Применение методов машинного обучения для решения задачи классификации эмоции на изображении по ключевым точкам

Abstract

In this paper, was considered the application of the three most popular methods of machine learning, which are used to classify images (support vector method, artificial neural network, and convolutional neural network). These methods were used to solve the problem of recognition and classification of emotions on image the face of a person. Emotions were recognized using facial landmarks (78), which were determined using the Active Appearance Model algorithm. For training and testing, the Extended CohnKanade Database (CK +) was used. The algorithm developed using convolution layers (mean about 91%) showed the best accuracy. It was also revealed that the use of convolution layers reduces the network error for the same number of training eras.