A new sparse representation algorithm based on kernel spatial non-minimum residual error for classification

In the original feature space, the reconstruction error of each class tends to be closer when the data don’t satisfy the isodirectional distribution or linear-separable. So choosing the minimum mean square error as decision function will misclassify the samples easily and decline algorithm's pe...

Full description

Saved in:
Bibliographic Details
Published in:Optik (Stuttgart) 2015-12, Vol.126 (23), p.4665-4670
Main Authors: Hu, Zheng-ping, Peng, Yan, Zhao, Shuhuan
Format: Article
Language:English
Subjects:
Face recognition
Kernel space
Nonlinearly map
Sparse coefficient accumulation
Sparse representation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the original feature space, the reconstruction error of each class tends to be closer when the data don’t satisfy the isodirectional distribution or linear-separable. So choosing the minimum mean square error as decision function will misclassify the samples easily and decline algorithm's performance. To address this issue, a new sparse representation algorithm based on kernel spatial non-minimum residual error is proposed. First, to make the data spatial-separable, all the samples are mapped into a high-dimensional kernel feature space by non-linear operator. And then, owing to the very high or even infinite dimensions of the kernel space, dimensionality reduction method is introduced to convert the space into a low-dimension subspace. Finally, sparse representation is performed and the sparse coefficient accumulation decision rule is used for classification. Experiments on ORL, Extended Yale B and AR database demonstrate the robustness and effectiveness of the proposed approach.
ISSN:0030-4026
1618-1336
DOI:10.1016/j.ijleo.2015.08.088