Design Method of Linear Discriminant Sparse Representation Classifier Based on Kernel Space

Abstract

The invention relates to a design method of a linear discriminative sparse representation classifier based on kernel space, the steps are: read training samples, perform nonlinear transformation on the training samples, transform them into high-dimensional kernel space, and perform each A class of training samples is used for learning, and the contribution (ie weight) made by each individual in this class of training samples to construct the subspace of this class of training samples is found. The product of this class of training samples and the weight matrix forms a dictionary, and all categories of The dictionaries are arranged in turn to form a large dictionary matrix; the linear discriminant sparse code of the test sample in the kernel space is obtained through the dictionary matrix, and the test sample is fitted with each type of dictionary and the linear discriminant sparse code corresponding to the dictionary, and the fitted Error; the class with the smallest fitting error is the class of the test sample. The invention can ensure that the sparse codes of samples of the same type are concentrated, and the sparse codes of samples of different types are dispersed, thereby effectively increasing the discrimination of samples and improving the performance of classifiers.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and in particular relates to a design method of a kernel space-based linear discrimination sparse representation classifier. Background technique [0002] In addition to showing great charm in the field of computer vision, the dictionary learning method based on sparse representation has been successfully applied in face recognition, image classification, image noise reduction and other fields. The pattern recognition process usually includes two stages: feature extraction stage and classification stage. The quality of the classifier directly affects the recognition rate of the pattern recognition system, and the design of the classifier has always been one of the core issues in the pattern recognition research. [0003] At present, the main classifier design methods are as follows. [0004] 1. Support vector machine method (English: Support Vector Machine) [0005] The support vec...

AI Technical Summary

Problems solved by technology

[0015] The present invention aims at the above-mentioned shortcomings of large fitting error and low precision in the classifier designed by the existing classifier design method, and provides a design method of a linear discrimination sparse representation classifier based on kernel space, and the classifier designed by the method The output is both sparse and discriminative, which significantly improves the performance of pattern recognition

Images