3D face recognition with descriptor images and shallow convolutional neural networks
Autor(a) principal: | |
---|---|
Data de Publicação: | 2020 |
Tipo de documento: | Tese |
Idioma: | eng |
Título da fonte: | Repositório Institucional da UFSCAR |
Texto Completo: | https://repositorio.ufscar.br/handle/ufscar/13463 |
Resumo: | Nowadays, there is an increasing need for systems that can accurately and quickly identify a person. Traditional identification methods utilize something a person knows or something a person has. This kind of methods has several drawbacks, being the main one the fact that it is impossible to detect an impostor who uses genuine credentials to pass as a genuine person. Besides, in some cases it is necessary to discover the identity of people in a covert manner. One way to deal with these types of problems is to use biometric identification. Face is one of the biometric features that best suit the covert identification since the current technology is able to provide high resolution 2D face images captured by low cost cameras, in a secret way, at a distance and without cooperation from the people being identified. However, in general, biometric systems based on 2D face recognition perform very poorly in certain scenarios when the input images present variations in pose, illumination, and facial expressions. One way to mitigate this problem is to use 3D face data, but the current 3D scanners are expensive and require a lot of cooperation from people. The use of deep convolutional neural networks is another way to mitigate the traditional 2D facial recognition drawbacks, but it can be unfeasible, due to their large training data and huge computational power requirements. Therefore, in this thesis, we introduce a hybrid approach, based on Shallow Learned Feature Representation, for 3D face recognition, which is focused on minimizing the amount of data, the computational power and the processing time required in the training stage, while being able to operate close to state-of-the-art methods and being able to transfer the learning made on high-resolution data to low-resolution data. Another important aspect of the proposed hybrid approach is the possibility to operate in both classification or feature-extraction modes. Experimental results obtained by our hybrid approach on EURECOM Kinect Face dataset, a low resolution depth dataset, showed a rank-1 recognition rate of 90.75% on the hardest case of classification mode, and 73.26% on the feature extraction mode, which are better than the rates obtained by related state-of-the-art methods with the same protocol and dataset. So, we conclude that the proposed hybrid approach helps to attenuate the cross-resolution differences and that the utilization of an input built with more discriminative data, such as low-level hand-crafted features, allows the utilization of shallow CNN for 3D face recognition. |
id |
SCAR_d0b20d4b806e63cebadea72470277514 |
---|---|
oai_identifier_str |
oai:repositorio.ufscar.br:ufscar/13463 |
network_acronym_str |
SCAR |
network_name_str |
Repositório Institucional da UFSCAR |
repository_id_str |
4322 |
spelling |
Cardia Neto, João BaptistaMarana, Aparecido Nilceuhttp://lattes.cnpq.br/6027713750942689Berretti, Stefanohttp://lattes.cnpq.br/6844092194166622b2a56b98-daee-4d3f-8f45-bc17767638802020-11-18T09:06:41Z2020-11-18T09:06:41Z2020-11-05CARDIA NETO, João Baptista. 3D face recognition with descriptor images and shallow convolutional neural networks. 