Research On Noisy Image Classification Based On Least Square Framework Of Correlation Analysis

With the rapid development of science and technology,the scale of image data is increasing.These data have the characteristics of redundancy,high dimension and unstructured,etc.,so that it is difficult to find the rules and associations among them.In order to analyze and process the correlation data reasonably,and enable users to quickly obtain reliable and effective low-dimensional feature information from massive data,this thesis proposes a least square correlation analysis framework based on sample penalty and feature selection technology.The main work is as follows:In this thesis,a least square correlation framework based on sample penalty is proposed.We incorporate four correlation analysis algorithms into the least square framework,and the sample penalty factor is added to obtain better features in noise image data.Different from the traditional correlation analysis method,the least square correlation framework based on sample penalty not only considers reducing reconstruction error,but also punishes sample points to distinguish the influence of different sample points on the correlation analysis framework.The experimental results show that the proposed method is superior to the traditional correlation analysis method in the accuracy of face image classification and the effect of noise suppression on the model,which proves the superiority of the proposed method in the aspect of noise suppression on the image.In this thesis,a least square correlation framework which combines sample penalty and feature selection is proposed.The proposed method aims to achieve feature and sample selection in sample space and dimension space.We use the feature selection technology based onL_2,1 norm to further eliminate the redundant features brought by sample selection,solve the over fitting problem in the training process,and improve the robustness of noise image.The classification results of handwriting datasets MNIST and USPS show that the proposed method has superior classification performance compared with the traditional feature selection algorithms.At the same time,the result of image classification in multi noise environment proves the proposed method has stronger generalization ability and feature expression ability,and can effectively against the impact of noise on the image.A noise image classification system based on the least square correlation analysis framework of sample penalty and feature selection is designed and implemented.The system is divided into user management module and image classification module.The system has a good test running interface,correct function,and can cope with the impact of different kinds of noise on the image,which verifies the correctness and rationality of the proposed method.