Biometric System for Identification Person by Face Image

The article provides a classification of existing approaches to face recognition in an image and describes the implementation of a system of biometric identification of a person by his or her face. To search for a face in an image, use the pyramid of histogram of orientation gradients method in combination with a Support Vector Machine (SVM) classifier. A method for constructing a unique human descriptor is presented. To build a descriptor using a trained cascade of regressors, 68 key points of the face are allocated. After that, the coordinates of the key points are passed to the input of the ResNet34 type neural network with the classification layers removed. At the output of the neural network, a descriptor consisting of 128 numbers is formed. Using a neural network allows you to compensate for the rotation of the face in the image to a certain extent. Euclidean distance is used to estimate the proximity of two descriptors. The accuracy of the method on marked faces in the wild (LFW) benchmark was 99.38%. The proposed method is compared with the results of the Microsoft Cognitive Services service.