2020. Tese (Doutorado em Ciência da Computação) – Universidade Federal de São Carlos, São Carlos, 2020. Disponível em: https://repositorio.ufscar.br/handle/ufscar/13463.https://repositorio.ufscar.br/handle/ufscar/13463Nowadays, there is an increasing need for systems that can accurately and quickly identify a person. Traditional identification methods utilize something a person knows or something a person has. This kind of methods has several drawbacks, being the main one the fact that it is impossible to detect an impostor who uses genuine credentials to pass as a genuine person. Besides, in some cases it is necessary to discover the identity of people in a covert manner. One way to deal with these types of problems is to use biometric identification. Face is one of the biometric features that best suit the covert identification since the current technology is able to provide high resolution 2D face images captured by low cost cameras, in a secret way, at a distance and without cooperation from the people being identified. However, in general, biometric systems based on 2D face recognition perform very poorly in certain scenarios when the input images present variations in pose, illumination, and facial expressions. One way to mitigate this problem is to use 3D face data, but the current 3D scanners are expensive and require a lot of cooperation from people. The use of deep convolutional neural networks is another way to mitigate the traditional 2D facial recognition drawbacks, but it can be unfeasible, due to their large training data and huge computational power requirements. Therefore, in this thesis, we introduce a hybrid approach, based on Shallow Learned Feature Representation, for 3D face recognition, which is focused on minimizing the amount of data, the computational power and the processing time required in the training stage, while being able to operate close to state-of-the-art methods and being able to transfer the learning made on high-resolution data to low-resolution data. Another important aspect of the proposed hybrid approach is the possibility to operate in both classification or feature-extraction modes. Experimental results obtained by our hybrid approach on EURECOM Kinect Face dataset, a low resolution depth dataset, showed a rank-1 recognition rate of 90.75% on the hardest case of classification mode, and 73.26% on the feature extraction mode, which are better than the rates obtained by related state-of-the-art methods with the same protocol and dataset. So, we conclude that the proposed hybrid approach helps to attenuate the cross-resolution differences and that the utilization of an input built with more discriminative data, such as low-level hand-crafted features, allows the utilization of shallow CNN for 3D face recognition.A crescente necessidade de sistemas que possam identificar uma pessoa com precisão e rapidez se torna muito evidente nos dias de hoje. Existem algumas aplicações em que a necessidade de descobrir a identidade das pessoas de forma sigilosa é primordial. Pensando nessas aplicações e na utilização de características biométricas a face é uma das características que melhor se adequa a esse tipo de identificação. Isso pois a tecnologia atual é capaz de fornecer imagens faciais 2D de alta resolução capturadas por câmeras de baixo custo a distância e sem a cooperação dos sujeitos. No entanto, em geral, os sistemas biométricos baseados no reconhecimento de face 2D têm sua performance afetada em certos cenários, quando as imagens das faces apresentam variações na pose, iluminação e expressões faciais. Uma maneira de atenuar esse problema é usar dados faciais em 3D, mas os scanners 3D atuais são caros e exigem muita cooperação dos sujeitos. O uso de redes neurais convolucionais profundas é outra forma de mitigar as desvantagens do reconhecimento facial 2D tradicional, mas pode ser inviável, devido à necessidade de grandes conjuntos de dados rotulados para o treinamento das redes e computadores com enorme capacidade de processamento e armazenamento de dados. Portanto, nesta tese, uma abordagem híbrida para reconhecimento de faces 3D é apresentada. Essa abordagem, que tem como foco minimizar a quantidade de dados, o poder computacional e o tempo de processamento necessário na fase de treinamento, é baseada em redes neurais convolucionais rasas e é capaz operar próximo aos métodos do estado da arte e ser capaz de transferir o aprendizado feito em dados de alta resolução para dados de baixa resolução. Outro aspecto importante da abordagem híbrida proposta é a possibilidade de operar nos modos de classificação