Abordagens neurais para controle de conteúdo pornográfico
Autor(a) principal: | |
---|---|
Data de Publicação: | 2019 |
Tipo de documento: | Tese |
Idioma: | por |
Título da fonte: | Biblioteca Digital de Teses e Dissertações da PUC_RS |
Texto Completo: | http://tede2.pucrs.br/tede2/handle/tede/9162 |
Resumo: | The adult content available on the internet generates health problems and behavioral disorders. The consumption of pornography is favored by the ease of access, low cost and anonymity of Internet users. Breaking at last one of these factors can minimize the consumption of this content, however, given the volume, it is necessary to analyze the content automatically. In this sense, Deep Learning can perform complex tasks automatically. This thesis attacks the ease of access to pornography by applying automatic censorship through 3 Deep Learning approaches: classification, object detection and automatic generation. In the classification approach, 8 predictive models of different neural network architectures were trained and evaluated, where the predictive results reached accuracy above 99%, processing up to 40 FPS. It was observed that the most significant regions for pornography classification are related to the intimate body parts. The second approach censored pornography with object detection methods. An intimate body parts detection dataset was constructed which allowed the training of models for censoring intimate body parts that achieved mAP = 0.6961. A neural network for detection, called CensorNet, was built, generating promising predictive results. We build CensorPlus, a network composed by a second output for classification. This network creates a hybrid method for object detection and image classification. Finally, the third approach to this thesis presents AttGAN, a method based on image-to-image translation that uses neural networks to generate automatic censorship. The method utilizes attention masks generated by AttNET, a classification-trained neural network converted to generate such masks. Three AttGAN variations were developed, and we designed an online survey where 21 participants compared the results. The results indicate an advantage to the AttGAN+ method, pointed as the best method in 1, 050 opinions collected. The AttGAN+ method was incremented by merging the input image with the censored output, giving rise to the AttGAN++ method, resulting in a censored image that preserves peripheral characteristics of the original image. |
id |
P_RS_771efecc556b1ed5c2b67c575f272803 |
---|---|
oai_identifier_str |
oai:tede2.pucrs.br:tede/9162 |
network_acronym_str |
P_RS |
network_name_str |
Biblioteca Digital de Teses e Dissertações da PUC_RS |
repository_id_str |
|
spelling |
Barros, Rodrigo Coelhohttp://lattes.cnpq.br/8172124241767828http://lattes.cnpq.br/1923513996340726Simões, Gabriel da Silva2020-07-16T19:39:54Z2019-11-08http://tede2.pucrs.br/tede2/handle/tede/9162The adult content available on the internet generates health problems and behavioral disorders. The consumption of pornography is favored by the ease of access, low cost and anonymity of Internet users. Breaking at last one of these factors can minimize the consumption of this content, however, given the volume, it is necessary to analyze the content automatically. In this sense, Deep Learning can perform complex tasks automatically. This thesis attacks the ease of access to pornography by applying automatic censorship through 3 Deep Learning approaches: classification, object detection and automatic generation. In the classification approach, 8 predictive models of different neural network architectures were trained and evaluated, where the predictive results reached accuracy above 99%, processing up to 40 FPS. It was observed that the most significant regions for pornography classification are related to the intimate body parts. The second approach censored pornography with object detection methods. An intimate body parts detection dataset was constructed which allowed the training of models for censoring intimate body parts that achieved mAP = 0.6961. A neural network for detection, called CensorNet, was built, generating promising predictive results. We build CensorPlus, a network composed by a second output for classification. This network creates