REDE NEURAL PROFUNDA DE MÚLTIPLAS ENTRADAS COM DADOS 3D APLICADOS AO RECONHECIMENTO FACIAL

Facial recognition is increasingly being used in commercial applications in a variety of scenarios, and uses 2D or 3D biometrics of the face to identify people. We propose a methodology using multi input neural network to perform facial recognition of the individuals from the 3D features extracted from frontal images. For 3D features extraction the images were at first submitted to the pix2vertex network in order to perform 3D reconstruction of the facial geometry of each individual. After the 3D reconstruction 275 points were extracted, to each point there are x, y and z coordinates, and the SIFT descriptor. Finally, this information is processed in a multiple-input artificial neural network to predict the classification of each individual. Evaluation of the results shows that the network was able to correctly classify individuals with an accuracy of 95.79% in the validation set.