Detecting Image Forgeries using Geometric Cues

This chapter presents a framework for detecting fake regions by using various methods including watermarking technique and blind approaches. In particular, we describe current categories on blind approaches which can be divided into five: pixel-based techniques, format-based techniques, camera-based techniques, physically-based techniques and geometric-based techniques. Then we take a second look on the geometric-based techniques and further categorize them in detail. In the following section, the state-of-the-art methods involved in the geometric technique are elaborated. INTRODUCTION Today's digital technology has begun to erode our trust on the integrity of the visual imagery since image editing software can generate highly photorealistic images (Farid, 2009). Doctored photographs are appearing with a growing frequency and sophistication in tabloid magazines, mainstream media outlets, political campaigns, photo hoaxes, evidences in a courtroom, insurance claims, and cases involving scientific fraud (Farid, 2009). With the rapid advancement in image editing software, photorealistic images will become increasingly easier to be generated and it becomes difficult for people to differentiate them from photographic images (Lyu & Farid, 2005). If we are to have any hope that photographs can hold the unique stature of being a definitive recording of events, we must develop technologies that can detect the tampered images. Therefore, authenticating the integrity of digital image's content has become particularly important when images are used as critical evidence in journalism and security surveillance applications. Over the past several years, the field of digital forensics has emerged to authenticate digital images by enforcing several authentication methods. The presence or absence of the watermark in interpolated images captured by the camera can be employed to establish the authenticity of digital color images. Digital watermarking (I.J. Cox & M.L. Miller & J.A. Bloom, 2002; H. Liu & J. Rao & X. Yao, 2008) has been proposed as a means to authenticate an image. However, a watermarking must be inserted at the time of recording, which would limit this approach to specially equipped digital cameras having no capabilities to add a watermarking at the time of image capture. Furthermore, the watermarking would be destroyed if the image is compressed and the ruin of watermark would make the method failed. Passive (nonintrusive) image forensics is regarded as the future direction. In contrast to the active methods, blind approaches need no prior information that is used in the absence of any digital watermarking or signature. Blind approaches can be roughly grouped into five categories (Farid, 2009): (1) pixel-based techniques that analyze pixel-level correlations arising from tampering. Efficient algorithms based on pixels have been proposed to detect cloned (B. Mahdian & S. Saic, 2007; A. Popescu & H. Farid, 2004; J. Fridrich & D. Soukal & J. Lukas, 2003), re-sampled (A. C. Popescu & H. Farid, 2005), spliced (T. T. Ng & S. F. Chang, 2004; T. T. Ng & S. F. Chang & Q. Sun, 2004; W. Chen, & Y. Shi, & W. Su, 2007) images.Statistical properties (H. Farid & S. Lyu, 2003; S. Bayram, & N. Memon, & M. Ramkumar, & B. Sankur, 2004) in natural images are also utilized; (2) format-based techniques detect tampering in lossy image compression: unique properties of lossy compression such as JPEG can be exploited for forensic analysis (H. Farid, 2008; J. Lukas & J. Fridrich, 2003; T. Pevny & J. Fridrich, 2008). (3) camera-based techniques exploit artifacts introduced by the camera lens, sensor or on-chip post-processing (J. Lukas, & J. Fridrich & M. Goljan, 2005; A. Swaminathan & M. Wu & K. J. Ray Liu, 2008). Models of color filter array (A. C. Popescu & H. Farid, 2005; S. Bayram & H. T. Sencar & N. Memon, 2005), camera response (Y. F. Hsu & S. F. Chang, 2007; Z. Lin & R. Wang & X. Tang & H.Y. Shum, 2005) and sensor noise (H. Gou & A. Swaminathan & M. Wu, 2007; M. Chen & J. Fridrich &M. Goljan & J. Lukas ,2008; J. Lukas, & J. Fridrich & M. Goljan, 2005) are estimated to infer the source digital cameras and reveal digitally altered images. Other work such as (A. Swaminathan & M. Wu & K. J. Ray Liu, 2008) trace the entire in-camera and post-camera processing operations to identify the source digital cameras and reveal digitally altered images using the intrinsic traces. (4) physically-based techniques model and detect anomalies using physical rules. For example, three dimensional interaction between physical objects, light, and the camera can be used as evidence of tampering (M.K. Johnson & H. Farid, 2005; M. K. Johnson & H. Farid, 2007). (5) geometric-based techniques make use of geometric constraints that are preserved or recovered from perspective views (M. K. Johnson & H. Farid, 2006; M. K. Johnson, 2007; W. Wang & H. Farid, 2008; W. Zhang & X. Cao & Z. Feng & J. Zhang & P. Wang, 2009; W. Zhang & X. Cao & J. Zhang & J.Zhu.