a hybrid method for object detection and image classification. Finally, the third approach to this thesis presents AttGAN, a method based on image-to-image translation that uses neural networks to generate automatic censorship. The method utilizes attention masks generated by AttNET, a classification-trained neural network converted to generate such masks. Three AttGAN variations were developed, and we designed an online survey where 21 participants compared the results. The results indicate an advantage to the AttGAN+ method, pointed as the best method in 1, 050 opinions collected. The AttGAN+ method was incremented by merging the input image with the censored output, giving rise to the AttGAN++ method, resulting in a censored image that preserves peripheral characteristics of the original image.O crescente volume de conteúdo adulto disponível na internet gera problemas de saúde e desordens comportamentais. O consumo de pornografia é favorecido pela facilidade de acesso, pelo baixo custo e pela anonimidade dos internautas. Quebrando um destes fatores, pode-se minimizar o consumo deste tipo de conteúdo, por outro lado, dado o volume, é necessário analisar o conteúdo automaticamente. Neste sentido, Deep Learning permite realizar tarefas complexas automaticamente. Esta tese ataca a facilidade de acesso à pornografia aplicando censuras automáticas através de 3 abordagens de Deep Learning: classificação, detecção de objetos e geração automática. Na abordagem de classificação, foram treinados e avaliados 8 modelos preditivos com diferentes arquiteturas de redes neurais, onde os resultados preditivos atingiram acurácias superiores a 99%, processando até 40 FPS. Observou-se que as regiões mais significativas para classificação de pornografia estão relacionadas especificamente às partes íntimas do corpo. A segunda abordagem censurou pornografia utilizando métodos de detecção de objetos. Foi construído um dataset para detecção de partes íntimas que permitiu o treinamento de modelos que atingiram resultados preditivos com mAP = 0, 6961, censurando partes íntimas de corpo. Foi construída uma rede neural para detecção, chamada CensorNet, gerando resultados preditivos promissores. Foi construída também CensorPlus, uma rede composta por uma segunda saída para classificação, criando um método híbrido para detecção de objetos e classificação de imagem. Finalmente, a terceira abordagem desta tese apresenta AttGAN, um método baseado em tradução imagem-para-imagem que utiliza redes neurais para gerar censuras automáticas. O método utiliza máscaras de atenção geradas por AttNET, uma rede neural treinada para classificação, convertida para a geração de tais máscaras. Foram desenvolvidas 3 variações de AttGAN, comparadas por meio de uma avaliação online onde 21 participantes compararam os resultados. Os resultados evidenciaram vantagem para o método AttGAN+, escolhido como melhor método em 1.050 opiniões coletadas. O método AttGAN+ foi incrementado, aplicando a mescla da imagem de entrada com a saída censurada, dando origem ao método AttGAN++, resultando em uma imagem censurada que preserva características periféricas da imagem original.Submitted by PPG Ciência da Computação (ppgcc@pucrs.br) on 2020-05-06T16:01:52Z No. of bitstreams: 1 GABRIEL DA SILVA SIMOES_TES.pdf: 7740163 bytes, checksum: e7ebcecb933a5b70bf7876b1566cbbff (MD5)Approved for entry into archive by Lucas Martins Kern (lucas.kern@pucrs.br) on 2020-07-16T19:13:56Z (GMT) No. of bitstreams: 1 GABRIEL DA SILVA SIMOES_TES.pdf: 7740163 bytes, checksum: e7ebcecb933a5b70bf7876b1566cbbff (MD5)Made available in DSpace on 2020-07-16T19:39:54Z (GMT). No. of bitstreams: 1 GABRIEL DA SILVA SIMOES_TES.pdf: 7740163 bytes, checksum: e7ebcecb933a5b70bf7876b1566cbbff (MD5) Previous issue date: 2019-11-08application/pdfhttp://tede2.pucrs.br:80/tede2/retrieve/178303/GABRIEL%20DA%20SILVA%20SIMOES_TES.pdf.jpgporPontifícia Universidade Católica do Rio Grande do SulPrograma de Pós-Graduação em Ciência da ComputaçãoPUCRSBrasilEscola PolitécnicaNeural networksconvolutional neural networksclassificationdetectionpornographicensorshipRedes neuraisRedes convolucionaisClassificaçãoDetecçãoGeração automáticaCensuraPornografiaCIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAOAbordagens neurais para controle