基于两级分类器串行的人脸识别

Face recognition based on two-level serial classifier

在分析人脸特征提取和分类器的基础上,提出一种两级分类器串行结合的模型进行人脸识别。在第一级分类器中利用极坐标傅立叶变换提取全局特征通过相似度匹配进行粗略的筛选,第二级分类器中利用改进的协同神经网络,基于原始灰度图像的小波变换提取内在特征,进行精细识别。研究分析了分类器串行结合模型中阈值的选取与系统精度、速度间的关系。在自建人脸库和Yale B人脸库上的实验结果表明,两级分类器串行的识别模型在保证较高系统识别率的前提下可以提升系统的速度。

After analyzing feature extraction and classifier, model of two serial classifier for face recognition is proposed. In the first level classifier, uses polar Fourier transform to extract global features and similarity matching is used for coarse screening. In the second level classifier, uses the improved synergetic neural network to fine recognition based on the intrinsic features that extracted directly from the gray scale extracted by wavelet transforms. Then the research on relationship between threshold selection and system accuracy, speed based on serial combination model. Finally, experiments on self face database and Yale B face database validate that the proposed two serial classifier face recognition model can improve the system speed under the conditions of a higher recognition rate.