Research Of Feature Extraction And Classification Algorithm Based On Face Images

With the rapid development of artificial intelligence,face recognition technology has been widely applied in many fields to verify the identity information.Feature extraction is of great importance for face recognition.In today's information age,how to extract primary message from redundant information is a problem demanding prompt solution.Principal component analysis?PCA?is a widely used algorithm for extracting useful information from original face images.PCA method can find a projection matrix so as to projects the high-dimensional to a lower dimensional space with as less as possible loss of information.Based on the PCA algorithm,researchers have proposed many improved methods,which indicates that using PCA for feature extraction is becoming one of the most popular research fields.In addition,the effective classification of face images is another indispensable work.Face recognition based on linear expression model is a hot sot,and there are many classical algorithms in this field,such as:Lasso regression and ridge regression.In practical recognition condition,there is a lot of noise in face images,such as emotion,gesture,illumination,occlusion and so on.Therefore,how to improve robustness of the model in the practical environment has become a problem.Based on the above discussion,this thesis focuses on the feature extraction algorithm and face image recognition and classification algorithm.The highlights and main contributions of this thesis are as follows:1.This thesis studies the PCA based methods,and proposes a novel algorithm for feature extraction based on the symmetry characteristics of face pictures.It is well known that human face structure has natural symmetry,which is not considered in traditional PCA method.The proposed method fully considers the symmetry of face structure,and it can extract features more effectively compared to other PCA based methods.The extracted feature space is composed of two parts:the low-dimensional projection matrix of the left face samples and the difference matrix of the left and right faces,through which the whole face image can be reconstructed.2.A face image classification algorithm based on sparse representation is studied and a robust face image recognition and classification algorithm is proposed in this thesis.On the basis of linear expression,this algorithm proposes an error detection function for the fidelity term in the linear expression model to calculate the expression error of test samples.The traditional Lasso regression model and ridge regression model both use₂l-norm as the loss function,and this method proposes to use Huber function as the loss function in the linear regression model to obtain the sparse coding of samples more accurately.Finally,the expression of the test samples was obtained through the sparse coding,so as to learn the label of the unknown samples.