Uma abordagem para a deformação de faces em vídeos visando à caricaturização.

Abstract

Facial caricature can be defined as the representation of an individual by exaggerating the most distinct features of the face and simplifying the most common features. Computer-generated cartoons, in turn, are developed with the intention of producing automatic or semi-automatic cartoons. This is due, in large part, to the growing advancement in the areas of Computer Graphics and Computer Vision. Within this context, the present study has as its main goal the proposition of an approach for the automatic generation of facial distortions in digital videos that resemble facial caricatures. The main challenges of the study were the design of a model for the detection of facial points in videos, obtaining general rules that visually satisfy the automatic generation of caricatures, as well as the treatment of possible problems related to the generation of the distorted videos. In this study, an adaptive approach was designed to detect fiducial points in videos, and a strategy was developed so that the temporal coherence of deformations was maintained. After comparing the adaptive approach developed in this study with other approaches, it was possible to observe a lower rate of error in locating the fiducial points of the face. It was also possible to determine a configuration that best met the criteria listed in this study, based on an analysis of the results obtained through a questionnaire to survey the opinions of a set of volunteers.