The article has been automatically translated into English by Google Translate from Russian and has not been edited.
Переклад цього матеріалу українською мовою з російської було автоматично здійснено сервісом Google Translate, без подальшого редагування тексту.
Bu məqalə Google Translate servisi vasitəsi ilə avtomatik olaraq rus dilindən azərbaycan dilinə tərcümə olunmuşdur. Bundan sonra mətn redaktə edilməmişdir.

How the US Department of State checks applications and photos of participants of the Green Card Lottery

The Green Card Lottery since its inception until now has been and remains the easiest way to immigrate to the United States. For the millions of people participating in the DV program, winning a green card in the lottery is the only way of immigration they can count on. And here it is important not to miss your chance because of annoying little things - mistakes in the profile or wrong photo. Photos when checking profiles are given huge Attention. Let's figure out what should be given special attention when filling out the questionnaire.

Photo: Shutterstock

The drawing process itself is not interesting. The computer randomly selects a certain number of winners from the general database of the region, all over the world - from 100 to 150 thousand. Considering that there are only 50 thousand visas, a rather large reserve is being made, which will cover the elimination of many in the future, namely: "athletes" who play to test their luck; people who, having calmly estimated the pros and cons of immigration, will refuse to continue the process; convicted; did not pass the medical commission; those who cannot collect the necessary documents or enough money; disqualified due to detected errors and deception, and so on.

In addition to all these people, out of 100-150 thousand initially selected winners, those who submitted not one but several applications will be eliminated.

The very first check when applying for a green card lottery

From 7 to 15 million applications are submitted annually to participate in the green card lottery (depending on the participating countries). It is clear that processing and checking such a number of documents by hand is a thankless job, it is fraught with a lot of errors. For this reason, almost the entire processing process is automated and entrusted to special computer programs.

At the time of submitting the application, the built-in validator checks whether all the required fields of the questionnaire are filled in and whether the attached photos correspond to the specified technical parameters. This is the very first technical check. It will skip any applications that formally comply with the rules. For example, it is quite possible for a person named Yyyyyyy Xxxxxxx to submit a profile with a photograph that does not contain a person's image. If this is a 600 by 600 pixel photo, a .jpeg file is less than 240 kilobytes, a color image with a 24-bit depth - all the technical parameters of the file are met, the application will be accepted, and a Confirmation Number will be issued.

However, the participant Yyyyyyy Xxxxxxx will never receive a notification of a win - his application will be rejected at the next stage of verification, and she will not take part in the green card drawing.

The drawing of the green card lottery and the elimination of "junk" applications

In fact, not one general draw is held, but six separate ones - in each of the six regions into which the world for the DV-lottery program is conditionally divided. For each region, a certain quota of winning people is calculated, the sum of these six quotas gives the total number of winnings - 100-150 thousand.

At the time of the draw, all regional applications are randomly assigned new numbers (case numbers).

Then, each application is checked for compliance with the technical requirements of the DV lottery rules. The program takes application number 1, checks it and gives an opinion on its compliance with the requirements. Then it moves on to claim number 2 and so on. If the program detects an application that does not correspond to the specified parameters, it marks it as "junk" and skips its number - the person who submitted it will not see the notification of the prize. So the program checks applications in a row until the quota is filled with “good” numbers.

Since no supporting documents are attached to the questionnaires when submitting an application, it is not possible to establish the accuracy of the data specified at this stage of the verification. A person may well be called Yyyyyyy Xxxxxxx. Therefore, in the future, only submitted photos are checked.

Each image file is analyzed, it establishes whether the person is depicted in the photo and whether its image meets the compositional requirements.

The area of ​​the image in which the head should be located is determined by the rules — the program imposes a virtual mask on this area, which is the average display of the main parts of a human face: eyes, nose, lips, etc. If the parameters of the analyzed area match according to certain criteria with a mask, the photo is defined as containing a face image. At the same time, the program checks that it is a photograph in front of it, and not, for example, a watercolor drawing. Image quality is also checked.

At this stage, questionnaires with attached pictures of cats, cars, clouds, black squares, etc. are discarded. People usually submit such questionnaires in order to see how the site works, to practice before submitting their present application.

