Paper Title
Labeled Expression Recognition on Face Drawings

Abstract
In the following paper we analyze the performances of two training schemes of a system that performs based on FER+ dataset, testing the timing for GPU vs CPU execution on the standard system, and picking the best, to experiment on a drawing dataset created from drawing like images of a small sample from the UTK dataset in the wild. The results show that Probabilistic Label Drawing training model performs better in reaching the best test accuracy with a value of 71.2%, but when involving the best validation accuracy – test accuracy pair, Majority Vote system is better with a corresponding value of 68.48% for a validation accuracy of 86.37%. The originality of the system consists in the experiment of a new idea of drawing dataset integration on a labeled emotion recognition system. Keywords - Expression Recognition, Sketch Recognition, Drawing, Face