Research On Image Multi Classification Based On Manifold Regularized Linear Regression

Computer vision is an important branch of machine learning.Image classification,as a basic research in machine vision,has played a huge role in promoting the whole field of machine learning.However,due to the complexity and variability of the real environment,image classification has many technical difficulties.In this paper,the robustness of the algorithm is the goal,and the shortcomings of traditional algorithms are overcome.Combined with linear regression and manifold learning,the main work and research results are as follows:(1)To solve the problem that the linear regression classifier(LRC)does not pay attention to the local structure information of the data,and ignores the differences between the samples within the class,which leads to the poor classification performance when the facial image has changes in expression,illumination,angle,occlusion and so on.A linear regression classifier based on local weighted representation(LWR-LRC)is proposed.Firstly,LWR-LRC constructs a weighted representative sample for each class of samples based on the similarity between test samples and all samples,then decomposes the test samples into linear combinations of weighted representative samples,finally classifies the test samples into the category with the largest reconstruction coefficient.LWR-LRC considers the local structure of samples,constructs the optimal representative samples of each class of samples,and uses the representative samples to calculate,which improves the robustness and greatly time cost.(2)To solve the problem that the discriminant least squares regression(DLSR)is not robust to sparse noise,the robust principal component analysis(RPCA)is integrated into the DLSR framework,and a denoising low-Rank discriminative based Least squares regression(DLRDLSR)is proposed.In this method,noise reduction and regression are combined to construct a joint optimization function to remove the sparse noise from the original data and make the data involved in the regression cleaner.Then in the regression process,the projection matrix discrimination ability is enhanced by using the ?-dragging technique.In addition,some additional constraints are imposed on the noise reduction part to further preserve the information to improve the classification accuracy.(3)To solve the problem that the previous image classification algorithms based on linear regression lose the nearest neighbor relationship of data when dealing with noise,a linear regression image classification algorithm based on regularized latent subspace denoising is proposed(LRDLSR),this algorithm introduces the idea of locality preserving projections(LPP),and adds a regular term for manifold structure optimization to preserve the nearest neighbor relationship.The algorithm proposed in this chapter is superior to the existing image classification methods based on linear regression.