一种用于人脸识别的仿射包集识别模型仿真

An affine package sets used in face recognition recognition model simulation

针对传统建模方法无法建立人脸图像与训练集之间完整的映射关系,导致建模准确度低、效率差的问题。提出采用基于仿射包集识别法的人脸识别方法。建立人脸图像集的仿射包模型,促使图像中的点与仿射包上的点一一对应后,定义仿射包间最近2点间的距离,在融合奇异值分解法组建人脸标准特征矩阵,获取人脸振幅谱表征,将获取的表征投影在标准特征矩阵中,将其投影系数定义为人脸的模式特征,构建最相似人脸特征类别候选集,利用最近邻分类器进行分类,有效实现了对人脸的识别。仿真实验结果证明,采用改进的方法进行人脸识别,其识别效率高,对人脸表情的各种变化都具有较强的鲁棒性。

The traditional modeling method cannot establish the complete mapping relation between face image and the training set,which results in low accuracy and poor efficiency of modeling. A face recognition method based on affine set is proposed. The affine package model of the face image set is established,the point in image and in affine packet one- to- one correspondence is prompted,and the nearest distance between the two points of affine packet is defined. Then,the singular value decomposition method is fused to establish the facial standard feature matrix,and the representation of facial amplitude spectrum is obtained. The obtained representation is projected in the standard feature matrix,and the projection coefficient is defined as the model features of the face to construct the candidate set of the most similar facial feature category. The nearest neighbor classifier is used for classification,and the face recognition is effectively realized. The simulation results show that the improved method has high recognition efficiency is and strong robustness against the changes of the facial expression.