基于改进型DCT和Gabor分块的人脸特征提取与识别

Face Feature Extraction and Recognition Based on Modified Discrete Cosine Transform and Gabor Block

人脸特征提取是人脸识别中重要的一个环节。提出了一种新的方法,利用DCT对人脸图像压缩降维,然后对DCT系数用20组Gabor小波滤波,滤波后的结果采用选择性分块统计方法提取特征向量。最后把特征向量用改进型感知器算法进行分类。以VC++6.0为开发平台在Yale人脸库和0RL人脸库上对该方法进行了测试。实验表明,该方法与常用的PCA、LDA等特征提取方法相比可以有效降低运算时间,并提高识别率。

Face feature extraction is an important part of face recognition. A new method is presented which u ses discrete cosine transform（DCT） for face images compression and dimension reduction, then the DCT coeffi cients is filtered by using 20 groups of Gabor wavelet, the selective block statistics method is used to extract eigenvectors from the result of filtering. Finally, the eigenvectors with modified perception algorithm are classi fied. It is tested on the Yale face database and ORL face database with VC ＋ ＋ 6.0. Experiment shows that compared with the commonly used feature extraction methods such as PCA and LDA, the new method can re duce the calculation and increase the identification effectively.