Semester: 10th Semester, Master Thesis Title

10.Conclusion

This thesis sought to find a solution and if possible an answer to the above problem, namely whether or not a computer could be programmed to interpret and reason a smile with the same accuracy and understanding as a human.

The implemented software solution was able to rate the smile in the pictures used for test phase one with a 50% correct estimation measured by peaks and valleys of the two graphs. As mentioned in the discussion and introduction to test phase two, errors in the facial detection algorithm prevented the software from rating every picture, thus resulting in only a 50% correct estimation. Furthermore ambiguous smiles, where the physical change in the shape of the mouth was not present also attributed to the low success rate. In test phase two the criteria for the selected pictures were that the computer software could properly identify the smile in the pictures. Furthermore, as test phase one revealed, the mouth is not the only factor that influences the level of perceived smile. Therefore the pictures used in test phase two were required to display exaggerated facial expressions in order to ensure the computer software would be able to rate the pictures. The results from test phase two revealed, that the computer could assess the same level of smile as the test participants, though of note is that only 10 test subjects were subjected to test phase two. In terms of accuracy and understanding the implemented solution worked when analysing exaggerated facial expressions. When factors such as the eye gaze are considered, the algorithm was not able to correctly assess the level of smile. The accuracy of the algorithm was not determined besides establishing, that the ratings from humans and the computer were similar in their graphical representation.

The understanding of the smile is twofold. As the articles from the analysis publicized, a clear definition of what constitutes a smile is vague at best. A general understanding of the smile and what it constitutes from a human perception standpoint is feasible, but determining an emotion is highly individual and ambiguous. Since this thesis tried to categorise the smile on a numerical scale, only a general understanding of the level of smile was achieved. Furthermore as only 42 test subjects undertook test phase one, the number of test subjects is not sufficient to determine if i.e. the rating of picture 4 would be applicable to the rating of an entire population.

The test also revealed that a clear definition of a smile is not possible as even humans have troubles defining what exactly constitutes a smile. With that in mind, the computer software can be viewed as a general rating system that is capable of rating the smile with the same inaccuracy as the human test subjects.

Therefore this thesis suggests that a new computer software should be developed and a new test devised, that should include an analysis of the mouth and eyes in order to try to determine which of these influence the perception of the smile. The following future perspectives chapter will outline the next steps in order to achieve a solution that could possibly bring the detection rate and smile rating accuracy closer to that of humans.