ou extração de características. Os resultados experimentais obtidos por nossa abordagem híbrida utilizando o dataset EURECOM Kinect Face, com dados de profundidade de baixa resolução, mostraram uma taxa de reconhecimento em rank-1 de 90,75% no caso mais difícil do modo de classificação e 73,26% no modo de extração de características, sendo o desempenho melhor que outras técnicas utilizando o mesmo protocolo e conjunto de dados. Assim, concluímos que a abordagem híbrida proposta ajuda a atenuar as diferenças de resolução e que a utilização de uma entrada construída com dados mais discriminativos, como extratores de característica de baixo nível, permite a utilização de CNN rasas para reconhecimento facial 3D.Não recebi financiamentoengUniversidade Federal de São CarlosCâmpus São CarlosPrograma de Pós-Graduação em Ciência da Computação - PPGCCUFSCarAttribution-NonCommercial-NoDerivs 3.0 Brazilhttp://creativecommons.org/licenses/by-nc-nd/3.0/br/info:eu-repo/semantics/openAccessBiometriaReconhecimento de faces 3DCNN rasas3DLBPSigmoid 3DLBPDescriptor ImageShallow Learned Feature Representation (SLFR)Biometrics3D face recognitionShallow CNNCIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO3D face recognition with descriptor images and shallow convolutional neural networksReconhecimento de faces 3D com descriptor images e redes neurais convolucionais rasasinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesis6006007130220c-6ef2-41e9-bc45-cc368a9c6597reponame:Repositório Institucional da UFSCARinstname:Universidade Federal de São Carlos (UFSCAR)instacron:UFSCARORIGINALTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdfTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdfTexto final da teseapplication/pdf7390949https://repositorio.ufscar.br/bitstream/ufscar/13463/1/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdfac6ce78a6e481e215c9e1469f76ea571MD51Copy of PPGCC_Template_dec_BCO (1).pdfCopy of PPGCC_Template_dec_BCO (1).pdfCarta Comprovante assinada pelo Orientadorapplication/pdf119138https://repositorio.ufscar.br/bitstream/ufscar/13463/2/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf1a66645407ec0b6aa895d55d83c4d2f9MD52CC-LICENSElicense_rdflicense_rdfapplication/rdf+xml; charset=utf-8811https://repositorio.ufscar.br/bitstream/ufscar/13463/3/license_rdfe39d27027a6cc9cb039ad269a5db8e34MD53TEXTTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.txtTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.txtExtracted texttext/plain158244https://repositorio.ufscar.br/bitstream/ufscar/13463/4/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.txte872c614ee160df87565070ef5012d1cMD54Copy of PPGCC_Template_dec_BCO (1).pdf.txtCopy of PPGCC_Template_dec_BCO (1).pdf.txtExtracted texttext/plain1637https://repositorio.ufscar.br/bitstream/ufscar/13463/6/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf.txt90289f1839b4d9880acd414661377294MD56THUMBNAILTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.jpgTese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.jpgIM Thumbnailimage/jpeg8262https://repositorio.ufscar.br/bitstream/ufscar/13463/5/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.jpg351a0e42e8ae642ec0074f244e684e03MD55Copy of PPGCC_Template_dec_BCO (1).pdf.jpgCopy of PPGCC_Template_dec_BCO (1).pdf.jpgIM Thumbnailimage/jpeg13148https://repositorio.ufscar.br/bitstream/ufscar/13463/7/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf.jpgd1965b82aebe2e266c7cac3bb91fe589MD57ufscar/134632023-09-18 18:32:03.925oai:repositorio.ufscar.br:ufscar/13463Repositório InstitucionalPUBhttps://repositorio.ufscar.br/oai/requestopendoar:43222023-09-18T18:32:03Repositório Institucional da UFSCAR - Universidade Federal de São Carlos (UFSCAR)false |
dc.title.eng.fl_str_mv |
3D face recognition with descriptor images and shallow convolutional neural networks |
dc.title.alternative.por.fl_str_mv |
Reconhecimento de faces 3D com descriptor images e redes neurais convolucionais rasas |
title |
3D face recognition with descriptor images and shallow convolutional neural networks |
