TBM offers an interesting premise. Without intending to provide a comprehensive explanation of the TBM framework, a key point is that it introduces degrees of belief and transfer (giving rise to the name of the method: the transferable belief model), which allows the model to make the necessary assumptions required to perform adequate classification (of the expressions). Basically, this means it scores its assumptions, that is, the assumption that the expression is a happy expression is determined to have a n percentage chance of being correct (we'll see this in action later in this chapter when we review the results of our project).

Further (and I'm oversimplifying), TBM looks to use quantified beliefs to make its classification decisions. Something perhaps more easily understood is that facial expression analysis extracts an expression skeleton of facial features (the mouth, eyes and eyebrows) and then derives simple distance coefficients from facial images. These characteristic distances are then fed to a rule-based decision system that relies on TBM in order to assign a facial expression to the face image. 

Again, the goal is not to define the theory behind TBM, or even the intimate details of a facial expression analysis solution, but more to show a working example of such; therefore, we will go on to the next section and our use case example and leave to the reader the further work of researching this topic.