Image authenticity recognition
In the formation of the image authentication is directly related to pattern recognition. But these are different processes. The main difference is that two images of the same image can differ in geometric properties and brightness. The image of the image also varies depending on the angle of the subject, light hit, differences in gestures or facial expressions.
Possible problems
Recognition technology is based on the fact that each pixel of an image is an axis in space, and its brightness affects the position of an object in space. Therefore, most recognition programs do not take into account the two-dimensional properties of images. So rotation of a picture in a plane (change of topology in a two-dimensional plane) or an increase in scale "knocks down" the recognition process. It is unrealistic to program the software so as to take into account all the scales of the object and all its positions because of their unlimited number.
Another problem with authentication is brightness variation. In this case, the entire image shifts to another point in space. And the program does not respond to the content of the image, but to its brightness and size.
However, modern image processing programs have learned to correct the geometric position and brightness of the object. There are also programs that transform images (moments) so that their coefficients are identical regardless of the scale and location of the object. This is why image normalization must be run to authenticate an object.
Another obstacle in the way of image recognition is the curvature of the two-dimensional projection due to the three-dimensional characteristics of the object (angle and lighting), as well as the deformation of the object itself (posture or gesture). Therefore, it is recommended to start image processing before recognition.
How to avoid problems
For this, a training set of images is formed, fixed in different angles and with different lighting. In this case, the desired image is within the specified conditions and is quickly recognized. Extensive training allows other input objects to be recognized without variations in lighting or angles. But this method is cumbersome, therefore it is applicable primarily to statistics, and not to practice.
The second method is to restore a three-dimensional object in two-dimensional space. In this case, the angle and lighting do not affect recognition, and the pose or gesture is quickly processed. But it is difficult to implement the method in practice. This is due to the fact that during the restoration of a three-dimensional object, a generalized model of the image is used, otherwise the recognition process is disturbed.
A high-quality image recognition program overcomes deformations on its own without training models and accurate images. The authenticator splits the image into 2D slices and configures itself to change the content of the slices.
Specialized software is divided into two groups: general recognition programs for a wide range of tasks and programs for a specific industry, which differ in clearer results of image authentication.
Online authentication
Simple online tools allow you to quickly verify the authenticity of images and photos, and extract the most information about the person in the photo.
Simple resources offer to download a picture or a link and display data: information, when and on what device the photo was taken, the location of the photo, image parameters.
More "advanced" online programs allow you to identify the added image fragments. After processing, a picture appears on the page in the browser with the selected parts that were edited.
A simple Google image search performs the reverse finding of the image: the search engine finds the source of the uploaded image and indicates the resources where it has already been published.
Multifunctional applications for Windows process images in AVI, DNG, PDF, THM formats and detect edited image fragments and errors in the warped file.
Recognition algorithm
The process of establishing the authenticity of a photograph or picture takes place according to a certain algorithm of three stages.
At the first stage, the authorship of the photo or its original source is established. The easiest way is to contact the person who uploaded or uploaded the image and ask a question about the author directly. Another option is to upload the image to Google search engine, which finds identical images, and sometimes allows you to determine if the image has been edited. The original source of the image is a link to the photo in the highest resolution.
At the second stage, the user who has distributed the picture or photo is checked. There are several resources available to collect information about a person. An example site for finding an Internet footprint is Pipl.com. The resource allows you to find the user and his photos. The service searches all social networks: Facebook, LinkedIn, MySpace. To do this, you need to know the last name and first name of the person in Latin transcription. The advantage of the resource is a search on the "deep" Internet, inaccessible to standard search engines and ordinary Internet users. Another interesting site - WebMi - searches by the name of a person, forms a web visibility rating to identify fake accounts. The resource allows you to detect the mention of a name on English-language sites. The Russian analogue of such an Internet search is People.yandex.ru.
At the third stage, the time, date and place of the photo is confirmed. An easy way is to contact the author and check the accuracy of the information using one of the services mentioned above. Specialists working with digital images will "decode" the image parameters provided by the programs. Ordinary users will understand the time and location information. If information is not available, you should carefully review the image. Landscape, weather, license plates, advertisements, buildings - small details will allow you to determine the location of the shooting or confirm that the image is being edited.
An important point in photo recognition is to establish the coincidence of the image with its description. It is possible that the photo is genuine, but the description is not.
The resources of Google Maps, Wikimapia or Panoramio will help to determine the geolocation. These services allow you to match the image with geographic coordinates, thanks to the integration with Google maps.
Information about the weather conditions in the declared place on the set date will help to establish the authenticity of the image. To do this, you need the English-language search service Wolfram Alpha. This is not an ordinary search engine, but rather an information scientific base, on which a robot-"intellectual" is searching, capable of answering various questions in the field of exact encyclopedic knowledge. A feature of the service is the lack of links to other Internet resources. The user receives only the generated response.
Clear formulas, expert knowledge and generic models all help to authenticate images online.