Verification of compliance with compositional requirements

The rules set fairly stringent requirements for the image of a person in a photograph: the size of the head, the eye level, the position of the head - its tilt and rotation, the background color are strictly stipulated. In this case, nothing is said about the ears, shoulders, hair, beard and mustache, make-up.

Accordingly, we can expect that the program will check only what has been said.

If the head in the photo is larger or smaller than necessary, turned to the right or left, raised or lowered, eyes higher or lower than it should be, the application will be rejected during verification. If the ears are not visible behind the hairdo, the shoulders are not at the same level, the mustache and beard hide the mouth, but the proportions are met and the person looks directly into the camera - the application will be recognized as valid.

During this check, the correctness of the background and the absence of shadows on it are evaluated - the contours of the head should be clearly defined against the background. Shadows on the face can cause disqualification if the program because of them can not recognize some parts of the face.

Preparing for the first stage of facial recognition

Why are photos so thoroughly checked? The statement will not be a discovery: in order to increase your chance of winning in any lottery, the participant must submit several applications. This is prohibited by the DV lottery rules, but the temptation is great, and many people (and intermediary organizations) resort to various tricks to get around the ban.

For example, applications are submitted with different transliteration of the name and photo with different hairstyles. In order for the computer to understand that this is the same person, face recognition technology is used.

When checking photos submitted for participation in the green card lottery, several steps of machine face recognition are used. Each subsequent recognition algorithm is more complicated than the previous one, so the number of analyzed images decreases with each step - until there are photographs that the program can say with almost 100% certainty that they depict the same person.

Recognition is implemented based on the FaceIt technology developed by Visionic. FaceIt technology is used by the Department of State (DoS) to screen people applying for any visa to enter the United States.

FaceIt works with images compliant with ISO / IEC 19794-5.

Each photo attached to the application for participation in the green card lottery is considered and its suitability for the recognition process is determined, and the following parameters are evaluated:

  1. Is the head size large enough?
  2. Cropping - is the face completely visible on the image?
  3. Centering - is the face sufficiently centered?
  4. Exposure - is the image not overexposed or underexposed?
  5. The eyes are clearly visible - are there glasses on the person, and if so, are the eyes visible or hidden?
  6. Focus - is the image well focused?
  7. Compression - Was the image not overly compressed to remove skin details?
  8. Texture - Does the surface of the skin contain textures suitable for use in face recognition?
  9. Resolution - Does the image resolution exceed the minimum measured in pixels between the eyes?
  10. Faceness - can an object detected in an image be called a human face or not?

At the stage of preparation, the photos are normalized - the computer rotates the images so that the eyes on them are located strictly horizontally (aligns the face with respect to the vertical axis) and crops the images to a new size, cutting off all unnecessary. The images are cropped so that the distance between the centers of the eyes in all images is the same, the brightness and contrast of the images are equalized.

Electronic face recognition on photos

Electronic recognition is applied only to those winning entries that were found to be valid in the previous verification steps and the photographs from which were successfully normalized.

The first stage of identification is vector comparison (VFA)

At the first recognition stage, the vector comparison algorithm (Vector Feature Analysis - VFA) is applied.

The essence of this recognition method lies in the fact that the program presents each analyzed normalized face image as a linear combination of other, previously created special images, the so-called eigenvectors or eigenfaces. The resulting code contains information about this combination. This is how each image that needs verification is encoded.

In a very simplified way, “own faces” (“eigenvectors”) can be represented as a set of certain standard face components obtained by statistical analysis, subsequent sampling and processing of a large set of images of different faces. When using this method, it is taken as an axiom that any human face can be composed of an “average face” (a component that is the same for all faces) by adding to it a certain number of certain “proper faces”. Most of the faces can be obtained by adding a small number of "own faces".

Comparing then the obtained sets of eigenvectors, the system concludes that the original images are similar or different.

Each file with a code, or template, is a simple list of applied own persons (and the percentage expression of the contribution of each person to the image built from them) and has a very small size, so you cannot restore the face image using only the data of this template alone. However, due to the small size of the template files, comparing them with each other is very, very fast.

