Facial recognition systems very rarely benefit from an optimal environment, in real life, to acquire images. A good light or a good angle are almost a fantasy. Often, for the same person, the results of facial recognition are different only with one of these two variations of environment.
So imagine when it comes to dealing with human nature and its moods, especially when they influence the facial expressions of video-recognized individuals.
As a result, the research world dealing with facial recognition is working on modifying algorithms to train systems to recognize 6 facial attitudes, in variations of brightness and image quality, with respect to the absolute neutrality found on identity documents, which gives us the look of a convict: joy, sadness, surprise, anger, fear and disgust.
The number of scientific publications about facial expression variation is quite impressive, proof if any were needed that this is an important issue for future systems. Among these researches, some are surprising.
For example, a study conducted by a group of Saudi Arabia researchers revealed that one emotion in particular (somewhat) thwarts facial recognition systems.
It was a strange research, you may say, and yet the aim was to determine which of the common expressions could potentially thwart biometric systems, or significantly influence the result. This was not to encourage crime or to find out how to manufacture an emotional make-up, but to optimize the systems.
So, which facial expression alters facial recognition systems?
Without further ado, the correct answer is the expression of disgust.
This was determined by giving a similarity score to each expression with respect to the neutral expression.
Sadness scored 93.93% similarity to the neutral expression. This is certainly due to the lack of mobility of the facial muscles when this emotion hugs us.
Disgust, on the other hand, shows a more significant difference of almost 8% with the neutral facial expression, with a score of 92.01%.
Disgust is therefore the feeling that most alters the most striking features during recognition, which is based on fixed slopes and lengths for each individual. This is particularly the case for the gaps between right eye, left eye and mouth, the gap between eyes and nose and nose – mouth. Or the significant gap between ears and mouth or eyes and eyebrows (I will come back in a future article on constants useful for facial recognition).
Read Docteur Zagrouba and Madame Alrubaish study for more informations.