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
Recommended Citation
Tan, Yue; Wang, Bo; Li, Ming; Guo, Yanqing; Kong, Xiangwei; and Shi, Yunqing, "Camera source identification with limited labeled training set" (2016). Faculty Publications. 10818.
https://digitalcommons.njit.edu/fac_pubs/10818
