the area of the object and r- =1A ∫∫ Drds is the mean of r. The shifting, rotation and scaling invariance of the normalized moment m np and normalized central moment m ncp are proved theoretically. After that five derived invariant moments are employed as features in the recognition of objects. Examples are presented to illustrate the performance of these moments. In our experiments, we make use of the random movement of the pixels of the object to simulate the noise disturbance, and utilize the minimum distance rule to classify the shapes. In the comparing experiment of recognition of plane models, polar-radius-invariant-moments give a recognition rate of 100%, while that of contour sequence moments is 86% (p=0.3). When classifying a group of symmetric shapes, our approach arrives a rate of 91.4% whereas Hu’s moments reach a rate of 81.6%. These moments can be used for both the boundary shape recognition and the interior region shape recognition.
出处
《计算机学报》
EI
CSCD
北大核心
2004年第6期860-864,共5页
Chinese Journal of Computers
基金
国家自然科学基金 ( 60 2 72 0 94)
山东省自然科学基金(Y2 0 0 1G10 )资助
