Camera source identification with limited labeled training set

Document Type

Article

Publication Date

1-1-2016

Abstract

This paper investigates the problem of model-based camera source identification with limited labeled training samples. We consider the realistic scenario in which the number of labeled training samples is limited. Ensemble projection (EP) method is proposed by introducing prototype theory into semi-supervised learning. After constructing sub-sets of local binary patterns (LBP) features, several pre-classifiers are established for all labeled and unlabeled samples. According to the ranking of posterior probabilities, several prototype sets are constructed for the ensemble projection. Combining the outputs of all labeled samples from classifiers trained by prototype sets, a new feature vector is generated for camera source identification. Experimental results illustrate that the proposed EP method achieves a notable higher average accuracy than previous algorithms when labeled training samples is limited.

Identifier

84964026263 (Scopus)

ISBN

[9783319319599]

Publication Title

Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics

External Full Text Location

https://doi.org/10.1007/978-3-319-31960-5_2

e-ISSN

16113349

ISSN

03029743

First Page

18

Last Page

27

Volume

9569

Grant

2012RZJ01

Fund Ref

Fundamental Research Funds for the Central Universities

This document is currently not available here.

Share

COinS