Digital Tamper Detection Tools
Hochschule Karlsruhe, mastersthesis, (September 2005)Abstract
A digital image is a data representing a two dimensional scene. Digital images have surpassed
analog images in all fields of applications. In today’s world of advanced computer
technology, tampering and synthesis of digital images can be easily performed by a novice
with a number of available sophisticated image processing software’s like Adobe Photoshop,
Corel Draw etc... In the fields such as forensics, medical imaging, e-commerce, and industrial
photography, authenticity and integrity of digital images is essential. In medical field
physicians and researchers make diagnoses based on imaging. This motivates the need for
detection tools that are transparent to tampering and can tell whether an image has been
tampered just by inspecting the tampered image.
As a foundation for this thesis work, a study on various existing image tampering techniques
and existing state-of-art tamper detection techniques has been carried out. From the
knowledge gathered, tampering of images is performed using standard image processing
software. Copying parts of an image and pasting in the same image for covering unwanted
information or creating a fake image by splicing two or more images are most used techniques
in digital image manipulation. These are called copy-move and image-splicing techniques
respectively.
Some of the proposed state-of-the-art image tamper detection techniques have been selected
for implementation as a plug-in for ImageJ, a scientific image processing software. For
detecting copy-move technique, matching algorithms based on direct image pixel blocks,
quantized discrete cosine transform coefficients of pixel blocks and principal components
analyzed pixel blocks have been implemented. In order to reduce computational complexity
and better matching of pixel blocks, two dimensional array sorting method “lexicographical
sort” and data search logics were implemented. For detection of image-splicing technique,
noise variance estimator based on image statistical moments has been implemented and a
modification of the algorithm by initial usage of Laplace filter on the image has been
proposed for better detection of noise variations.
To evaluate the performance of detection algorithms, a database of 135 self tampered images
were created and passed for tamper detection with the four selected detection algorithms to
collect 540 resultant output images. Statistical hypothesis testing technique is selected as basis
for performance evaluation. 540 images collected were compared with their respective
reference images, average sensitivity and accuracy were calculated, and the robustness curves
were plotted against the affects of JPEG compression quality, additive Gaussian noise, and
Gaussian blur. From the plots of robustness it is observed that, on an average, 80% of the
tampered region is detected with an accuracy of 98% when copy-move technique is used for
tampering images and 50% of the tampered region was detected with an accuracy of 96%
when image-splicing technique is used for tampering images.
Thus the objectives of this thesis work have been successfully accomplished by implementing
a tool for digital image tamper detection and by evaluating its performance against various
factors.