PNIPU: Mathematical method for identifying twins

The main articles are:

• Facial recognition systems
• Biometric identification technologies

2022: Development of a mathematical method for face recognition

On July 8, 2022, Perm National Research Polytechnic University"" reported that university scientists have developed a mathematical method that is capable to distinguish not only of twins, but also of one person of different ages. The technology of polytechnics was presented at a scientific conference in. AOE By the way, the to data Wikipedia corporation Google spent more than two billion on dollars solving this problem, as a complex of the problem of recognition in general.

As of July 8, 2022, the facial recognition system is actively used in computer vision technologies in video analytics, in the permit-pass mode, perimeter protection and for educational and research purposes. However, most available technologies have an error in the identification of twins, thereby leaving the possibility of personality substitution.

According to scientists, the checked photo from the database or from the video stream is subjected to automatic correction of brightness and contrast with color alignment on RGB channels, is passed through the Sobel filter to obtain a black and white image. The technology finds the horizontal location of the eyebrows (At tilt angles, a module for returning faces to the vertical is used), then a grid of 6x6 lines is built in the machine (using the rules of the golden section), which gives 25 values ​ ​ in the form of a vector with 26 values, as an average, after which the array is sorted by 26 values, which allows for any size of the base for analysis to take only six values. The Fourier transform applies to each rectangle.

During the analysis, persons of different genders, ages, face change through the photobot, slopes and turns are taken into account. In addition, the program takes into account the different living conditions of the twins, which in different ways affected the elements of the face to a greater or lesser extent. So, from a base data consisting of dozens of people, the system was able to recognize 99%. Recognition time for one photo is 2 Msec. In flow with calculated vectors. The result allows you to assert competence algorithm in solving this problem and not only in it, and in the task identifications of people of different ages. An increase in the percentage of recognition is achieved due to the equality of units of all values ​ ​ of the vector out of 25, except for those responsible for the eyes, eyebrows, nose, part of the lips, said Associate Professor of the Department of Automation and Telemechanics of the Perm Polytechnic, Candidate of Technical Sciences Yuri Lipin.