The RamblingFreeMan rambles on both life and tech but mostly the latter. The RamblingFreeMan wishes to speak, so we shall shift to first person now. Hi guys, I am intrigued by ML and would like to research its applications in domains like computer vision. So I am going to start a tech series on how to build a simple facial emotion recognition system which can recognize up to six emotions. This happens to also be part of my club project that we are doing in college.
So the question arises as to why such a system is required? The importance lies in human computer interaction which the title suggests is the study of how humans interact with computers; It is used in industry to innovate and improve user experience. There are already companies like Affectiva which have amassed a huge database of faces, both videos and images and gives accurate readings of faces.
There are disadvantages with deciding emotions only from a face as context i.e situation is also important. For example, after the recent Paris attacks, an emotion analysis was done on one of the surviving victims but it wasn’t able to discern whether his slight smile was because of frustration or happiness but it marked his emotions as a mixture of neutral and happy. Another disadvantage is that a lot of emotions are communicated by our body language rather than our faces. So if scientists want to build a holistic emotion recognition system, it will have to take into account these factors too.
The one flaw is that our machine learning model we are going to use is only learning from images but if it does so from videos, we can improve on how to classify emotions such as smiles. If a smile forms and last longer, we can say its happy and if it suddenly forms and disappears we can say it is from frustration. This is what MIT researchers have done to differentiate such smiles.
Applications of such a system are numerous; for example we can check whether a customer is happy with a product just by their face. This will also help to make more interactive and intelligent computers.
That’s all for today. Do look up next week on the first part of the tech blog on facial emotion recognition system where I will talk about implementing face detection to identify the faces.