Forensic characterization of image capture devices

Nitin Khanna, Purdue University

Abstract

Forensic characterization of sensors or devices is important in many applications such as establishing the trust and verifying authenticity of data produced by a sensor or device and the sensor or device that created it. Recently there has been a great deal of interest using features intrinsic to a data-generating sensor for the purpose of source identification. Numerous methods have been proposed for various problems related to sensor forensics in general and image forensics in particular. Although a considerable amount of work has been done in forensic identification of digital cameras, more work needs to be done in forensic characterization of scanners, video cameras and other audio devices. This thesis is aimed at developing tools for forensic characterization of devices or sensors, in particular image capture devices. Statistical feature based classifiers are designed for imaging sensor classification and for source scanner identification for images acquired using flatbed desktop scanners. The methods are based on using imaging sensor pattern noise for scanned photographs and texture features for scanned documents, as device fingerprints. The statistical feature vector based methods provide high accuracies, both for native resolution and lower resolution scanned images. The proposed method perform well with images that have undergone JPEG compression with low quality factors, image sharpening, and contrast stretching. The proposed features are also robust to the scan area used for a particular scan so knowledge of the exact location of scanner's bed used for scanning is not needed. The sensor noise based source scanner identification scheme is extended for forgery detection in scanned photographs scanned at native resolution of the scanners. This method can be an effective tool for forgery detection in scanned images if used in co-ordination with other existing methods for forgery detection. The techniques used for both camera and scanner identification are dependent on having prior knowledge of the class of devices (cameras or scanners) that generated the image. If the image was generated by a digital camera, then the digital camera identification methods must be used. Similarly if the image was generated by a scanner, the scanner identification methods must be used to obtain the best identification results. Use of the sensor pattern noise for classifying digital images based on their originating mechanism, a scanner or a digital camera or the use of computer graphics, is investigated. To achieve this, differences in the characteristics of the sensor noise are used. These differences arise between the two classes due to inherent mechanical differences between their respective sensors and image generation mechanisms. As shown by our results, the proposed scheme does not need the availability of the actual source device for training purposes. Thus, images generated by a completely unknown scanner or digital camera can be classified properly.

Degree

Ph.D.

Advisors

Delp, Purdue University.

Subject Area

Electrical engineering|Computer science

COinS