利用图像预处理手段提高2DPCA人脸识别性能

针对二维主成分分析在特征提取上存在的缺点,提出一种综合的预处理方法.先将图像对角化表示,保留图像对角方向结构信息;再通过小波变换得到一定尺度的图像;最后利用图像灰度的指数衰减策略,降低算法对光照等变化的敏感性.通过这样序贯处理,能够充分利用2DPCA特性提取图像结构特征,消除与识别无关的细节信息,不仅提高了算法识别精度,还降低了算法对计算机硬件的要求.基于ORL数据库的实验表明,采用预处理手段能获得比传统方法更好的识别性能.

Enhancement of performance of human physiognomy recognition with 2DPCA by means of image preprocessing method

Aimed at the disadvantages occurred in feature extraction with 2DPCA during the analysis of two dimensional principal component,a comprehensive preprocessing method was presented.First the image was transformed into diagonal one,reserving the correlations between variations of rows and those of columns of the image,then an image with definite size was obtained by using wavelet transform,and finally the susceptivity of algorithm to the change of illumination was reduced by using exponential decay strategy.In doing so,the characteristics of 2DPCA was fully used to extract the feature of image structure and get rid of irrelevant informations for recognition,so that not only the accuracy of recognition was improved but the requirement of computer hardware necessary for the algorithm was relaxed.The experiment based on ORL database showed that by using above-mentioned preprocessing approach,a better results of recognition could be obtained than by using traditional method.