&P. Wang, 2009). Several geometric-based techniques (M. K. Johnson & H. Farid, 2007; W. Wang & H. Farid, 2008; W. Zhang & X. Cao & Z. Feng & J. Zhang & P. Wang, 2009; M. K. Johnson & H. Farid, 2006) have been proposed in the field of image forgery detection. The estimation of internal camera parameters including principal point (M. K. Johnson & H. Farid, 2007) and skew (W. Wang & H. Farid, 2008) can be used as evidence of tampering. In (M. K. Johnson & H. Farid, 2007) the authors showed how translation in the image plane is equivalent to a shift of the principal point and differences in which can therefore be used as evidence of forgery. Wang and Farid (W. Wang & H. Farid, 2008) argued that the skew of the re-projected video is inconsistent with the expected parameter of an authentic video. The approach has the advantage that the re-projection can cause a non-zero skew in the camera intrinsic parameters, but there are also some drawbacks that it only applies to frames that contain a planar surface. Zhang et al. (W. Zhang & X. Cao & Z. Feng & J. Zhang & P. Wang, 2009) described a technique for detecting image composites by enforcing two-view geometrical constraints. The approach can detect fake regions efficiently on pictures at the same scene but requires two images correlated with H (planar homography) or F (fundamental matrix) constraints. Metric measurements can be made from a planar surface after rectifying the image. In (M. K. Johnson & H. Farid, 2006), the authors reviewed three techniques for the rectification of planar surfaces under perspective projection. They argued that knowledge of polygons of known shape, two or more vanishing points, and two or more coplanar circles can be used to recover the image to world transformation of the planar surface, thereby allowing metric measurements to be achieved on the plane. Each method in (M. K. Johnson & H. Farid, 2006) requires only one single image but fails in measurements for objects out of the reference plane. Wang et al. (G. Wang & Z. Hu & F. Wu & H. T. Tsui, 2005) show how to use the camera matrix and some available scene constraints to retrieve geometrical entities of the scene, such as height of an object on the reference plane, measurements on a vertical or arbitrary plane with respect to the reference plane, etc. The single view metrology using geometric constraints has been addressed in (A. Criminisi & I. Reid & A. Zisserman, 1999). The authors demonstrated that the affine 3D geometry of a scene may be measured from a single perspective image using the vanishing line of a reference plane and the vertical vanishing point. However, they are only concerned with measurements of the distance between the plane which is parallel to the reference plane and measurements on this plane. This chapter is organized as follows. After reviewing the background in section 2, the involved methods based on geometric technique are described in sections 3. The future research direction is given in section 4 and the final conclusions are drawn in section 5. BACKGROUND Photographic alterations have existed about as long as photography itself. However, before the digital age, such deceptions required mastery of complex and time-consuming darkroom techniques. Nowadays, anyone who has a little of computer skill can use powerful and inexpensive editing software to create tampered images as he or she likes. Therefore, as sophisticated forgeries appear with fast and alarming frequency, people’s belief in what they see has been eroded (H. Farid, 2009). A more recent example of photo tampering came to light in July 2008. Sepah News, the media arm of Iran’s Revolutionary Guard, celebrated the country’s military prowess by releasing a photo showing the simultaneous launch of four missiles. But only three of those rockets actually left the ground, a fourth was digitally added. The truth emerged after Sepah circulated the original photo showing three missiles in flight—but not before the faked image appeared on the front pages of the Chicago Tribune, the Financial Times, and the Los Angeles Times. Figure 1. A July 2008 photo shows four Iranian missiles streaking skyward. The right is the true image Sepah News replaced the faux photo with the original without explanation. Over the past few years, the field of digital-image forensics has emerged to challenge this growing problem and return some level of trust in photographs. Nearly every digital forgery starts out as a photo taken by a digital camera. The camera’s image sensor acts as the film. By using computer methods to look at the underlying patterns of pixels that make up a digital image, specialists can detect the often-subtle signatures of manipulated images that are invisible to the naked eye. Traditionally, watermarking is added into the images or video to give the validating information for image authentication. However, the watermarking can be easily destroyed in the process of image compression. Recently, digital blind techniques emerge in the field of image forgery detection. These techniques work on the assumption that although digital forgeries may leave no visual clues that indicate tampering, they may alter the underlying statistics of an image. MAIN FOCUS AND CONTRIBUTION OF THE CHAPTER In this chapter, we focus on the category on the geometric-based techniques in image forgery detection. Geometric techniques, which appear as a new application: the nonintrusive digital image forensic can be further divided into four categories: (1) techniques based on the camera’s intrinsic parameters; (2) techniques based on metric measurement; (3) techniques based on multiple view geometry; (4) techniques based on other geometrical constraints. After reviewing the literature in the field, we propose a potential solution for image integrity’s authentication, which is based on the published geometric method on 3D height measurement, measurements on the vertical or an arbitrary plane with respect to a reference plane. Our proposed solution enriches the detecting methods for image forgery, which provides a prospect to build a integrated framework incorporating various methods for fake region detection. Solutions and Recommendations Preliminary Camera model The general pinhole camera can also be written as: