-
Yves Augusto Lima Romero
Departamento de Telemática, Instituto Federal do Ceará, CE
-
Ajalmar Rêgo da Rocha Neto
Departamento de Telemática, Instituto Federal do Ceará, CE
Keywords:
deep learning, convolutional neural networks, person reid, one-shot learning, triplet loss, machine learning, biometric systems
Abstract
Recenlty, many face reid strategies have been developped. These strategies create vectorial spaces capable of representing data on reduced dimensions. Such representations are produced by deep learning models that learn to maximize the intra-class similarity and minimize the inter-class similarity. The method described in this article proposes a new face reid strategy based on the Facenet model. It consists of training support vector machines on the euclidean distances calculated between the embeddings of the Facenet model, notably reducing the false positive rate.
