Improved Algorithm Of Support Vector Machine Based On Compressive Sensing And Its Application In Image Processing

SVM is a common machine learning method,which is widely used in the realization of function regression and classification,but the model of this method is complicated to solve.LS-SVM converts the SVM model solution problem into a linear equation solution,so as to reduce the computational complexity.This paper proposes improved algorithms for LS-SVM and applies them to image inpainting and super-resolution reconstruction.The paper includes the following contents:(1)Considering the correlation of training data,LS-SVM based on additive high-order kernel functions is proposed to improve the regression effect of the model by making full use of the correlation between multi-dimensional training data.For small sample problems,this paper increases the mount of training data by introducing gradient information to construct a regression model;The function regression results show that LS-SVM based on additive high-order kernel functions improves the regression effect of multidimensional data,and LS-SVM based on gradient information effectively constrains the regression curve of sample data.(2)The use of sparse compressive sensing LS-SVM were studied based on sparse samples LS-SVM and LS-SVM based on sparse support vector.Among them,based on sparse samples of LS-SVM in experiment for periodic signals of good performance,but the performance is not ideal for non periodic signal;based on sparse support vector LS-SVM in experiment whether periodic signal or non periodic signal with regression better effect,improve the generalization ability of the model.(3)Applying the improved algorithm of LS-SVM to the image inpainting,the additive higher order kernel function can make full use of the correlation between the training data and improve the effect of image restoration.The introduction of gradient information makes the damaged images clear details and more natural edges.The LS-SVM algorithm based on support vector sparse improves the computational efficiency of the model under the objective and subjective conditions.(4)The improved algorithm of LS-SVM is applied to the super-resolution reconstruction,compared with the common interpolation algorithm,the LS-SVM algorithm based on the additive higher order kernel function makes the image of the super-resolution reconstruction more smooth and the detail information is not prominent;the image of super-resolution reconstruction is not only more clear by introducing gradient information,and the image is closer to the original image.The LS-SVM algorithm based on support vector sparse improves the computational efficiency of the model,but the results of super-resolution reconstruction are vague and the high-frequency information is less.