Research On Image Classification Algorithm Based On Metric Learning

Image classification,as a basic research problem in the field of machine learning and computer vision,has gradually become an increasingly important research in the field of image processing and other research areas.In the era of rapid development of Internet and Internet of things,the rapid development of information technology has brought more and more convenience and development.How to efficiently organize,retrieve and distinguish massive image data in this network has become a very important and key problem.Therefore,a very important and key way to solve this problem is image classification.Therefore,in-depth study of image classification is of great practical value and profound theoretical significance.After many years of research,many algorithms of image classification have been put forward in succession.At present,metric learning plays a very important role in image classification.In the field of machine learning and computer vision,metric learning has attracted more and more attention from researchers in the field of machine learning and computer vision.In this paper,image classification is based on metric learning.The main work of this paper is as follows:(1)This paper describes research situation and significance of the research direction of image classification based on metric learning.This paper studies Mahalanobis metric learning under two criteria and several image classification methods based on Mahalanobis metric learning.(2)Image classification algorithm based on ellipse mahalanobis metric learning is proposed.The ellipse mahalanobis learning problem in this paper gives training sample data,then learns a linear transformation method which can reflect the spatial structure of the information or the semantic information through training,thus makes elliptic mahalanobis metric learning method have a better distinction.It is known that linear transformation is a special form of fractional linear transformation,thus the method based on ellipse mahalanobis metric learning is more widely used.In this paper,the image classification algorithm has better effect.Firstly,extract HSV color feature and LBPs texture feature,combining the image feature to build features of the images in this paper;then calculate the initial ellipse mahalanobis metric matrix and define ellipse mahalanobis metric according to the statistical characteristics of the sample data.Then obtain the optimal elliptic mahalanobis metric matrix with the use of ellipse mahalanobis metric learning;finally calculate the distance between the image features by learning elliptic mahalanobis metric and achieve the image classification.(3)Image classification algorithm based on ellipse metric learning with quadratic discriminant analysis is proposed.In view of the key problems of metric in image classification,the image classification algorithm based on ellipse metric learning firstly extracts the color histogram and scale invariant Scale Invariant Local Ternary Patterns histogram,and obtains effective image feature using the local maximum occurrence processing.Then ellipse metric is introduced,the Gauss distribution is modeled through the difference of intra-class and inter-class,and the Quadratic Discriminant Analysis model is constructed.Then,the elliptic metric matrix is defined according to the logarithmic likelihood ratio of the two models.In order to maintain the maximum classification information,a supervised linear discriminant analysis is used to generate new feature subspace,and the optimal elliptic metric matrix is learned.Finally,the ellipse metric is used to compute the distance between the features of the images to complete the classification.Experiments are carried out on the three datasets of VIPeR,QMULGRID and CUHK03,experiments show that the algorithm improves the performance of image classification.