Graduate School and Research Center in Digital Sciences

Cross-spectrum thermal to visible face recognition based on cascaded image synthesis

Mallat, Khawla; Damer, Naser; Boutros, Fadi; Kuijper, Arjan; Dugelay, Jean-Luc

ICB 2019, 12th IAPR International Conference On Biometrics, 4-7 June, Crete, Greece

Face synthesis from thermal to visible spectrum is fundamental to perform cross-spectrum face recognition as it simplifies its integration in existing commercial face recognition systems and enables manual face verification. In this paper, a new solution based on cascaded refinement networks is proposed. This method generates visible-like colored images of high visual quality without requiring large amounts of training data. By employing a contextual loss function during training, the proposed network is inherently scale and rotation invariant. We discuss the visual perception of the generated visible-like faces in comparison with recent works. We also provide an objective evaluation in terms of cross-spectrum face recognition, where the generated faces were compared against a gallery in visible spectrum using two state-of-the-art deep learning based face recognition systems. When compared to the recently published TV-GAN solution, the performance of the face recognition systems, OpenFace and LightCNN, was improved by a 42.48% (i.e. from 10.76% to 15.37%) and a 71.43% (i.e. from 33.606% to 57.612%), respectively.

Document Bibtex

Title:Cross-spectrum thermal to visible face recognition based on cascaded image synthesis
Department:Digital Security
Eurecom ref:5920
Copyright: IAPR
Bibtex: @inproceedings{EURECOM+5920, year = {2019}, title = {{C}ross-spectrum thermal to visible face recognition based on cascaded image synthesis}, author = {{M}allat, {K}hawla and {D}amer, {N}aser and {B}outros, {F}adi and {K}uijper, {A}rjan and {D}ugelay, {J}ean-{L}uc}, booktitle = {{ICB} 2019, 12th {IAPR} {I}nternational {C}onference {O}n {B}iometrics, 4-7 {J}une, {C}rete, {G}reece}, address = {{C}rete, {GREECE}}, month = {06}, url = {} }
See also: