Research On Isometric Projection Algorithm And Its Application In Image Recognition

With the advent of the era of big data,the value of data has gradually attracted attention in many fields.However,image data obtained in real life usually has a high dimensional and complex non-linear structures.In many fields,such as machine learning and computer vision,how to deal with high dimensional data has become a common problem we faced.There is a method to solve the above problem to reduce the dimension of the data on the premise of reducing the loss of original information.Recently,manifold learning,as a method of dimensionality reduction of non-linear data,has the characteristics of being able to learn complex non-linear structures in data and find its essential dimensions.In recent years,a large number of research results about manifold learning have emerged.However,non-linear dimensionality reduction methods based on manifold learning,such as isometric mapping algorithms,cannot easily process newly added image sample data.Therefore,classic manifold learning algorithms are difficult to apply to classification problems such as image recognition.The linear generalization of the classic manifold learning algorithm provides ideas for solving this problem.The predecessors have already studied the linearization of the manifold learning algorithm to some extent.However,some work needs to be done to apply it to everyday life.The main work of this thesis is as follows:(1)The isometric projection algorithm is researched in this thesis.Since the isometric projection algorithm combines the advantages of the traditional linear dimensionality reduction algorithm and the classic nonlinear manifold learning algorithm,we first did some research on the two types of algorithms.Aiming at the small-samplesize problem of the isometric projection algorithm,based on the isometric projection algorithm,the exponential matrix and the maximum marginal criterion are introduced,and some existing linear mapping algorithms are improved to some extent.The thesis proposes an algorithm called isometric projection base on maximal margin criterion.(2)The experimental comparison of the isometric projection base on maximal margin criterion proposed in this thesis is carried out.Handwritten Digits database(handwritten numbers database and handwritten letters database),ORL database,Georgia Tech database and COIL-20 database were selected in the experiment.For principal component analysis algorithm,isometric projection algorithm,regularized isometric projection algorithm,exponential isometric projection algorithm and linearity based on maximum marginal criterion the isometric projection base on maximal margin criterion algorithm was compared experimentally.Different numbers of training samples and subspace dimensions were selected as a series of experiments.The experimental results prove that the isometric projection base on maximal margin criterion algorithm has a better recognition accuracy,which is 88.69%,78.85%,97.50%,73.12%,and 99.68%.(3)This thesis designs and implements an image recognition system,and applies the isometric projection base on maximal margin criterion algorithm proposed in this thesis to image recognition.In the test,it shows that this image recognition system can recognize image samples of a certain size.To sum up,this thesis focuses on the problems of isometric projection algorithm in image recognition and improves it.And the proposed algorithm has been well verified in general image databases.