Fakery Image Identifying Using Lighting Direction

We are living in a world where seeing (or hearing) is no longer believing. The technology that allows for digital media to be manipulated and distorted is developing at break-neck speeds. And at the same time, our understanding of the technological, ethical, and legal implications is lagging behind. Therefore,developing new methods to detect forgery images or tampering medias is imperative.Generally speaking, forgery image detection can be divided into classes: passive method and positive methods. In fact, the paper I wrote is belonged to the former.Digital watermark is a typical method belonged to the positive method, and the signature is inserted into the picture when it is formed. The signature is power evidence to prove that the image is true or not. Now, the newest software photoshop can insert a signature into any image.There are five basic classes that make of passive digital image forensic technology: pixel-based, format based, camera based, physics based and geometry based. Though there are many ways to identify an image, there is no method that can deal with every tampering picture. Such as the format based method which mainly detects the tampering information in the JPG/JPEG, and it can not deal with lossless compression format. Another example is the inconsistencies in lighting which is reasonable if a picture is shot in a shinny day. If it is a cloudy day, this means can not be used. However, a method that can identify all fakery images does not exist all over the world. So, we have a long way to go.In the paper, I have introduced the principle of inconsistencies in lighting in detail, and some improvements are proposed in the paper too.In the first chapter, the urgency of developing digital forensic technology is expressed. Then many passive identifying methods and their principles are introduced. I think these are very useful for my future work.The lighting inconsistency which can be used in fakery image detection is detailed carefully. Because of it's the basis of the paper, I not only introduces the principle, but also the experimental result.The standard approaches for estimating light source direction begin by making some simplifying assumptions: (1) the surface of interest is Lambertian (the surface reflects light isotropically); (2) the surface has a constant reflectance value; (3) the surface is illuminated by a point light source infinitely far away; and (4) the angle between the surface normal and the light direction is in the range 0°to 90°. Under these assumptions, the image intensity can be expressed as: where R is the constant reflectance value, L is a 3-vector pointing in the direction of the light source, N(x,y) is a 3- vector representing the surface normal at the point (x,y), and A is a constant ambient light term.In order to evaluate the lighting direction from above model, we firstly need to obtain I(x,y) and N(x,y). It is obvious that the normal direction of a point on a boundary can t be computed from single image.Fortunately, however, ones have observed that the z-component of the surface normal is zero, i.e. N z? 0at the occluding boundary. It is a very important condition which makes the latter solution is tractable.As the statement above, if we fit the occluding boundary using a curve, and the normal vector can be evaluated easily. Therefore, the remaining work is how to compute the intensity at a point at the occluding boundary. Though Farid et al [JF05] have introduced two ways to evaluated it, the mistake exists in one model. Following is the modified model:Now, let us begin to evaluate the lighting direction.In the third chapter, the necessary of building an intensity model is stated firstly. The author analyzes the shortcoming in the former model, and proposed my own model to evaluate the intensity at the occluding boundary. In the experiment, the author found that if the place of a point we chose is not proper, and the fitting result would be bad which of course could affect the experimental results. Position constraint is an excellent solution for such problem. On the other hand, the author has observed that there is no ways to deal with some light sources such as strip lighting source, ring light source and so on so far. The author divides these light sources into many point light sources , and therefore the problem is transformed into a multiple light source problem which we have detailed in the second chapter. The author gives a lot of experimental results in the paper which shows that the method proposed by me is effective at the end.The author analyzes the current condition in the field of fakery image identifying at the last chapter, and concludes the incomplete works in the paper, and outlooks the future developments of fakery image detection.People in China pay a little attention to fakery image detection because they don t value their portrait rights unless their profits are not affected. But fakery image detection is imperative with the improvement of people s consciousness.The paper to all method is just like a drop of water to the sea. The method introduced in the paper just only deals with specific image which limits its application. A more deep and convenient algorithm is needed for future application.