spellingShingle |
3D face recognition with descriptor images and shallow convolutional neural networks Cardia Neto, João Baptista Biometria Reconhecimento de faces 3D CNN rasas 3DLBP Sigmoid 3DLBP Descriptor Image Shallow Learned Feature Representation (SLFR) Biometrics 3D face recognition Shallow CNN CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
title_short |
3D face recognition with descriptor images and shallow convolutional neural networks |
title_full |
3D face recognition with descriptor images and shallow convolutional neural networks |
title_fullStr |
3D face recognition with descriptor images and shallow convolutional neural networks |
title_full_unstemmed |
3D face recognition with descriptor images and shallow convolutional neural networks |
title_sort |
3D face recognition with descriptor images and shallow convolutional neural networks |
author |
Cardia Neto, João Baptista |
author_facet |
Cardia Neto, João Baptista |
author_role |
author |
dc.contributor.authorlattes.por.fl_str_mv |
http://lattes.cnpq.br/6844092194166622 |
dc.contributor.author.fl_str_mv |
Cardia Neto, João Baptista |
dc.contributor.advisor1.fl_str_mv |
Marana, Aparecido Nilceu |
dc.contributor.advisor1Lattes.fl_str_mv |
http://lattes.cnpq.br/6027713750942689 |
dc.contributor.advisor-co1.fl_str_mv |
Berretti, Stefano |
dc.contributor.authorID.fl_str_mv |
b2a56b98-daee-4d3f-8f45-bc1776763880 |
contributor_str_mv |
Marana, Aparecido Nilceu Berretti, Stefano |
dc.subject.por.fl_str_mv |
Biometria Reconhecimento de faces 3D CNN rasas 3DLBP |
topic |
Biometria Reconhecimento de faces 3D CNN rasas 3DLBP Sigmoid 3DLBP Descriptor Image Shallow Learned Feature Representation (SLFR) Biometrics 3D face recognition Shallow CNN CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
dc.subject.eng.fl_str_mv |
Sigmoid 3DLBP Descriptor Image Shallow Learned Feature Representation (SLFR) Biometrics 3D face recognition Shallow CNN |
dc.subject.cnpq.fl_str_mv |
CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
description |
Nowadays, there is an increasing need for systems that can accurately and quickly identify a person. Traditional identification methods utilize something a person knows or something a person has. This kind of methods has several drawbacks, being the main one the fact that it is impossible to detect an impostor who uses genuine credentials to pass as a genuine person. Besides, in some cases it is necessary to discover the identity of people in a covert manner. One way to deal with these types of problems is to use biometric identification. Face is one of the biometric features that best suit the covert identification since the current technology is able to provide high resolution 2D face images captured by low cost cameras, in a secret way, at a distance and without cooperation from the people being identified. However, in general, biometric systems based on 2D face recognition perform very poorly in certain scenarios when the input images present variations in pose, illumination, and facial expressions. One way to mitigate this problem is to use 3D face data, but the current 3D scanners are expensive and require a lot of cooperation from people. The use of deep convolutional neural networks is another way to mitigate the traditional 2D facial recognition drawbacks, but it can be unfeasible, due to their large training data and huge computational power requirements. Therefore, in this thesis, we introduce a hybrid approach, based on Shallow Learned Feature Representation, for 3D face recognition, which is focused on minimizing the amount of data, the computational power and the processing time required in the training stage, while being able to operate close to state-of-the-art methods and being able to transfer the learning made on high-resolution data to low-resolution data. Another important aspect of the proposed hybrid approach is the possibility to operate in both classification or feature-extraction modes. Experimental results obtained by our hybrid approach on EURECOM Kinect Face dataset, a low resolution depth dataset, showed a rank-1 recognition rate of 90.75% on the hardest case of classification mode, and 73.26% on the feature extraction mode, which are better than the rates obtained by related state-of-the-art methods with the same protocol and dataset. So, we conclude that the proposed hybrid approach helps to attenuate the cross-resolution differences and that the utilization of an input built with more discriminative data, such as low-level hand-crafted features, allows the utilization of shallow CNN for 3D face recognition. |
