Exploring inconsistencies on illumination and details generation for exposing computer generated images containing people

• Grant number: 16/13011-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): August 01, 2016
• Effective date (End): July 31, 2017
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Tiago Jose de Carvalho
• Grantee: Matheus Teles de Paiva Alves
• Host Institution: Instituto Federal de Educação, Ciência e Tecnologia de São Paulo (IFSP). Campus Campinas. Campinas , SP, Brazil

Abstract

Computer Graphics methods for creation of digital games and special effects have been strongly improved in the last years, making possible generate images so realistic that are able to deceive even the most skilled viewers. Trying to help experts that, many times, needs to be able to distinguish between computer generated and real images, this research project propose to investigate inconsistencies presents in different kinds of telltales, in a way to make possible detect, as automatically as possible, each one of these aforementioned groups of images. Exploring illumination inconsistencies and imperfections present in regions as eyes and hair to propose discriminative features, this project intend to use machine learning methods to discern computer generated from real images.