Reconhecimento de adulterações em imagens digitais: uma abordagem passiva

• Ano de defesa: 2016
• Autor(a) principal: Cunha , Lucas Marques da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Integridade; Autenticidade; Reconhecimento; Análise Forense; Integrity; Authenticity; Recognition; Forensics Analyses; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/tede/9270

The creation and marketing of image editing software allowed ordinary people to perform any kind of manipulation in digital images. In a judicial context, where authenticity and data integrity are crucial, the development of techniques to ensure such attributes are needed. Forensic analysis of digital image aims to use computational scientific methods, such as analysis of a sensor device and JPEG (Joint Photographic Experts Group) artifacts, in order to recognize the presence or absence of such attributes. This paper presents a passive approach to Tampering Recognition in Digital Images with and without JPEG compression using two approaches. The first approach is based on analysis of the 4-pixel neighborhood that may be classified as interpolated or not. Based on such analysis, we obtain information about the standard CFA (Color Filter Array) pattern to investigate the authenticity and integrity of images with low or no compression according to misclassification of pixels. The second approach is based on inconsistency analysis of BAG (Block Grid Artifact) pattern in images with high compression created under tampering techniques like composition and cloning. The image's BAG is the distinction of JPEG blocks. Furthermore, segmentation techniques have been defined for precise location of the tampered area. The method selects one of the approaches according to the image compression ratio. The analysis is performed in agreement with the values of accuracy, sensitivity, specificity, and accuracy. The accuracy rates ranged from 85.1% to 95.4% and precision rates between 41.7% to 74.3%. Values from 32.3% to 82.2% were obtained for sensitivity rates and between 85.9% to 99.2% for specificity in an image database composed by 960 images interpolated by different algorithms and tampered by composition and cloning. The methods described in the literature have a limited scope related to the formats of the images tested and how they gauge their effectiveness. The approach proposed differs from these techniques presenting a most wide scope in the mentioned questions, covering images with and without compression, and assessing the efficiency from metrics able to prove the assumptions during the research.