Detecção automática de fases temporais de emoção em vídeos a partir de características da face

Abstract

Computational techniques employed to solve real-world problems in the real world, such as the recognition of human emotions,has become more frequent. In this context, it is possible to use computational concepts and techniques to analyze and identify human emotions by applying features extracted from different data modalities such as face, gesture and writing. When data are obtained from video, according to the literature, any represented human emotion is composed of four temporal phases involving five steps (neutral, onset, apex, offset and neutral), these phases represent the entire "life" cycle of an emotion. Therefore, defining temporal phases is a very important step, since it supports video emotion recognition systems. This work presents an approach based on appearance and similarity techniques to automatically identify the temporal phases of emotions in videos taking into account face data. The experimental results provided in this work show that the proposed method is able to identify a pattern of emotions temporal phases recognition in videos based on face data features. The learned pattern is independent of the database used.