In this paper, we propose a new framework for facerecognition from depth images, which is both effective andefficient. It consists of two main stages: First, a hand-crafted low-level feature extractor is applied to the rawdepth data of the face, thus extracting the corresponding de-scriptor images (DIs); Then, a not-so-deep (shallow) con-volutional neural network (SCNN) has been designed thatlearns from the DIs. This architecture showed two main ad-vantages over the direct application of deep-CNN (DCNN)to the depth images of the face: On the one hand, the DIsare capable of enriching the raw depth data, emphasizingrelevant traits of the face, while reducing their acquisitionnoise. This resulted decisive in improving the learning ca-pability of the network; On the other, the DIs capture low-level features of the face, thus playing the role for the SCNNas the first layers do in a DCNN architecture. In this way,the SCNN we have designed has much less layers and canbe trained more easily and faster. Extensive experiments onlow- and high-resolution depth face datasets confirmed usthe above advantages, showing results that are comparableor superior to the state-of-the-art, using by far less trainingdata, time, and memory occupancy of the network

Deep Learning from 3DLBP Descriptors for Depth Image Based Face Recognition / Cardia Neto, J. B.; Marana, A. N.; Ferrari, C.; Berretti, S.; Del Bimbo, A.. - STAMPA. - (2019), pp. 1-1. (Intervento presentato al convegno IAPR International Conference on Biometrics tenutosi a Crete. Greece nel 4-7 June, 2019).

Deep Learning from 3DLBP Descriptors for Depth Image Based Face Recognition

C. Ferrari;
2019-01-01

Abstract

In this paper, we propose a new framework for facerecognition from depth images, which is both effective andefficient. It consists of two main stages: First, a hand-crafted low-level feature extractor is applied to the rawdepth data of the face, thus extracting the corresponding de-scriptor images (DIs); Then, a not-so-deep (shallow) con-volutional neural network (SCNN) has been designed thatlearns from the DIs. This architecture showed two main ad-vantages over the direct application of deep-CNN (DCNN)to the depth images of the face: On the one hand, the DIsare capable of enriching the raw depth data, emphasizingrelevant traits of the face, while reducing their acquisitionnoise. This resulted decisive in improving the learning ca-pability of the network; On the other, the DIs capture low-level features of the face, thus playing the role for the SCNNas the first layers do in a DCNN architecture. In this way,the SCNN we have designed has much less layers and canbe trained more easily and faster. Extensive experiments onlow- and high-resolution depth face datasets confirmed usthe above advantages, showing results that are comparableor superior to the state-of-the-art, using by far less trainingdata, time, and memory occupancy of the network
2019
Deep Learning from 3DLBP Descriptors for Depth Image Based Face Recognition / Cardia Neto, J. B.; Marana, A. N.; Ferrari, C.; Berretti, S.; Del Bimbo, A.. - STAMPA. - (2019), pp. 1-1. (Intervento presentato al convegno IAPR International Conference on Biometrics tenutosi a Crete. Greece nel 4-7 June, 2019).
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11381/2900766
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 3
  • ???jsp.display-item.citation.isi??? 1
social impact