Challenges

We propose two contest tasks, that can be performed with methods based on the image, on the text or either, depending on the choice of participants. These tasks are :

  1. The detection of the falsified documents among others. The provided dataset will contain both genuine and modified documents. The first task aims at discovering which ones are forged.
  2. The spotting of this falsification in the document (many falsifications can occur in one document). The aim of the contest is to collect a maximum of methods to be compared. To follow this idea, we suggest that the contest may be opened to two communities :
  • People working at image level : disruption in fonts, localisation of visual anomalies, default of texture, etc..
  • People working  at text level to discover inconsistencies in association rules (abnormal price for a topic), errors in cumulative sums, information checking, etc…