基于分块SIFT和GMM的人脸识别方法

Abstract

当前人脸识别方法对采集环境要求严格,多是基于不变特征格,普适性不强.提出一种基于分块SIFT和GMM的人脸识别方法,首先在人脸整体和各分块图像上分别提取SIFT特征,并采用PCA方法进行降维处理,得到鲁棒性强和区分能力好的人脸特征;然后借鉴假设检验的思想,用通用的背景模型描述生物特征识别问题,并通过GMM方法和本文提出的成对模型构建人脸的背景模型和个体模型,据此计算待识别样本与注册样本之间的相似度,求取分类指数,采用分层决策框架实现人脸识别.仿真实验表明,本文方法对环境变化的识别鲁棒性强、识别性能好.

The current face recognition methods have strict requirements on the acquisition environment,mostly based on invariant feature lattices,which are not universal.This paper presents a face recognition method based on partition SIFT and GMM.First,we extract SIFT features on the face as a whole and for each sub-block images,and use PCA method for reducing the dimension of features,to obtain face features that are robust and easy to distinguish;then according to the ideological of hypothesis testing,we use a common Background Model to describe the problem of biometric identification,and construct Background Model and Individuality Model by using GMM method and paired models generating method proposed in this paper.On this basis,we calculate the similarity between the sample to be identified and the one from registration database,compute the classification index,and execute hierarchical decision to achieve face recognition.Experiments show that the proposed method is robust to changes in the environment,and have good recognition performance.