A low-rank sparse representation image feature learning method based on Laplacian regularization

Abstract

The invention discloses a low-rank sparse representation image feature learning method based on Laplacian regularization, comprising the following steps: (1) randomly dividing a data set into a training set and a test set; (2) constructing an undirected weight of the training set Graph, and calculate its Laplacian matrix; (3) Initialize the feature extraction matrix, and perform initial feature extraction on the training set; (4) Design a learning model of non-negative low-rank sparse representation; (5) Use the LADMAP optimization method to optimize Learn the model to obtain the optimal feature extraction matrix and optimal classifier model parameters; (6) Predict and identify the test set samples to verify the feature extraction effect and classification accuracy. The invention has the advantages of strong robustness, high recognition rate, wide adaptability, etc., extracts features from image samples, retains more sample information, and has stronger discrimination, and can be widely used in target recognition, image classification, and the like.

Description

technical field [0001] The invention relates to a face image recognition method, in particular to a low-rank sparse representation image feature learning method based on Laplacian regularization. Background technique [0002] At present, in large-scale image feature learning, it is difficult to obtain labeled training samples, which makes it difficult to use some existing supervised feature learning techniques, and training samples with noise further limits their performance. [0003] Existing methods usually assume that image sample data are distributed in independent low-dimensional subspaces, or approximately span multiple low-dimensional subspaces, and have a low-rank sparse structure. Some methods utilize low-rank sparse constrained representation learning methods to learn the spatial structure of the original data while denoising the data. It finds the low-rank structure of the data through the nuclear norm, through the l 1 The norm finds the sparse structure of the ...

AI Technical Summary

Problems solved by technology

However, the existing feature learning methods usually do not associate feature learning with classification tasks, making the learned features less discriminative and less stable, resulting in damage to the overall performance of these methods.

Images