Font Size: a A A

Image Recognition Algorithm Based On Graph Embedding And Its Application

Posted on:2021-05-19Degree:MasterType:Thesis
Country:ChinaCandidate:J LiFull Text:PDF
GTID:2428330611473198Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Feature extraction plays an important role in computer vision and pattern recognition.Principal component analysis(PCA)and linear discriminant analysis(LDA)are two wellknown feature extraction modes,but they are difficult to find the underlying manifold structure of the data.In recent years,graph-based feature extraction modes have attracted much attention.The key issue is how to construct graphs to discover the essential structure of the data.This article describes the classic feature extraction algorithm and proposes some new feature extraction algorithms based on graph embedding.The effectiveness of each algorithm is verified on some data sets,including:(1)This paper proposes a local discriminative projection feature extraction algorithm based on competitive cooperative representation.This method uses a method based on competitive cooperative representation to construct inter-class graphs and intra-class graphs.Considering the influence of various types of coefficients in adjacent graphs,we introduce the idea is to sparse adjacency graphs with positive representation coefficients,and to calculate the local structure of the image by calculating the intra-class divergence matrix and inter-class divergence matrix and obtain its optimal projection matrix.Experimental results on some data sets show that this method can effectively improve the efficiency of image recognition.(2)It is considered that the matrix regression construction graph using kernel norms has been widely studied in feature extraction algorithms.However,it will inevitably produce negative coefficients when solving the coefficients.Physically,it is not suitable to use addition and subtraction to reconstruct realistic applications.For this,this paper proposes a non-negative representation matrix regression feature extraction algorithm based on kernel norm.This algorithm constructs an inter-class graph and an intra-class graph through kernel norm matrix regression with non-negative constraints.The intra-class divergence matrix and the inter-class divergence matrix are used to characterize the local structure of the image and obtain its optimal projection matrix.The effectiveness of the algorithm is verified on multiple image data sets.(3)In order to learn effective features and give up useless features,and make full use of the local neighborhood relationship of the training samples,this paper proposes a feature extraction of discriminant analysis based on similarity weighted local collaborative representation.Maintain the similarity between training samples while retaining local information and overall discriminative information,and use the ratio of inter-class divergence and intra-class divergence as a criterion to learn the optimal projection matrix and extract the most effective features,thereby further improving the classification of the algorithm effect.Experiments show that the algorithm can improve the performance of image recognition.
Keywords/Search Tags:feature extraction, cooperative representation, image recognition, graph embedding, non-negative representation
PDF Full Text Request
Related items