Semantic Segmentation of Facial Images in Biometric Authentication Systems for Personnel of Critical Infrastructure Facilities

The problem of the article is to increase the efficiency of biometric authentication of personnel of critical infrastructure facilities. It is shown that one of the main directions of increasing efficiency is to improve the procedure for selecting facial contours in the test image, the result of which is the determination of a rectangular area that does not provide accurate selection of facial contours and interference during video recording. To overcome these limitations, it is advisable to use neural network semantic segmentation tools that allow you to accurately select facial contours, the eye area, as well as areas with overlaps or background elements. At the same time, known solutions in the field of semantic segmentation show that most of them do not provide the possibility of effective functioning in the conditions of critical infrastructure facilities. In order to overcome these shortcomings, the article has developed a method for determining the architectural parameters of a neural network model of semantic segmentation of a facial image during biometric authentication at critical infrastructure facilities. The method allows taking into account the specifics of the expected conditions for the use of semantic segmentation tools and provides an opportunity to reduce the number of experimental studies related to the determination of the architectural parameters of the neural network model by up to 10 times. The results of experimental studies confirm that the model developed as a result of the implementation of the proposed method provides a 1.1-1.2-fold increase in the accuracy of semantic segmentation of a face image and allows reducing the influence of background diversity and interference, which improves the quality of input data, which is subsequently used for face recognition.