Towards multimodal emotion recognition : A new approach

                        Multimedia indexing is about developing techniques allowing       people to effectively find media. Content–based methods       become necessary when dealing with large databases as people       cannot possibly annotate all available content. Emotions       are intrinsic in human beings and are known to be very important       for natural interactions, decision making, memory,       and many other cognitive functions. Current technologies       allows exploring the emotional space by mean of content–       based analysis of audio and video, but also thanks to other       modalities such as the human physiology.       In this paper, we present the latest development in the       emotion recognition part of SAMMI [18] by mean of an       extensive study on feature selection and the application of       many of the principles we have presented in [17] and [15].       Then, we present the concepts of output thresholding, inverse       thresholding, and profiling which we used for improving       the results of the recognition. Finally, we present a study       on the robustness to rotations and zoom of our feature point       tracking system.       Our experiments on the six prototypical emotions by Ekman       and Friesen presented in the eNTERFACE’05 database       result in as much as 75% correct recognition rate.

Document Doi Bibtex

Permalink: http://www.eurecom.fr/publication/3107