Towards methodological evaluation of affective video content annotation: First steps

Internet ambitioned as an information platform has encouraged the exponential increase of available multimedia data. Such data volume exceeds the limitations of the existing indexing techniques, generating new challenges to ease and customise information retrieval. New research lines have focused on exploiting human emotions for automatic indexing of multimedia information in an attempt to approach indexing schemas and human reasoning. Despite several techniques have been presented along the literature, their evaluation is rather subjective, presenting numerous difficulties for rigorous evaluation and comparison. Thus, the focus of this paper is two-folded. Firstly, a non-intrusive automatic indexing technique based on human emotion modelling is presented. And, secondly, a rigorous quantitative evaluation methodology is proposed establishing a platform for automatic affective indexing analysis, evaluation and comparison.