@techreport{TR-IC-PFG-18-27,
   number = {IC-PFG-18-27},
   author = {Gabriel {Bertocco} and Anderson {Rocha}},
   title = {{Event Repurposing Detection}},
   month = {December},
   year = {2018},
   institution = {Institute of Computing, University of Campinas},
   note = {In English, 23 pages.
    \par\selectlanguage{english}\textbf{Abstract}
       Nowadays  one  of  the  greatest  problems  faced in social and
       electronics  media  is  the  attempt  to  change the purpose of
       images in order to increase or change the impact of some event.
       For  example,  a malicious person can use a picture depicting a
       specific  event to illustrate another completely event, leading
       to  misurderstanding  and changing the public opinion about the
       event  or  related topics. This repurpose of the meaning of the
       original  picture is called Event Repurposing. In this work, we
       propose two methods to detect if a image is being repurposed or
       not  (binary  classification  problem)  based on the full scene
       analysis  (using  the  whole  image) and on the analysis of the
       objects (cars, people, trucks, and so on) present in the scene.
       In  order  to  model  the  binary  classification,  we  extract 
       features  from  the  images  using  a Deep Convolutional Neural
       Network  (DCNN)  and  train  an  One-Class  SVM.  To  check the
       robustness of the proposed methods, we consider eight different
       events with different objects and landscapes to get variability
       in terms of scenario, context and purpose of the events.
  }
}