Classification of digital camera-models based on demosaicing artifacts

Sevinc Bayram, Husrev T. Sencar, Nasir Memon

Research output: Contribution to journalArticlepeer-review

Abstract

We utilize traces of demosaicing operation in digital cameras to identify the source camera-model of a digital image. To identify demosaicing artifacts associated with different camera-models, we employ two methods and define a set of image characteristics which are used as features in designing classifiers that distinguish between digital camera-models. The first method tries to estimate demosaicing parameters assuming linear model while the second one extracts periodicity features to detect simple forms of demosaicing. To determine the reliability of the designated image features in differentiating the source camera-model, we consider both images taken under similar settings at fixed sceneries and images taken under independent conditions. In order to show how to use these methods as a forensics tool, we consider several scenarios where we try to (i) determine which camera-model was used to capture a given image among three, four, and five camera-models, (ii) decide whether or not a given image was taken by a particular camera-model among very large number of camera-models (in the order of hundreds), and (iii) more reliably identify the individual camera, that captured a given image, by incorporating demosaicing artifacts with noise characteristics of the imaging sensor of the camera.

Original languageEnglish (US)
Pages (from-to)49-59
Number of pages11
JournalDigital Investigation
Volume5
Issue number1-2
DOIs
StatePublished - Sep 2008

Keywords

  • Camera
  • Camera-model
  • Color filter array
  • Demosaicing
  • Image forensics
  • Source identification

ASJC Scopus subject areas

  • Pathology and Forensic Medicine
  • Information Systems
  • Computer Science Applications
  • Medical Laboratory Technology
  • Law

Fingerprint Dive into the research topics of 'Classification of digital camera-models based on demosaicing artifacts'. Together they form a unique fingerprint.

Cite this