Image Classification Based On SVM

The so-called image classification is an image processing method for a given image,in which the category matching the image is selected as the output.Image classification is a basic problem in the field of image processing.It is a very active research direction in the fields of computer vision,machine learning and pattern recognition,and has many practical applications.Therefore,it is of great theoretical and practical significance to study image classification algorithms with strong robustness and high accuracy.This paper chooses support vector machine as a classifier in image classification for the following reasons: SVM is a structural risk minimization model,which solves the problem of over-fitting when the number of samples is small,and effectively improves the generalization ability of the model.In the case of non-linear separability of samples,SVM maps the features of samples to high-dimensional space,so that the samples can be linearly separable in high-dimensional space,and the non-linear problems that are difficult to solve can be transformed into more easily solved linear problems.Moreover,SVM introduces a kernel function,which can efficiently calculate the inner product of high-dimensional vectors and effectively reduce the complexity of the model.The innovation of this paper is as follows: The selection of kernel function has a great influence on the generalization ability of SVM.The commonly used method of selecting kernel function is exhaustive method,which has a certain randomness and depends on the training of the model,so its efficiency is low.To solve this problem,a kernel function selection algorithm based on kernel-target alignment is proposed in document On Kernel-Target Alignment.Kernel functions are often regarded as similarity measures of samples in feature space.Kernel-target alignment is the premise.This paper proves that some kernels do not satisfy the three properties of similarity measures.Therefore,some kernels can not be regarded as similarity measures of samples in feature space.Aiming at this problem,this paper proposes a kernel function selection algorithm based on feature distance,which proves that feature distance is more suitable for similarity measurement of samples in feature space than kernel function,and solves the problems existing in the alignment of core objects.And a new evaluation function is defined.Comparedwith the evaluation function in the core target alignment,the evaluation function does not need to be normalized and has lower computational complexity.Finally,the experiment shows that the model generalization ability is very good by selecting kernel function to construct SVM based on feature distance.