The need of reliable authentication and forensic analysis tools for multimedia contents has recently led to the investigation of phylogenetic analysis strategies for near-duplicate (ND) images, i.e., pictures generated from a common one through processing operations. As an example, it is possible to detect which image within a group has been used to generate the others. Unfortunately, the accuracy of these algorithms is significantly impaired when the analyzed set includes semantically-similar (SS) images, i.e., pictures reproducing the same subject from different viewpoints that can be easily confused as NDs. This paper presents an image phylogenetic analysis strategy that is able to distinguish between SS and ND pictures thanks to the estimation of the geometric localization of the acquiring viewpoints. The proposed solution has been tested under different experimental setups showing an improved accuracy and a lower computational load with respect to other state-of-The-Art strategies.
Phylogenetic analysis of near-duplicate and semantically-similar images using viewpoint localization
BESTAGINI, PAOLO;TUBARO, STEFANO
2016-01-01
Abstract
The need of reliable authentication and forensic analysis tools for multimedia contents has recently led to the investigation of phylogenetic analysis strategies for near-duplicate (ND) images, i.e., pictures generated from a common one through processing operations. As an example, it is possible to detect which image within a group has been used to generate the others. Unfortunately, the accuracy of these algorithms is significantly impaired when the analyzed set includes semantically-similar (SS) images, i.e., pictures reproducing the same subject from different viewpoints that can be easily confused as NDs. This paper presents an image phylogenetic analysis strategy that is able to distinguish between SS and ND pictures thanks to the estimation of the geometric localization of the acquiring viewpoints. The proposed solution has been tested under different experimental setups showing an improved accuracy and a lower computational load with respect to other state-of-The-Art strategies.File | Dimensione | Formato | |
---|---|---|---|
wifs2016ndss.pdf
Accesso riservato
:
Pre-Print (o Pre-Refereeing)
Dimensione
4.12 MB
Formato
Adobe PDF
|
4.12 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.