publishDate |
2020 |
dc.date.accessioned.fl_str_mv |
2020-11-18T09:06:41Z |
dc.date.available.fl_str_mv |
2020-11-18T09:06:41Z |
dc.date.issued.fl_str_mv |
2020-11-05 |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/doctoralThesis |
format |
doctoralThesis |
status_str |
publishedVersion |
dc.identifier.citation.fl_str_mv |
CARDIA NETO, João Baptista. 3D face recognition with descriptor images and shallow convolutional neural networks. 2020. Tese (Doutorado em Ciência da Computação) – Universidade Federal de São Carlos, São Carlos, 2020. Disponível em: https://repositorio.ufscar.br/handle/ufscar/13463. |
dc.identifier.uri.fl_str_mv |
https://repositorio.ufscar.br/handle/ufscar/13463 |
identifier_str_mv |
CARDIA NETO, João Baptista. 3D face recognition with descriptor images and shallow convolutional neural networks. 2020. Tese (Doutorado em Ciência da Computação) – Universidade Federal de São Carlos, São Carlos, 2020. Disponível em: https://repositorio.ufscar.br/handle/ufscar/13463. |
url |
https://repositorio.ufscar.br/handle/ufscar/13463 |
dc.language.iso.fl_str_mv |
eng |
language |
eng |
dc.relation.confidence.fl_str_mv |
600 600 |
dc.relation.authority.fl_str_mv |
7130220c-6ef2-41e9-bc45-cc368a9c6597 |
dc.rights.driver.fl_str_mv |
Attribution-NonCommercial-NoDerivs 3.0 Brazil http://creativecommons.org/licenses/by-nc-nd/3.0/br/ info:eu-repo/semantics/openAccess |
rights_invalid_str_mv |
Attribution-NonCommercial-NoDerivs 3.0 Brazil http://creativecommons.org/licenses/by-nc-nd/3.0/br/ |
eu_rights_str_mv |
openAccess |
dc.publisher.none.fl_str_mv |
Universidade Federal de São Carlos Câmpus São Carlos |
dc.publisher.program.fl_str_mv |
Programa de Pós-Graduação em Ciência da Computação - PPGCC |
dc.publisher.initials.fl_str_mv |
UFSCar |
publisher.none.fl_str_mv |
Universidade Federal de São Carlos Câmpus São Carlos |
dc.source.none.fl_str_mv |
reponame:Repositório Institucional da UFSCAR instname:Universidade Federal de São Carlos (UFSCAR) instacron:UFSCAR |
instname_str |
Universidade Federal de São Carlos (UFSCAR) |
instacron_str |
UFSCAR |
institution |
UFSCAR |
reponame_str |
Repositório Institucional da UFSCAR |
collection |
Repositório Institucional da UFSCAR |
bitstream.url.fl_str_mv |
https://repositorio.ufscar.br/bitstream/ufscar/13463/1/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf https://repositorio.ufscar.br/bitstream/ufscar/13463/2/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf https://repositorio.ufscar.br/bitstream/ufscar/13463/3/license_rdf https://repositorio.ufscar.br/bitstream/ufscar/13463/4/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.txt https://repositorio.ufscar.br/bitstream/ufscar/13463/6/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf.txt https://repositorio.ufscar.br/bitstream/ufscar/13463/5/Tese_JoaoCardia_versao_Agosto_2020_versao_final_enviar.pdf.jpg https://repositorio.ufscar.br/bitstream/ufscar/13463/7/Copy%20of%20PPGCC_Template_dec_BCO%20%281%29.pdf.jpg |
bitstream.checksum.fl_str_mv |
ac6ce78a6e481e215c9e1469f76ea571 1a66645407ec0b6aa895d55d83c4d2f9 e39d27027a6cc9cb039ad269a5db8e34 e872c614ee160df87565070ef5012d1c 90289f1839b4d9880acd414661377294 351a0e42e8ae642ec0074f244e684e03 d1965b82aebe2e266c7cac3bb91fe589 |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 MD5 MD5 MD5 MD5 |
repository.name.fl_str_mv |
Repositório Institucional da UFSCAR - Universidade Federal de São Carlos (UFSCAR) |
repository.mail.fl_str_mv |
|
_version_ |
1802136382853873664 |