After all the template files are ready, the program compares the first successful bid with all submitted non-trash bids. She compares the templates obtained from the photographs at the encoding stage, evaluating the coincidence of the codes as a percentage. If the coincidence of the checked code with any other does not exceed a certain value, the program concludes that no duplicates were found in the application.

Then the program proceeds to check the next application - checking the code of the photo from it with all the other codes of all photos from all submitted valid applications. So, one by one, the program checks all applications that won the green card lottery.

If the computer detects that the match of the codes of the two compared photos exceeds the threshold value (this value is one of the adjustable parameters of the identification system), the program makes a note about it and continues checking the code under study.

Thus, as a result of the first stage of recognition, pairs (or triplets, fours, etc.) of photographs are obtained, which the system suspected of being the same person.

Vector comparison is one of the oldest methods used for face recognition. It is extremely demanding to ensure that all analyzed images are strictly normalized. Significant errors in deciding on the similarity of compared faces, inherent in this method, due to the small number of eigenvectors used, force it to be used only as the crudest tool, the basis for further search for duplicates.

The second stage of identification is an algorithm for comparing hierarchical face graphs (HGM)

Hierarchical Graph Matching (HGM) is an algorithm for comparing individuals based on an analysis of the location of control nodes and the distances between them.

A lot of over 2000 points are defined on the face, starting from the centers of the eyes, the bridge of the nose, the tip and wings of the nose, the left and right corners of the mouth, and so on, which, connecting with each other, form the so-called graph, or mask, individual for each face. The distances between the points are alternately entered into the file, while encoding the face image and making it suitable for mathematical comparison.

This method of analysis does not depend on the texture of the face, it examines only its shape. The created mathematical models are built according to the principle “from large to small”, which significantly speeds up the process of comparing files.

The HGM algorithm shows fairly good recognition results in relation to the green card lottery program, where strictly normalized high-quality images are analyzed.

The third stage of identification is the analysis of local differences (LFA)

Local Feature (Local Feature) is a portion of the image, which differs from other neighboring neighboring areas. The site may vary in intensity, color or texture, but does not have to be localized precisely by this change. Local differences may be points, edges, small parts of the image.

The LFA procedure describes a set of local interconnected perceptual fields defined at each point of a virtual grid of receptors superimposed on the facial image. These fields are different from each other, they are optimally added to the original image, and at the output they differ as much as possible. The algorithm for creating files describing the dependence of these fields, and then comparing these files is the basis used to check photos on the FaceIt system green card.

During LFA, the program, by re-creating the encoded file, recognizes and evaluates many local differences in the image of facial structures. The system compares the generated encoded files and marks those that match above a predetermined threshold.

One of the factors that impede the identification of persons is their variability depending on facial expressions. Even a small half-smile sets in motion a large number of facial muscles, while almost all distances between control points of the face change. The LFA method, analyzing the differences in the local areas of the face, which are moreover determined with a large excess, is practically free from the problems associated with facial expressions.

The fourth stage of identification is the analysis of the face surface texture (STA)

Surface Texture Analysis (STA) should be understood as a combination of technologies and methods of identification using photographs that allow a sufficiently detailed examination of the skin texture of recognizable faces. Accordingly, it is the quality of the images that is of paramount importance for successful recognition.

The STA is used as a further development of the method of analyzing local differences, which makes it possible to use for comparison even smaller details that the surface of any person’s skin abounds in.

As a rule, a facial area free from excessively dense vegetation is analyzed - from the bottom of the eyes to the beginning of the upper lip.

In short, the STA method can be described as follows: first, the average brightness of each pixel is determined as the average gray scale of the pixels surrounding it. Then, the image of the face, by comparing the gray scale values ​​of pixels in their average brightness, is converted into a binary image with 1 or 0 values ​​assigned to those pixels whose average brightness is higher than the specified gray value boundary.

Then the selected face zone is divided into small blocks. For each block from the first image being compared, the system searches for the corresponding block in the second image, which is best matched with the compared block.

