在人脸识别过程中,基于2DPCA特征提取方法具有直接、高效等特点。但它只包含了二阶统计信息,因而丢失了可能对分类很有用的高阶统计信息而使识别率受到一定影响。SVM采取升维的方法把线性不可分问题转变为线性可分问题,识别率较高,但直接对图像分类时运算量大、运行时间长。文章结合两者的优点,使用了2DPCA和SVM相结合的人脸识别方法,即先利用2DPCA进行特征提取,然后把降维后的数据输入SVM进行分类识别。该方法在ORL、YALE人脸库上的实验表明,不但可以提高识别率,而且所用时间明显减少。
The two-dimensional PCA method extracts feature directly and rapidly in the process of face regnition. However, this method only involes two-dimentional statistical information and lacks useful high-order statistical information for classifying, this may influence the recognition rate. SVM changes the nonlinear question change into the linear question by promoting the dimensions, thus making recognition rate even higher. However, the computational amount is large when this method is used. Thus a new approach in combination with two dimensions PCA and SVM, is proposed to recognize the human face, that is, the 2DPCA is first used to deal with feature extraction, then SVM to make use of the feature to do classification. Experiments with ORL and YALE face-databases show that the proposed method has achieved a higher recognition rate with a reasonable time cost.