Results for person identification by face
The article was published in issue №3
This paper discusses modern methods and technologies for identification techniques by facial image.
All known approaches are based on the selection and analysis of face parameters in the image and their further processing. Processing the obtained face parameters is mainly based on neural network and mathematical approaches. The disadvantages of the neural network approach are mathematical problems (retraining, choosing
the optimization step, getting into a local optimum).
The problems of the mathematical approach are low speed and low resistance to the image defects, while
the neural network method can correct image defects at the stage of image alignment. Also, the mathematical method is very demanding in terms of computing resources.
During the work on the paper, the method of active form models and the neural network method of identifying a person by the image of a person were selected and tested on two data samples. The active shape model method is used to detect facial features and obtain key points on faces. The neural network method using a convolutional neural network retrieves a descriptor that describes a person, which is a vector of 128 features. Further, by determining the distance between the vectors, the most similar vector is located in the database.
During testing, the speed of the method and the accuracy of the work were measured. The test results showed a performance of approximately 2 secs in two samples and an accuracy of 97 %. These studies are related to the development and implementation of a module for streaming identification of people by video stream, where the reaction speed of the method is very important, as well as its accuracy.