A Novel Deep Network Architecture for Reconstructing RGB Facial Images from Thermal for Face Recognition

Andre Litvin, Kamal Nasrollahi, Cagri Ozcinar, Sergio Escalera Guerrero, Thomas B. Moeslund, Gholamreza Anbarjafari

Publikation: Bidrag til tidsskriftTidsskriftartikelForskningpeer review

2 Citationer (Scopus)

Resumé

This work proposes a fully convolutional network architecture for RGB face image generation from a given input thermal face image to be applied in face recognition scenarios. The proposed method is based on the FusionNet architecture and increases robustness against overfitting using dropout after bridge connections, randomised leaky ReLUs (RReLUs), and orthogonal regularization. Furthermore, we propose to use a decoding block with resize convolution instead of transposed convolution to improve final RGB face image generation. To validate our proposed network architecture, we train a face classifier and compare its face recognition rate on the reconstructed RGB images from the proposed architecture, to those when reconstructing images with the original FusionNet, as well as when using the original RGB images. As a result, we are introducing a new architecture which leads to a more accurate network.
OriginalsprogEngelsk
TidsskriftMultimedia Tools and Applications
Vol/bind78
Udgave nummer18
Sider (fra-til)25259-25271
Antal sider13
ISSN1380-7501
DOI
StatusUdgivet - 30 sep. 2019

Fingerprint

Face recognition
Network architecture
Convolution
Decoding
Classifiers
Hot Temperature

Citer dette

@article{7eac05a2cb1c42698c75bd2ac727e0cc,
title = "A Novel Deep Network Architecture for Reconstructing RGB Facial Images from Thermal for Face Recognition",
abstract = "This work proposes a fully convolutional network architecture for RGB face image generation from a given input thermal face image to be applied in face recognition scenarios. The proposed method is based on the FusionNet architecture and increases robustness against overfitting using dropout after bridge connections, randomised leaky ReLUs (RReLUs), and orthogonal regularization. Furthermore, we propose to use a decoding block with resize convolution instead of transposed convolution to improve final RGB face image generation. To validate our proposed network architecture, we train a face classifier and compare its face recognition rate on the reconstructed RGB images from the proposed architecture, to those when reconstructing images with the original FusionNet, as well as when using the original RGB images. As a result, we are introducing a new architecture which leads to a more accurate network.",
keywords = "Face recognition, Fully convolutional networks, FusionNet, Thermal imaging",
author = "Andre Litvin and Kamal Nasrollahi and Cagri Ozcinar and Guerrero, {Sergio Escalera} and Moeslund, {Thomas B.} and Gholamreza Anbarjafari",
year = "2019",
month = "9",
day = "30",
doi = "10.1007/s11042-019-7667-4",
language = "English",
volume = "78",
pages = "25259--25271",
journal = "Multimedia Tools and Applications",
issn = "1380-7501",
publisher = "Springer",
number = "18",

}

A Novel Deep Network Architecture for Reconstructing RGB Facial Images from Thermal for Face Recognition. / Litvin, Andre; Nasrollahi, Kamal; Ozcinar, Cagri; Guerrero, Sergio Escalera; Moeslund, Thomas B.; Anbarjafari, Gholamreza.

I: Multimedia Tools and Applications, Bind 78, Nr. 18, 30.09.2019, s. 25259-25271.

Publikation: Bidrag til tidsskriftTidsskriftartikelForskningpeer review

TY - JOUR

T1 - A Novel Deep Network Architecture for Reconstructing RGB Facial Images from Thermal for Face Recognition

AU - Litvin, Andre

AU - Nasrollahi, Kamal

AU - Ozcinar, Cagri

AU - Guerrero, Sergio Escalera

AU - Moeslund, Thomas B.

AU - Anbarjafari, Gholamreza

PY - 2019/9/30

Y1 - 2019/9/30

N2 - This work proposes a fully convolutional network architecture for RGB face image generation from a given input thermal face image to be applied in face recognition scenarios. The proposed method is based on the FusionNet architecture and increases robustness against overfitting using dropout after bridge connections, randomised leaky ReLUs (RReLUs), and orthogonal regularization. Furthermore, we propose to use a decoding block with resize convolution instead of transposed convolution to improve final RGB face image generation. To validate our proposed network architecture, we train a face classifier and compare its face recognition rate on the reconstructed RGB images from the proposed architecture, to those when reconstructing images with the original FusionNet, as well as when using the original RGB images. As a result, we are introducing a new architecture which leads to a more accurate network.

AB - This work proposes a fully convolutional network architecture for RGB face image generation from a given input thermal face image to be applied in face recognition scenarios. The proposed method is based on the FusionNet architecture and increases robustness against overfitting using dropout after bridge connections, randomised leaky ReLUs (RReLUs), and orthogonal regularization. Furthermore, we propose to use a decoding block with resize convolution instead of transposed convolution to improve final RGB face image generation. To validate our proposed network architecture, we train a face classifier and compare its face recognition rate on the reconstructed RGB images from the proposed architecture, to those when reconstructing images with the original FusionNet, as well as when using the original RGB images. As a result, we are introducing a new architecture which leads to a more accurate network.

KW - Face recognition

KW - Fully convolutional networks

KW - FusionNet

KW - Thermal imaging

UR - http://www.scopus.com/inward/record.url?scp=85066282792&partnerID=8YFLogxK

U2 - 10.1007/s11042-019-7667-4

DO - 10.1007/s11042-019-7667-4

M3 - Journal article

VL - 78

SP - 25259

EP - 25271

JO - Multimedia Tools and Applications

JF - Multimedia Tools and Applications

SN - 1380-7501

IS - 18

ER -