Geometry / physics based techniques Sample Clauses

Geometry / physics based techniques. Up to now we have presented only works that tackle editing detection from a signal processing point of view. Inconsistencies or footprints left by processing operations are searched using statistical tools and models. In this section we introduce a different kind of approach, usually referred to as “geom- etry / physics based”. Instead of looking at signal properties, the idea is to reveal inconsistencies introduced by tampering at the “scene” level (e.g. inconsistencies in lighting, shadows, colors, per- spective, etc.). One of the main advantages of these techniques is that, being fairly independent on low-level characteristics of images, they are extremely robust to compression, and remain applicable even when the quality is low. Going to extremes, these techniques can be applied even for old analog photographs: for example, Farid carefully examined a famous photo of Xxx X. Oswald1 that dates back to 1963 [180]. To the best of our knowledge, methods of this kind have only been developed to target forgery detection. The basic consideration underlying these techniques is that human brain is not good at all in creating forgeries that are consistent from a geometric/physic point of view. This leads to think that most forgeries will likely contain slight errors that can be detected. However, humans are not good at detecting these kinds of inconsistencies. As a consequence of these facts, the great majority of the works in this category provide tools for helping a human analyst to search for inconsistencies at the scene level. Therefore, it is usually assumed that the analysis is assisted. Furthermore, some kind of physical footprints are left by the acquisition device (like chromatic aberration, see later) and do not depend on the scene content but on the device characteristics. Notice that the kind of inconsistencies searched by these methods are likely to be introduced when a cut-and-paste attack is performed. Conversely, a copy-move attack is usually hard to reveal es- pecially when targeted to hide something, except for a few techniques (for example, those based on lens distortion). Similarly to the previous section, we adopt a simple classification based on the kind of footprint. Notice that it is not easy to objectively assess the performance of these techniques because, being human-assisted, they can not be tested on massive amounts of data. As a conse- quence, while each paper shows very good results on all of the reported examples, the validity of t...