An ablation study on 360-degree saliency estimation

Omni-directional multimedia is increasing in popularity thanks to the diffusion of head mounted displays for its fruition. The shift from the 2D to omni-directional paradigm implies a change in the way users interact with the visual content. In order to analyze how users explore this new type of stimulus, saliency maps can be exploited. Although several approaches have been proposed in the literature, a complete study on how the different image features contribute to the overall saliency is lacking. For this reason, this work considers a set of features commonly employed in the state-of-the-art, and analyzes their impact on saliency estimation. The performed analysis highlights that increasing the number of features does not necessarily imply better results, and saliency estimation gains from feature combination depending on the feature type.