Strategy for the detection of types of facial gestures using SOM neural network

Raúl Eduardo H.Z.; Yensi V.L.; Edward José F.M.; Llanos Chacaltana K.S.; Mónica D.R.; Miguel Alfredo L.S. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2022) · DOI: 10.18687/leird2022.1.1.80

Resumen

This research aims to cover a need to be able to classify gestures, specifically the gestures of people's faces, which reflects the emotions of each person such as anger, fear, happiness and sadness. To be able to identify these gestures, it is necessary to apply a strategy, which is to prepare the digital image matrices in a sequence, such as converting to gray tone, finding the orientation, applying the sobel and medfilt2 algorithm, so that this result can enter to a SOM neural network and be able to be classified according to the gestures. Labeling as 0 to anger, 1 to fear, 2 to happiness and 3 to sadness. To corroborate this strategy, a public database of faces has been taken, being 160 images of faces for the training and for the tests 15 images were used that were not part of the training and each image obtained in .jpg format in different dimensions, achieving demonstrate with this strategy an affectivity of 96.0% certainty in the identification of gestures.

Instituciones: Universidad Nacional de Trujillo; Universidad Nacional Federico Villarreal; Universidad Nacional Tecnológica de Lima Sur; Universidad Privada Norbert Wiener; Universidad Privada del Norte