A robust visual image classification method and system

Abstract

The invention discloses a robust visual image classification method and system. In order to effectively realize the category prediction of unlabeled samples in training samples and the rapid induction and reasonable dimensionality reduction of samples to be tested, the error measurement based on elastic regression analysis is integrated into Train an out-of-sample label propagation model, and normalize the manifold regularization term through parameter trade-offs, based on soft labels l 2,1 Norm regularized label fitting term and based on l 2,1 The impact of the norm regularized elastic regression residual term on the sample description and category identification completes the establishment of the label propagation model; and then obtains the projection matrix used to determine the category of the sample to be tested through the iterative optimization of the label propagation model. Therefore, in this application by introducing based on l 2,1 Norm regularized regression error term and soft label l 2,1 Norm regularization can effectively improve the robustness of the system while inheriting the advantages of the label propagation classification method, making the induction process of the samples to be tested fast and accurate.

Description

technical field [0001] The present invention relates to the technical field of data mining, and more specifically, relates to a robust visual image classification method and system. Background technique [0002] Classification technology has always been the most fundamental and core research topic in data mining. In recent years, classification technology has developed rapidly, and many excellent classification methods have emerged, such as SVM, decision tree, association rules, neural network and deep learning, etc. Many classification systems have been put into use and have produced huge social and economic benefits. In the field of label propagation, a projection-based fast inductive classification method has emerged, which greatly improves the accuracy and speed of label propagation classification. [0003] Label propagation classification is a semi-supervised learning method based on similarity graph construction. In actual operation, the supervised data (data with c...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a robust visual image classification method and system to solve the problem of slow classification speed in the label propagation classification method of the prior art

Images