After that, the continuity of neighboring blocks is estimated. If the relative change in the positions of a pair of neighboring blocks is below a predetermined threshold, they are considered as continuous. The more continuous pairs of blocks, the greater the likelihood that the analyzed images belong to the same person. This probability can be formulated as a function of the number of continuous pairs of blocks.

The analysis of the surface texture requires, in comparison with other recognition methods, more time for calculations, therefore, it is applied at the final stage of identification to those images for which there are sufficient grounds to be considered as belonging to the same person. According to the developer's data, the use of the STA method as an addition to the LFA increases the recognition accuracy by 20-25%, making it possible to confidently distinguish even identical twins.

Recognize photo retouching for DVlottery

When checking applications for participation in the lottery of green cards, in order to prevent changes to the images in photographs using special graphics programs and other technical means, retouching is used.

The instruction for the green card lottery directly says that the retouching of the submitted photos is prohibited - applications that have shown any changes in the photos (show manipulation in any way) will be disqualified.

To detect retouching (the term “retouching” here means any changes in a photo, except for cropping to size), special programs are used that use a combination of several algorithms when searching for changes - from color transformations with the imposition of various filters and recognition of the continuity of chains of blocks of pixels formed during compression images in jpeg, before analyzing the code of the jpeg file.

When making any changes to photographs, you should remember that, with a high degree of probability, retouching will be detected. Of course, you can try to make it difficult to detect retouching by printing on paper and scanning modified photos, photographing the monitor screen with a retouched image, and similar tricks, but you need to understand that the technologies used for verification may include tools to protect against such manipulations.

The result of checking photos for the lottery green cards

The computer cannot give a definite answer whether the same person is depicted in two different photographs. He is only able to estimate the probability of such a coincidence (of course, this probability can be close to 100%). However, the final decision is always made by a person.

In the KCC, for each winner, a case (case) is set up, where the submitted documents and the results of the checks are collected. If, according to the result of checking the photos, duplicates are found in the application (the suspicion that these are duplicates exceeds a certain amount), all these suspicious applications are also included in the case, an entry is made in the case description file, and a red label is pasted on the paper folder with the personal file with an indication of the reason.

The completed file is sent to the consular section where the interview will take place. During the interview, the consul will assess all the circumstances of the case, get acquainted with the results of checking the photographs and visually compare the found suspicious questionnaires, seeing in front of him a living person, the applicant. The result will be a decision whether there was a violation of the rules, whether the applicant submitted several questionnaires. Accordingly, the consul will either approve the issuance of a visa or refuse.

In the first year of DV-2006 (Face Recognition System) operation, 5221 fraudulent applications were found among the winners.

Cheating a duplicate finder on a green card lottery

Submission of several applications by one person is a violation of the rules of the lottery, leading to the automatic disqualification of the applicant.

For a lie to the consul under oath for an interview, you can get a life ban on entering the United States.

The consul, seeing in your case several applications filed with different photos recognized by the verification program, will certainly take an interest in the circumstances of the incident. If a person starts to lie, it will lead to a denial of a visa, and a life ban on entry.

The problem is that the applicant cannot know exactly why the consul asks about this - did the program really recognize his applications or the consul asks a simple routine question.

Based on this, the methods of cheating the recognition system can only be talked about theoretically, for research purposes, and also in order not to accidentally do something forbidden.

Online registration for the lottery Green Card 2022 started October 7, 2020 and will run until 12:00 PM ET November 10, 2020. ForumDaily has prepared all the details and tips for filling out the questionnaire for you here.

Miscellaneous check a photo Educational program Editor's Choice green card lottery

Read also on ForumDaily:

15 significant disadvantages of living in the United States, according to a Russian-speaking immigrant

In Illinois, elderly illegal immigrants will receive free medical insurance

Registration for the DV-2022 green card lottery has opened: where and how to participate

How to fill in the form on the green card: detailed instructions

Do you want more important and interesting news about life in the USA and immigration to America? Subscribe to our page in Facebook. Choose the "Display Priority" option and read us first. And don't forget to subscribe to ForumDaily Woman and ForumDaily New York - there you will find a lot of interesting and positive information.

1042 requests in 2,049 seconds.