Image classification on projection based multilayer sparse representation

This paper describes a novel multilayer-sparse-representation based image classification. This method designs a dictionary for sparse coefficients of each layer with ADMM (Alternating Direction Method of Multipliers) referring training images. For the classification stage, sparse coefficients should...

Full description

Saved in:
Bibliographic Details
Published in2019 IEEE 8th Global Conference on Consumer Electronics (GCCE) pp. 1069 - 1070
Main Authors Hirakawa, Tomoya, Kuroki, Yoshimitsu
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2019
Subjects
ADMM
Atomic layer deposition
Convex functions
Dictionaries
Dictionary Learning
Image classification
Multilayer Sparse Coding
Nonhomogeneous media
Principal component analysis
Training
Transforms
Online AccessGet full text
DOI10.1109/GCCE46687.2019.9014638

Cover

More Information
Summary:This paper describes a novel multilayer-sparse-representation based image classification. This method designs a dictionary for sparse coefficients of each layer with ADMM (Alternating Direction Method of Multipliers) referring training images. For the classification stage, sparse coefficients should also be calculated with ADMM for test images, which needs computational burden. To reduce the burden, this work proposes to project inputs onto dictionary atoms of each layer instead of solving sparse coefficients. This alternative method is inspired by CNNs (Convolutional Neural Networks), and is also faster than solving sparse coefficients. Experimental results show that our method can predict coefficient vectors faster than the conventional methods with almost equivalent classification accuracy.
DOI:10.1109/GCCE46687.2019.9014638