Image Feature Extraction In The Application Of Face Recognition

In recent years,computer and network technology develops rapidly.After the NPC &CPPCC in 2015,'Internet plus initiative' has been a national strategy.At present,biological recognition technology has become an important research direction,and face recognition is popular with the majority of scholars because it is friendly and it will not be recognized easily.Since the number of the face image's dimension is so high,the original image processing takes longer and consumes more resources.However,the nature of the data structure is often only a small part of parameters behind these high-dimensional data,so feature extraction plays a vital role in the process of face recognition.The possibility of extracting the vector with best discrimination ability also becomes the main standard to judge the face recognition algorithm.In view of the variety of facial expressions,facial expression recognition also plays an important role in face recognition.This paper studies the traditional linear dimension reduction algorithm,thus improved dimensionality reduction algorithms are proposed,giving birth to new models and algorithms for expression recognition.The main research contents and innovative work of this paper are as follows:(1)Improvement of LDA Algorithm Based on MMCAn optimization criterion of LDA is given,and the robustness of this criterion is achieved.To be specific,the null space of the total scatter matrix is first removed to remedy the singularity problem.Then the eigen-subspace corresponding to each specific eigenvalue is achieved.Finally,in each eigen-subspace,the discriminability of each eigenvector is measured by the maximum margin criterion and the discriminant vectors are obtained by optimizing this criterion.We also conduct extensive experiments to evaluate the proposed method on various well-known datasets,and we compare it with other dimension reduction methods.Experimental results demonstrate the effectiveness of the proposed method.(2)Feature Extraction Technology Based on Tangent VectorThe concept of tangent vector is introduced in the traditional LDA,and the concrete process of obtaining the tangent vector is given.Then the tangent vector is added as additional information to the original image information.Finally combined with the KNN algorithm,it aims to reduce the dimension and classifies the image information.Experiments on human face images are compared with other dimensionality reduction algorithms.The experimental results show that the proposed method outperforms the traditional LDA,and in most cases better than other linear dimensionality reduction algorithms.The new method is effective for different data types,and has strong robustness.(3)Kernel Partial Least Squares Regression with Applications to Facial Expression RecognitionFacial expression recognition,as a classification problem,has received much attention from researchers due to its important roles in the human-computer interaction.In this paper,I analyze(kernel)partial least squares regression and obtain a new method to solve them.Moreover,it is worth noting that the first stage of kernel partial least squares regression is equivalent to generalized discriminant analysis of feature extraction.At last,we also demonstrate that the kernel partial least squares regression can be directly applied to the classification problems by defining the dummy matrix.In conclusion,experimental results show that the proposed algorithm outperforms other linear dimensionality reduction algorithm,which is commonly used in most cases;the kernel partial least squares regression algorithm,which is based on facial expression recognition,also achieves good classification performance.