Improving face recognition of artificial social companions for smart working and living environments

In this paper we address the topic of improving the performance of human-machine interaction functionalities, focusing on the example of face recognition, using a context based approach to reduce the search space for algorithms performing face identification in large face data sets. We apply this feature in the customization of the behaviour of a virtual assistant, which performs distinct "animations" depending on the identified user. The results presented in the paper refer mainly to the comparison of a face identification algorithm, Eigenfaces, while performing in various scenarios, without and with the context based approach. The conclusions of this work, point in the direction that clustering the search space by defining constrains based on context features lead to improved performance of the identification algorithm while adding some degree of simple first-order logic to the actions performed afterwards (i.e. the behaviour performed by the virtual assistant).