Detecção de evidências de pornografia infantojuvenil em imagens digitais com estratégias formadas por técnicas computacionais integradas

• Ano de defesa: 2024
• Autor(a) principal: Evangelista, João Rafael Gonçalves
• Orientador(a): Sassi, Renato José
• Banca de defesa: Sassi, Renato José , Chalco, Jesús Pascual Mena , Souza, Edson Melo de , Dias, Cleber Gustavo , Araújo, Sidnei Alves de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Nove de Julho
• Programa de Pós-Graduação: Programa de Pós-Graduação em Informática e Gestão do Conhecimento
• Departamento: Informática
• País: Brasil
• Palavras-chave em Português: pornografia infantojuvenil; conteúdo sensível; imagens digitais; técnicas computacionais; exame pericial
• Palavras-chave em Inglês: child pornography; sensitive content; digital images; computational techniques; expert examination
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://bibliotecatede.uninove.br/handle/tede/3514

Criminals use the internet to make cybercrimes, such as sharing files with child pornography. Detecting evidence of this type of crime is a task made by police authorities in an expert examination. A difficulty in detecting evidence is the variety and quantity of files present on a device to be examined. One way to increase the chances of success in the forensic examination is to use Strategies made up of integrated computational techniques from Forensic Computing, Artificial Intelligence, and Computer Vision. Thus, the objective of this work was to develop and apply Strategies made up of integrated computational techniques from the areas of Forensic Computing, Artificial Intelligence, and Computer Vision applied to the detection of evidence of child pornography in digital images, to support the execution of expert examinations. The name Fenrir was given to all these Strategies. Four Strategies were developed: Detection and recovery of Perceptual Hash values (A), Detection of people (B), Detection of textual content related to child pornography (C), and Detection of objects related to child pornography (D). The results obtained with the development and application of the Strategies were considered promising because they achieved the objective proposed for each Strategy and, consequently, the general objective. Strategy A formed by the Perceptual Hash Algorithm and Hopfield Neural Networks obtained an accuracy rate of 89.36%, making it possible to calculate and recover Perceptual Hash values to detect similar or altered images. Strategy B, formed by Skin Color Detection and Random Forest, achieved an accuracy of 99.98%, making it possible to identify skin colors in image pixels and detect people. Strategy C formed by Metadata Extraction, OCR, LSTM, and Natural Language Processing obtained CER and WER Error Rate values ranging between 0.1 and 10.0, making it possible to detect textual content related to Child Pornography in images, finally, Strategy D formed by Object Detection, Convolutional Artificial Neural Networks, and Generative Adversarial Networks, it obtained a 60% success rate in identifying and classifying objects, making it possible to detect objects related to Child Pornography in images. It was concluded that the development and application of Fenrir supported the execution of expert examinations to detect evidence of child pornography.