Microe- A Comprehensive Dataset of Facial Micro Expressions for Building Responsible Software Systems with Emotion Recognition

Abstract

Micro expressions are facial expressions that occur for a short period of time, usually less than a second. These expressions usually go unnoticed but represent the true expression that is felt by the person because they are involuntary. There have been recent attempts to make datasets that contain micro expressions in an attempt to carry out accurate Facial Emotion Recognition. However, many of these datasets have had various limitations, ranging from low diversity of the dataset, which introduced bias into the emotion detection models, low quality images, a limited number of images that resulted into low accuracy of the model, etc. The study introduces a dataset called Microe dataset, which contains images that were collected from various sources to ensure a high diversity and more features that are going to be discussed in sections below. Experiments were also carried out to evaluate this dataset and the results that were acquired have been included in this paper.