de conteúdo pornográficoinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisTrabalho não apresenta restrição para publicação-4570527706994352458500500-862078257083325301info:eu-repo/semantics/openAccessreponame:Biblioteca Digital de Teses e Dissertações da PUC_RSinstname:Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS)instacron:PUC_RSTHUMBNAILGABRIEL DA SILVA SIMOES_TES.pdf.jpgGABRIEL DA SILVA SIMOES_TES.pdf.jpgimage/jpeg4996http://tede2.pucrs.br/tede2/bitstream/tede/9162/4/GABRIEL+DA+SILVA+SIMOES_TES.pdf.jpg537897edc2f2e3dfd02ad0f1583cf752MD54TEXTGABRIEL DA SILVA SIMOES_TES.pdf.txtGABRIEL DA SILVA SIMOES_TES.pdf.txttext/plain281744http://tede2.pucrs.br/tede2/bitstream/tede/9162/3/GABRIEL+DA+SILVA+SIMOES_TES.pdf.txtfc11eb37907f70e10481066c3ea472caMD53ORIGINALGABRIEL DA SILVA SIMOES_TES.pdfGABRIEL DA SILVA SIMOES_TES.pdfapplication/pdf7740163http://tede2.pucrs.br/tede2/bitstream/tede/9162/2/GABRIEL+DA+SILVA+SIMOES_TES.pdfe7ebcecb933a5b70bf7876b1566cbbffMD52LICENSElicense.txtlicense.txttext/plain; charset=utf-8590http://tede2.pucrs.br/tede2/bitstream/tede/9162/1/license.txt220e11f2d3ba5354f917c7035aadef24MD51tede/91622020-07-16 20:01:26.183oai:tede2.pucrs.br:tede/9162QXV0b3JpemE/P28gcGFyYSBQdWJsaWNhPz9vIEVsZXRyP25pY2E6IENvbSBiYXNlIG5vIGRpc3Bvc3RvIG5hIExlaSBGZWRlcmFsIG4/OS42MTAsIGRlIDE5IGRlIGZldmVyZWlybyBkZSAxOTk4LCBvIGF1dG9yIEFVVE9SSVpBIGEgcHVibGljYT8/byBlbGV0cj9uaWNhIGRhIHByZXNlbnRlIG9icmEgbm8gYWNlcnZvIGRhIEJpYmxpb3RlY2EgRGlnaXRhbCBkYSBQb250aWY/Y2lhIFVuaXZlcnNpZGFkZSBDYXQ/bGljYSBkbyBSaW8gR3JhbmRlIGRvIFN1bCwgc2VkaWFkYSBhIEF2LiBJcGlyYW5nYSA2NjgxLCBQb3J0byBBbGVncmUsIFJpbyBHcmFuZGUgZG8gU3VsLCBjb20gcmVnaXN0cm8gZGUgQ05QSiA4ODYzMDQxMzAwMDItODEgYmVtIGNvbW8gZW0gb3V0cmFzIGJpYmxpb3RlY2FzIGRpZ2l0YWlzLCBuYWNpb25haXMgZSBpbnRlcm5hY2lvbmFpcywgY29ucz9yY2lvcyBlIHJlZGVzID9zIHF1YWlzIGEgYmlibGlvdGVjYSBkYSBQVUNSUyBwb3NzYSBhIHZpciBwYXJ0aWNpcGFyLCBzZW0gP251cyBhbHVzaXZvIGFvcyBkaXJlaXRvcyBhdXRvcmFpcywgYSB0P3R1bG8gZGUgZGl2dWxnYT8/byBkYSBwcm9kdT8/byBjaWVudD9maWNhLgo=Biblioteca Digital de Teses e Dissertaçõeshttp://tede2.pucrs.br/tede2/PRIhttps://tede2.pucrs.br/oai/requestbiblioteca.central@pucrs.br||opendoar:2020-07-16T23:01:26Biblioteca Digital de Teses e Dissertações da PUC_RS - Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS)false |
dc.title.por.fl_str_mv |
Abordagens neurais para controle de conteúdo pornográfico |
title |
Abordagens neurais para controle de conteúdo pornográfico |
spellingShingle |
Abordagens neurais para controle de conteúdo pornográfico Simões, Gabriel da Silva Neural networks convolutional neural networks classification detection pornographi censorship Redes neurais Redes convolucionais Classificação Detecção Geração automática Censura Pornografia CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO |
title_short |
Abordagens neurais para controle de conteúdo pornográfico |
title_full |
Abordagens neurais para controle de conteúdo pornográfico |
title_fullStr |
Abordagens neurais para controle de conteúdo pornográfico |
title_full_unstemmed |
Abordagens neurais para controle de conteúdo pornográfico |
title_sort |
Abordagens neurais para controle de conteúdo pornográfico |
author |
Simões, Gabriel da Silva |
author_facet |
Simões, Gabriel da Silva |
author_role |
author |
dc.contributor.advisor1.fl_str_mv |
Barros, Rodrigo Coelho |
dc.contributor.advisor1Lattes.fl_str_mv |
http://lattes.cnpq.br/8172124241767828 |
dc.contributor.authorLattes.fl_str_mv |
http://lattes.cnpq.br/1923513996340726 |
dc.contributor.author.fl_str_mv |
Simões, Gabriel da Silva |
contributor_str_mv |
Barros, Rodrigo Coelho |
dc.subject.eng.fl_str_mv |
Neural networks convolutional neural networks classification detection pornographi censorship |
topic |
Neural networks convolutional neural networks classification detection pornographi censorship Redes neurais Redes convolucionais Classificação Detecção Geração automática Censura Pornografia CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO |
dc.subject.por.fl_str_mv |
Redes neurais Redes convolucionais Classificação Detecção Geração automática Censura Pornografia |
dc.subject.cnpq.fl_str_mv |
CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO |
description |
