基于几何不变性和BP网络的二维目标识别算法

提出了一种基于几何不变性和BP网络的二维目标识别算法.该算法不仅能适应目标物体在旋转、缩放和平移变换(RST变换)下的不变性识别,而且能适应仿射及射影变换下的不变性识别.算法通过对目标物体边缘点进行规格化和对规格化后的边缘点进行5点不变量穷举计算解决了模型图像与目标图像的对应点选取问题;通过将不同观测方位和不同旋转角度的样本图像边缘点的5点不变量集合作为输入向量对BP网络进行训练解决了由于仿射和射影变换造成规格化边缘点间距变化对正确分类的影响.算法不仅能识别多边形而且能识别曲线轮廓的目标物体. 

2D Object Recognition Based on Geometry Invariant and BP Network

A 2D object recognition algorithm based on geometry invariant and BP Network. is proposed. It can be applied to object recognition under rotation scaling translation (RST) and projection transform. By normalizing the number of pixel in the edge of image of the object to be recognized and computing the 5 point invariant in exhausting way the problem of selecting correspond points between the training and the testing object is solved, The influence to the correct recognition of the object is removed by using 5 point invariant from different view and rotation of the object image as the training set of the BP network. These makes the algorithm can be used for the recognition of the object with curve edge.