Submitted by Rafael Estefano Vicentini null (rafaelvicentini@dee.feis.unesp.br) on 2017-12-14T18:01:32Z No. of bitstreams: 1 DISSERTAÇÃO-RAFAEL ESTÉFANO VICENTINI.pdf: 15039479 bytes, checksum: 43a2765c1d39e13b3435f194a64198ec (MD5) / Approved for entry into archive by Cristina Alexandra de Godoy null (cristina@adm.feis.unesp.br) on 2017-12-18T10:48:16Z (GMT) No. of bitstreams: 1 vicentini_re_me_ilha.pdf: 15039479 bytes, checksum: 43a2765c1d39e13b3435f194a64198ec (MD5) / Made available in DSpace on 2017-12-18T10:48:16Z (GMT). No. of bitstreams: 1 vicentini_re_me_ilha.pdf: 15039479 bytes, checksum: 43a2765c1d39e13b3435f194a64198ec (MD5) Previous issue date: 2017-10-20 / Na última década, o aumento da capacidade de processamento de informação em computadores e dispositivos de uso pessoal possibilitou o desenvolvimento de filtros e classificadores automatizados que operam em tempo real, aplicados em diversas áreas. No âmbito do Processamento Digital de Imagens associado às Redes Neurais Artificiais, os filtros emulam a percepção humana buscando por padrões para identificação de características de interesse. Filtros que têm por objetivo restringir o acesso a conteúdo impróprio partem da identificação de pele - principal indício de presença humana em uma imagem. Independentemente de sua complexidade e/ou robustez, caso o classificador não seja capaz de identificar as diferentes tonalidades de pele sob diferentes condições de captura, sua eficácia é prejudicada. Frente à diversificada forma de descrever uma tonalidade de pele usando diferentes espaços de cor, neste estudo foram destacados os espaços de cor RGB, YCbCr e HSV, amplamente utilizados em equipamentos de captura (por exemplo câmeras fotográficas e filmadoras digitais). A partir de exemplos apresentados durante a etapa de treinamento, as RNAs devem estar aptas para classificar as tonalidades em dois grupos distintos: pele e não pele. Dentre os espaços de cores indicados, seja utilizando ou descartando o aspecto da iluminação (critério amplamente discutido na literatura), este trabalho busca avaliar qual possui a maior taxa de detecção de pele em uma imagem. / Over the last decade, the increasing capacity of data processing in personal computers and devices could develop filters and automatic classifiers working in real time and applied in several areas. Considering Digital Image Processing and Artificial Neural Networks, these filters emulate the human perception searching for patterns to identify specific features. Filters which the main goal is to restrict the access to inappropriate content starts identifying skin tones - the main evidence of human presence in a picture. Although being complex and robust, if the classifier is not able to identify distinct skin tones under random capture conditions, the accuracy is minimal. Facing several ways on describing skin tones over different color spaces, this work uses the RGB, YCbCr and HSV color spaces which are widely applied in recording devices (photographic and digital cameras for example). Based on the examples shown during the training phase, the ANNs must be able to classify skin tones into two distinct groups: skin and non skin. Among the different color spaces used, considering or not the luminance aspect (widely discussed on papers), this work intends to evaluate which one has the highest detection accuracy to identify skin tone in such a picture.

Orientador: Anna Diva Plasencia Lotufo / Resumo: Na última década, o aumento da capacidade de processamento de informação em computadores e dispositivos de uso pessoal possibilitou o desenvolvimento de filtros e classificadores automatizados que operam em tempo real, aplicados em diversas áreas. No âmbito do Processamento Digital de Imagens associado às Redes Neurais Artificiais, os filtros emulam a percepção humana buscando por padrões para identificação de características de interesse. Filtros que têm por objetivo restringir o acesso a conteúdo impróprio partem da identificação de pele - principal indício de presença humana em uma imagem. Independentemente de sua complexidade e/ou robustez, caso o classificador não seja capaz de identificar as diferentes tonalidades de pele sob diferentes condições de captura, sua eficácia é prejudicada. Frente à diversificada forma de descrever uma tonalidade de pele usando diferentes espaços de cor, neste estudo foram destacados os espaços de cor RGB, YCbCr e HSV, amplamente utilizados em equipamentos de captura (por exemplo câmeras fotográficas e filmadoras digitais). A partir de exemplos apresentados durante a etapa de treinamento, as RNAs devem estar aptas para classificar as tonalidades em dois grupos distintos: pele e não pele. Dentre os espaços de cores indicados, seja utilizando ou descartando o aspecto da iluminação (critério amplamente discutido na literatura), este trabalho busca avaliar qual possui a maior taxa de detecção de pele em uma imagem. / Abstract: Over the last decade, the increasing capacity of data processing in personal computers and devices could develop filters and automatic classifiers working in real time and applied in several areas. Considering Digital Image Processing and Artificial Neural Networks, these filters emulate the human perception searching for patterns to identify specific features. Filters which the main goal is to restrict the access to inappropriate content starts identifying skin tones - the main evidence of human presence in a picture. Although being complex and robust, if the classifier is not able to identify distinct skin tones under random capture conditions, the accuracy is minimal. Facing several ways on describing skin tones over different color spaces, this work uses the RGB, YCbCr and HSV color spaces which are widely applied in recording devices (photographic and digital cameras for example). Based on the examples shown during the training phase, the ANNs must be able to classify skin tones into two distinct groups: skin and non skin. Among the different color spaces used, considering or not the luminance aspect (widely discussed on papers), this work intends to evaluate which one has the highest detection accuracy to identify skin tone in such a picture. / Mestre

