01 Mar
14:00 Master's Defense Fully distance
DeepFakes Detection from Computer Vision and Machine Learning Techniques
Camila Steffane Fernandes Teixeira de Moura
Advisor / Teacher
Anderson de Rezende Rocha
Brief summary
With technological advances and access to information in real time, a new class of problems has emerged, the fake news. An evolution and more worrying technique that has been used are Deepfakes, in which images or videos are artificially manipulated, through deep learning, which has been minimizing the time and the amount of data necessary to develop these synthetic / false contents. Analyzing this growing problem and the impacts that can occur, in political, entertainment areas, among others, we developed this project, which aims to analyze how machine learning and deep learning techniques can assist in the task of detecting Deepfakes. In the course of this work, we analyzed some available databases, in addition to creating our own database composed of fake sensitive content videos involving only women, Deepfakes' main target audience for adult entertainment. We evaluated the behavior of traditional techniques of deep learning for the task of classification between images with and without facial manipulations, as well as the influence that the specific characteristics of the images can directly impact on the quality of the generated detection model.
Examination Board
Anderson de Rezende Rocha IC / UNICAMP
John Paul Pope FC / UNESP
Julio Cesar dos Reis IC / UNICAMP
Sandra Eliza Fontes from Avila IC / UNICAMP
Jefersson Alex dos Santos DCC / UFMG