BIB-VERSION:: CS-TR-v2.0
ID:: ncstrl.dartmouthcs//TR2017-837
ENTRY:: October 02, 2017
ORGANIZATION:: Dartmouth College, Computer Science
REQUESTED-BY:: farid@cs.dartmouth.edu
REQUESTED-FOR:: farid@dartmouth.edu
REQUESTED-DATE:: Thu Sep 28 13:29:43 EDT 2017
TITLE:: A Statistical Prior for Photo Forensics: Object Removal
TYPE:: Technical Report (paper)
REVISION:: 1
AUTHOR:: Fan, Wei
AUTHOR:: Farid, Hany
DATE:: October 2017
RETRIEVAL:: For a paper copy, email
RETRIEVAL:: For a paper copy, write to
Technical Report Librarian
Department of Computer Science
Dartmouth College
6211 Sudikoff Laboratory
Hanover, NH 03755-3510
USA
RETRIEVAL:: PDF at http://www.cs.dartmouth.edu/reports/TR2017-837.pdf
ABSTRACT::
If we consider photo forensics within a Bayesian framework, then the probability
that an image has been manipulated given the results of a forensic test can be
expressed as a product of a likelihood term (the probability of a forensic test detecting
manipulation given that an image was manipulated) and a prior term (the probability
that an image was manipulated). Despite the success of many forensic techniques,
the incorporation of a statistical prior has not been previously considered. We describe
a framework for incorporating statistical priors into any forensic analysis and specifically
address the problem of quantifying the probability that a portion of an image is the result
of content-aware fill, cloning, or some other form of information removal. We posit that
the incorporation of such a prior will improve the overall accuracy of a broad range of
forensic techniques.
END:: ncstrl.dartmouthcs//TR2017-837