The adult content available on the internet generates health problems and behavioral disorders. The consumption of pornography is favored by the ease of access, low cost and anonymity of Internet users. Breaking at last one of these factors can minimize the consumption of this content, however, given the volume, it is necessary to analyze the content automatically. In this sense, Deep Learning can perform complex tasks automatically. This thesis attacks the ease of access to pornography by applying automatic censorship through 3 Deep Learning approaches: classification, object detection and automatic generation. In the classification approach, 8 predictive models of different neural network architectures were trained and evaluated, where the predictive results reached accuracy above 99%, processing up to 40 FPS. It was observed that the most significant regions for pornography classification are related to the intimate body parts. The second approach censored pornography with object detection methods. An intimate body parts detection dataset was constructed which allowed the training of models for censoring intimate body parts that achieved mAP = 0.6961. A neural network for detection, called CensorNet, was built, generating promising predictive results. We build CensorPlus, a network composed by a second output for classification. This network creates a hybrid method for object detection and image classification. Finally, the third approach to this thesis presents AttGAN, a method based on image-to-image translation that uses neural networks to generate automatic censorship. The method utilizes attention masks generated by AttNET, a classification-trained neural network converted to generate such masks. Three AttGAN variations were developed, and we designed an online survey where 21 participants compared the results. The results indicate an advantage to the AttGAN+ method, pointed as the best method in 1, 050 opinions collected. The AttGAN+ method was incremented by merging the input image with the censored output, giving rise to the AttGAN++ method, resulting in a censored image that preserves peripheral characteristics of the original image. |
publishDate |
2019 |
dc.date.issued.fl_str_mv |
2019-11-08 |
dc.date.accessioned.fl_str_mv |
2020-07-16T19:39:54Z |
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.uri.fl_str_mv |
http://tede2.pucrs.br/tede2/handle/tede/9162 |
url |
http://tede2.pucrs.br/tede2/handle/tede/9162 |
dc.language.iso.fl_str_mv |
por |
language |
por |
dc.relation.program.fl_str_mv |
-4570527706994352458 |
dc.relation.confidence.fl_str_mv |
500 500 |
dc.relation.cnpq.fl_str_mv |
-862078257083325301 |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
eu_rights_str_mv |
openAccess |
dc.format.none.fl_str_mv |
application/pdf |
dc.publisher.none.fl_str_mv |
Pontifícia Universidade Católica do Rio Grande do Sul |
dc.publisher.program.fl_str_mv |
Programa de Pós-Graduação em Ciência da Computação |
dc.publisher.initials.fl_str_mv |
PUCRS |
dc.publisher.country.fl_str_mv |
Brasil |
dc.publisher.department.fl_str_mv |
Escola Politécnica |
publisher.none.fl_str_mv |
Pontifícia Universidade Católica do Rio Grande do Sul |
dc.source.none.fl_str_mv |
reponame:Biblioteca Digital de Teses e Dissertações da PUC_RS instname:Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS) instacron:PUC_RS |
instname_str |
Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS) |
instacron_str |
PUC_RS |
institution |
PUC_RS |
reponame_str |
Biblioteca Digital de Teses e Dissertações da PUC_RS |
collection |
Biblioteca Digital de Teses e Dissertações da PUC_RS |
bitstream.url.fl_str_mv |
http://tede2.pucrs.br/tede2/bitstream/tede/9162/4/GABRIEL+DA+SILVA+SIMOES_TES.pdf.jpg http://tede2.pucrs.br/tede2/bitstream/tede/9162/3/GABRIEL+DA+SILVA+SIMOES_TES.pdf.txt http://tede2.pucrs.br/tede2/bitstream/tede/9162/2/GABRIEL+DA+SILVA+SIMOES_TES.pdf http://tede2.pucrs.br/tede2/bitstream/tede/9162/1/license.txt |
bitstream.checksum.fl_str_mv |
537897edc2f2e3dfd02ad0f1583cf752 fc11eb37907f70e10481066c3ea472ca e7ebcecb933a5b70bf7876b1566cbbff 220e11f2d3ba5354f917c7035aadef24 |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 MD5 |
repository.name.fl_str_mv |
Biblioteca Digital de Teses e Dissertações da PUC_RS - Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS) |
repository.mail.fl_str_mv |
biblioteca.central@pucrs.br|| |
_version_ |
1799765344549601280 |