Image clustering based on compressed camera fingerprints
Reference persons TIZIANO BIANCHI
Thesis type RESEARCH / EXPERIMENTAL
Description In every imaging sensor, slight variations in the properties of individual pixels produce a noise-like, yet deterministic pattern affecting every image or video taken by that sensor. This feature, called photo-response non-uniformity (PRNU), can be used as a unique fingerprint for identifying the specific acquisition device. However, using PRNU fingerprints can be challenging when only a large dataset of pictures is given, without any clue regarding the acquisition device, since this requires, at least in principle, to compare any possible pair of pictures for verifying whether they come from the same camera.
This research group has developed an innovative technique for compressing PRNU fingerprints based on random projections, which sensibly reduces storage requirements and matching complexity, with only a negligible degradation in terms of matching accuracy.
The objective of this thesis is to develop and test a reduced complexity clustering algorithm that uses compressed PRNU fingerprints for classifying a large dataset of pictures into sets of images taken by the same camera, assessing its performance with respect to existing techniques.
Required skills Signal processing skills, Matlab programming
Deadline 03/07/2020 PROPONI LA TUA CANDIDATURA