Spatiotemporal saliency model

Ioannis Cassagne, Marc Décombas

In our digital society, the amount of data is constantly increasing. Each day thousands of hours of video are recorded but there are not always people to watch and analyze them. A way to solve this problem is to use saliency model. It can be defined as the art of modeling the mechanism of human attention. This ability, which has allowed humanity to survive, will be an asset to detect the most important information, in videos. By this way, danger and abnormal events can be highlighted for video protection in jails, train station or airport CCTV by example.

Static saliency has been widely studied and some works has been extended for video applications. Based on the Spatio-Temporal RArity saliency model using Priors (STRAP), proposed by N. Riche et al., this project ,conducted by Dr Marc Décombas and Ioannis Cassagne, will improve the robustness of the model over the temporal dimension and four new keys ideas will be studied to contribute to the Spatio-Temporal Saliency problematic.