Article Preview
Top1. Introduction
With the advancements of image editing tools and softwares have made image tampering very easy consequently claiming their authenticity became uncertain. A very common tampering method with digital image is Copy-Move, where a continuous portion of pixels block is copied and pasted into a different location in the same image to hide or expose some object or scene. To make effective tampering, the copy move regions are often created with resampling, illumination and geometrical transformations. The example of such Copy-Move problem that appeared in newspaper, July 2008 is given in Figure 1.
Figure 1. An example of copy move image tampering that appeared in newspapers in July, 2008 shows four Iranian missiles but only three of them are real.(Tampered and original image from left to right).
In courts of law, where images are presented as basic evidence, its verification plays a crucial role. Images can be tampered to change its meaning and influence the judgment. Many prominent personalities of Film industry, Television are also being victimized by the image tampering attack. When someone does image tampering with malicious intention, it becomes forgery. Image tampering/forgery has begun such era where seeing is no longer believing. Subsequently, it has become important to prove the authenticity of the image and bring truth towards the world.
Image tampering detection techniques can be classified in two categories: Active and Passive. In Active image tampering detection techniques we have additional pre-embedding information in the image. If we have Active image tampering detection techniques then it is easy to detect tampering in the image by using pre-embedding information like Digital Signature (DS) and Digital Watermark (DW). However if we have no pre embedding information and source of image is also unknown, then it will be a very challenging problem to detect image tampering in such images. Generally, images found on internet don’t contain information regarding source camera, DS and DW, therefore Active image tampering detection techniques fail at these points. To overcome the problems of Active image tampering detection techniques, Passive techniques are introduced recently. Passive image tampering detection techniques utilize intrinsic properties of the images, this technique can be roughly categorized into five categories: (1) Pixel based (2) Format based (3) Hardware or Camera based (4) Physics based (5) Geometric based H.Farid (2009).
In Pixel based technique, the main emphasis is on the pixel which is the underlying building block of a digital image. There are many techniques for detecting various forms of tampering, each of which directly or indirectly analyzes pixel-level correlations or matching that arise from a specific form of tampering. The various images tampering approach in this category are Copy-Move, Splicing, and Resampling. Digital forensic analysts always try to preserve the evidence for investigation purpose, lossy image compression technique such as JPEG might be create big challenge to forensic analyst. Format based technique include Double JPEG compression, JPEG Blocking for finding out the clue regarding tampering. Hardware or Camera based tampering detection use Sensor Noise, Color Filter Array, Camera Response Function, Chromatic Aberration, Gamma Correction and White Balancing features of a camera to detect tampering in the image. In Physics based image tampering detection major attention are drawn on light direction and light environment. In Geometric based tampering detection Principal point and Metric measurement play important role.
Many copy-move tampering detection methods are based on direct matching of an image pixels block or transform coefficients and these are not effective when the copy-move regions have some scaling, rotation, scaling-rotation and illumination transforms. The reason behind considering Speed up Robust features (SURF) Bay et al.(2006), Scale Invariant Features Transform (SIFT) Lowe(2004), Hybrid features like SURF with Histogram of Oriented Gradients(HOG) (Dalal et al.,2005) and SIFT with HOG is that these features are invariant with respect to illumination and geometrical transformation and provide commendable results.