At the core of all facial expression analysis solutions is an expression database.
A (facial) expression database is a collection of images showing the specific facial expressions of a range of emotions. These images must be well annotated or emotion-tagged if they are to be useful to expression recognition systems and their related algorithms.
A major hindrance to new developments in the area of automatic human behavior analysis is the lack of suitable databases with displays of behavior and affect. There have been directed advances in this area, as in the MMI Facial Expression Database project, which aims to deliver large volumes of visual data of facial expressions to the facial expression analysis community.
The database is freely available to the scientific community. Find out more about the database online at https://mmifacedb.eu.
In other example projects, we've been able to create our own test data or alter existing datasets to use within a project. However, with expression analysis projects, it is really not realistic to create a reasonably sized database (stating with nothing), which would require the collection and processing of literally thousands of images, all appropriately documented.
After collection, each (facial) image needs to be reviewed and categorized based on the emotion shown into one of seven categories (angry, disgust, fear, happy, sad, surprise, and neutral). To further complicate this work, images may not be aligned and properly proportioned.
The bottom line is that, even if you have a large number of images, if the images are not correctly labeled or simply do not contain detectable facial images, the performance of the expression analysis and detection process will be compromised (it will perform poorly).
These types of challenges make the classification process more difficult because the model is forced to generalize.