Image Classification by Multilayer Feature Extraction Based on Nuclear Norm Minimization

This paper describes a novel multilayer image classification method based on a nuclear norm of dictionary atoms minimization; in each layer, the method designs dictionary atoms for training dataset by solving a nuclear norm regularized convex optimization problem. The problem promotes non-orthogonal...

Full description

Saved in:
Bibliographic Details
Published in2019 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS) pp. 1 - 2
Main Authors Hirakawa, Tomoya, Kubota, Shohei, Kuroki, Yoshimitsu
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2019
Subjects
dictionary learning
image classification
nuclear norm
Online AccessGet full text

Cover

More Information
Summary:This paper describes a novel multilayer image classification method based on a nuclear norm of dictionary atoms minimization; in each layer, the method designs dictionary atoms for training dataset by solving a nuclear norm regularized convex optimization problem. The problem promotes non-orthogonality of dictionary atoms and makes coefficient vectors non-negative; these contribute to the improvement of classification accuracies. Experiments result shows that our method is superior to the conventional KLT (Karhunen-Loéve Transform) based multilayer classification method in classification accuracy.
ISSN:2642-3529
DOI:10.1109/ISPACS48206.2019.8986349