用组合不变量进行模式识别的一种新方法

用与目标的位置、大小、方向和其他变化无关的特征来识别目标是模式识别领域的一个热点。现存的基于不变特征的二维模式识别方法在目标被模糊了情况下都无法精确识别。本文提出了一种可解决上述问题的新的模式识别方法。该方法用组合不变量作为图像特征，以加权规格化互相关作为分类技术。在分类过程中，使用每一类的所有原型的第k个特征的类内标准方差的均值作为加权因子以提高识别率。对头像的数字试验证实了组合不变量特征对图像的平移、伸缩、旋转和模糊变换的不变性和该模式识别方法的可行性。

Pattern Recognition by Combined Invariants

A feature based recognition of objects or patterns independent of their position, size, orientation and other variations is the goal of much recent research. The existing approaches to invariant two dimensional pattern recognition are useless when the pattern is blurred. A novel pattern recognition method is presented, which can solve the problem by using combined invariants as image features. The classification technique chosen for the system is the weighted and normalized cross correlation. The mean of the intraclass standard deviations of the k th feature over the total number of prototypes for each class is used as a weighting factor during the classification process to improve recognition accuracy. The feasibility of the pattern recognition system and the invariance of the combined features with respect to translation, scaling, rotation and blurring are proved by numerical experiments on head images.