Text in Afrikaans / Summaries in Afrikaans and English / In Afrikaans, soos in NederJands en Duits, word saamgestelde woorde aanmekaar geskryf. Nuwe woorde word dus voortdurend geskep deur woorde aanmekaar te haak Dit bemoeilik die proses van woordafkapping tydens teksprosessering, wat deesdae deur rekenaars gedoen word, aangesien die verwysingsbron gedurig verander. Daar bestaan verskeie afkappingsalgoritmes en tegnieke, maar die resultate is onbevredigend. Afrikaanse woorde met korrekte lettergreepverdeling is net die elektroniese weergawe van die handwoordeboek van die Afrikaanse Taal (HAT) onttrek. 'n Neutrale netwerk ( vorentoevoer-terugpropagering) is met sowat. 5 000 van hierdie woorde afgerig. Die neurale netwerk is verfyn deur 'n gcskikte afrigtingsalgoritme en oorfragfunksie vir die probleem asook die optimale aantal verborge lae en aantal neurone in elke laag te bepaal. Die neurale netwerk is met 5 000 nuwe woorde getoets en dit het 97,56% van moontlike posisies korrek as of geldige of ongeldige afkappingsposisies geklassifiseer. Verder is 510 woorde uit tydskrifartikels met die neurale netwerk getoets en 98,75% van moontlike posisies is korrek geklassifiseer. / In Afrikaans, like in Dutch and German, compound words are written as one word. New words are therefore created by simply joining words. Word hyphenation during typesetting by computer is a problem, because the source of reference changes all the time. Several algorithms and techniques for hyphenation exist, but results are not satisfactory. Afrikaans words with correct syllabification were extracted from the electronic version of the Handwoordeboek van die Afrikaans Taal (HAT). A neural network (feedforward backpropagation) was trained with about 5 000 of these words. The neural network was refined by heuristically finding a suitable training algorithm and transfer function for the problem as well as determining the optimal number of layers and number of neurons in each layer. The neural network was tested with 5 000 words not the training data. It classified 97,56% of possible points in these words correctly as either valid or invalid hyphenation points. Furthermore, 510 words from articles in a magazine were tested with the neural network and 98,75% of possible positions were classified correctly. / Computing / M.Sc. (Operasionele Navorsing)

Text in Afrikaans / Summaries in Afrikaans and English / In Afrikaans, soos in NederJands en Duits, word saamgestelde woorde aanmekaar geskryf. Nuwe woorde word dus voortdurend geskep deur woorde aanmekaar te haak Dit bemoeilik die proses van woordafkapping tydens teksprosessering, wat deesdae deur rekenaars gedoen word, aangesien die verwysingsbron gedurig verander. Daar bestaan verskeie afkappingsalgoritmes en tegnieke, maar die resultate is onbevredigend. Afrikaanse woorde met korrekte lettergreepverdeling is net die elektroniese weergawe van die handwoordeboek van die Afrikaanse Taal (HAT) onttrek. 'n Neutrale netwerk ( vorentoevoer-terugpropagering) is met sowat. 5 000 van hierdie woorde afgerig. Die neurale netwerk is verfyn deur 'n gcskikte afrigtingsalgoritme en oorfragfunksie vir die probleem asook die optimale aantal verborge lae en aantal neurone in elke laag te bepaal. Die neurale netwerk is met 5 000 nuwe woorde getoets en dit het 97,56% van moontlike posisies korrek as of geldige of ongeldige afkappingsposisies geklassifiseer. Verder is 510 woorde uit tydskrifartikels met die neurale netwerk getoets en 98,75% van moontlike posisies is korrek geklassifiseer. / In Afrikaans, like in Dutch and German, compound words are written as one word. New words are therefore created by simply joining words. Word hyphenation during typesetting by computer is a problem, because the source of reference changes all the time. Several algorithms and techniques for hyphenation exist, but results are not satisfactory. Afrikaans words with correct syllabification were extracted from the electronic version of the Handwoordeboek van die Afrikaans Taal (HAT). A neural network (feedforward backpropagation) was trained with about 5 000 of these words. The neural network was refined by heuristically finding a suitable training algorithm and transfer function for the problem as well as determining the optimal number of layers and number of neurons in each layer. The neural network was tested with 5 000 words not the training data. It classified 97,56% of possible points in these words correctly as either valid or invalid hyphenation points. Furthermore, 510 words from articles in a magazine were tested with the neural network and 98,75% of possible positions were classified correctly. / Computing / M.Sc. (Operasionele Navorsing)