摘要
利用计算机进行模式识别需要完成模式特征的选取、特征维数的压缩以及分类器的设计。本文在人脸识别的研究中,根据选取的代数特征,提出了一种基于正交小波变换的多分类器融合分类系统。首先利用正交小波变换将高维特征变换为多个低维的特征,达到特征维数压缩的目的;然后采用基于模糊的BP神经网络(FB-PNN)并行地对这些特征空间的模式进行分类;最后,利用FBPNN对这些分类结果进行融合,得到最终的分类结果。实验结果表明这种分类系统具有很好的分类效果。
In pattern recognition, feature extraction, ieature dimension compression, and classifier design need to be accomplished. In the area of face recognition, on the basis of algebraic feature extracted, a classifying system with multiple classifier fusion based on wavelet transformation is proposed. Firstly, the high- dimension feature space is transformed into several low - dimension feature spaces by orthonormal wavelet, with which the objective of the feature dimension compression is attained: then, the patterns of the feature spaces are parallelly classified by the fuzzy- based BP neural network(FBPNN): finally, those classified result is fusioned by a FBPNN and the final classified result is obtained. The experiments show that the proposed classifying system attain a satisfactory clasified effect.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2000年第1期22-27,共6页
Pattern Recognition and Artificial Intelligence
基金
国家"973"重点基础研究发展规划
国家"211"基金
关键词
模式识别
图像识别
小波变换
多分类器融合
Pattern Recognition, Features, Orthonormal Wavelet, Neural Network, Fusion
