Digital Image Tamper Detection Based on Multimodal Fusion of Residue Features - Advanced Concepts for Intelligent Vision Systems - page 79

pre-processing chain. By sliding segmentation of image frames, we extract intra-
frame and inter-frame pixel sub-block residue features, transform them into optimal
cross-modal subspace, and perform multimodal fusion to detect evolving image
tampering attacks, such as JPEG double compression, re-sampling and retouching.
The promising results presented here can result in the development of digital image
forensic tools, which can help investigate and solve evolving cyber crimes.
2 Background
Digital image tamper detection can use either active tamper detection techniques or
passive tamper detection techniques. A significant body of work, however, is
available on active tamper detection techniques, which involves embedding a digital
watermark into the images when the images are captured. The problem with active
tamper detection techniques is that, not all camera manufacturers embed the
watermarks, and in general, most of the customers have a dislike towards cameras
which embed watermarks due to compromise in the image quality. So there is a need
for passive and blind tamper detection techniques with no watermark available in the
images.
Passive and blind image tamper detection is a relatively new area and recently
some methods have been proposed in this area. Mainly these are of two categories
[1, 2, 3, 4]. Fridrich [4] proposed a method based on hardware aspects, using the
feature extracted from photos. This feature called sensor pattern noise is due to the
hardware defects in cameras, and the tamper detection technique using this method
resulted in an accuracy of 83% accuracy. Chang [5] proposed a method based on
camera response function (CRF), resulting in detection accuracy of 87%, at a false
acceptance rate (FAR) of 15.58%. Chen et al. [6] proposed an approach for image
tamper detection based on a natural image model, effective in detecting the change of
correlation between image pixels, achieving an accuracy of 82%. Gou et al [7]
introduced a new set of higher order statistical features to determine if a digital image
has been tampered, and reported an accuracy of 71.48%. Ng and Chang [8] proposed
bi-coherence features for detecting image splicing. This method works by detecting
the presence of abrupt discontinuities of the features and obtains an accuracy of 80%.
Popescu and Farid [3] proposed different CFA (colour filter array) interpolation
algorithms within an image, reporting an accuracy of 95.71% when using a 5x5
interpolation kernel for two different cameras. A more complex type of passive
tamper detection technique, known as “copy-move tampering” was investigated by
Bayram, Sencar, Dink and Memon [1,2] by using low cost digital media editing tools
such as Cloning in Photoshop. This technique usually involves covering an unwanted
scene in the image, by copying another scene from the same image, and pasting it
onto the unwanted region. Further, the tamperer can use retouching tools, add noise,
or compress the resulting image to make it look genuine and authentic. Finally,
detecting tampers based on example-based texture synthesis scheme was proposed by
Criminisi et al[9] that is based on filling in a region from sample textures. It is one of
the state-of-the-art image impainting or tampering schemes. Gopi et al in [10]
proposed a pattern recognition formulation and used auto regression coefficients and
neural network classifier for tamper detection.