Camera source identification with limited labeled training set

Yue Tan, Bo Wang, Ming Li, Yanqing Guo, Xiangwei Kong, Yunqing Shi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

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.

Original languageEnglish (US)
Title of host publicationDigital-Forensics and Watermarking - 14th International Workshop, IWDW 2015, Revised Selected Papers
EditorsIsao Echizen, Hyoung Joong Kim, Yun-Qing Shi, Fernando Pérez-González
PublisherSpringer Verlag
Pages18-27
Number of pages10
ISBN (Print)9783319319599
DOIs
StatePublished - Jan 1 2016
Event14th International Workshop on Digital-Forensics and Watermarking, IWDW 2015 - Tokyo, Japan
Duration: Oct 7 2015Oct 10 2015

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9569
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other14th International Workshop on Digital-Forensics and Watermarking, IWDW 2015
CountryJapan
CityTokyo
Period10/7/1510/10/15

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Keywords

  • Camera source identification
  • Ensemble projection
  • LBP features
  • Limited labeled training samples

Fingerprint Dive into the research topics of 'Camera source identification with limited labeled training set'. Together they form a unique fingerprint.

Cite this