一种新的协同模式识别学习算法

在协同模式识别中，学习可以归结为求原型向量和伴随向量．文中提出了一种基于核函数映射的学习算法，输入向量被隐式地映射到一个可分性有所提高的向量空间，然后计算变换后的原型向量．对伴随向量增加一个附加的约束以避免它的范数超过一定值，从而改善伴随向量的性能，减少误识别．通过对数字、英文字母和汉字等的训练实验表明，这种算法得到的伴随向量能更好地表示样本的特征，计算所得的初始序参量能更好地反映测试图像与训练样本之间的相似程度．

A Novel Learning Algorithm for Synergetic Pattern Recognition

Learning in synergetic pattern recognition is a process by which prototype vectors and adjoint vectors are calculated. A new kernel-based learning algorithm was presented. Input data are implicitly mapped into a high-dimensional feature space in which the patterns are more separable. The samples are represented as adjoint vectors on which an additive constraint is imposed to avoid the norm value of them being unreasonable large. The experimental results of applying the scheme to characters demonstrate its efficiency, and the calculated order parameters used in subsequent evolvement represent more truly the similarity between the test pattern and the prototypes.