A multimodal approach for the automatic assessment of viewer subjective perception of Youtube videos

This research project is focused on the understanding of how saliency could influence the impression perceived by a viewer of a video. However, this perception cannot be perfectly assessed since some inherent bias is inevitable. This problem is an existing error that should be understood and taken into account in any statistical analysis or experiments. For instance, and focused on this research, the perception of a viewer can be affected by tendencies, inclinations or feelings of the individual (Attentional bias). In order to lessen the impact of this issue it is important to define a simplified research field, limiting tendencies or feelings of the viewers. Due this, car commercials were chosen as the video domain for the study. These and other following choices were previously stablished by Alejandro Hernández García in his project called Aesthetics Assessment of Videos through Visual Descriptors and Automatic Polarity Annotation which will be named several times in this research. This car commercial database was extracted from YouTube and was simplified obtaining finally a collection of 138 car commercial videos. This collection will be used for this project too and will be the domain for the proposed computational model. This way, we will demonstrate the impact of the saliency in subjective perception, focusing this assumption on car commercial videos extracted from YouTube. As consequence of this assumption, saliency also would modify the video ratings in this platformIngeniería de Sistemas Audiovisuale