Skip to main content

Twelve Labours of Image Processing

Final Report Summary - TWELVE LABOURS (Twelve Labours of Image Processing)

In this project the methodology of image processing and image analysis has been tightened by making a systematic exploration of the state of the art of some twenty fundamental problems, and by deciding to publish online the most relevant methods for each one. The project has trained (or is currently training) some 40 researchers in the mathematics and algorithms of image processing and in their reproducible publication.
The main novelty of the project is the creation of the first journal for image processing which associates to the algorithm being published an online, executable and reproducible demonstration of the algorithm being published:
IPOL (Image Processing On line, www.ipol.im).
It is seemingly the first journal with these requirements. From 2011 to 2015, this journal has evolved from an experimental web platform to a permanent journal with 90 articles published, 23 submitted and 40 in preparation.
The journal also encourages the more informal exchange of executable algorithms by hosting “workshops”. Workshops are temporary publications of online algorithms which do not need to follow the publication rules, but allow researchers and research partners in a project to interact directly. This permits to avoid the software curse which hinders researchers using different platforms or languages to communicate easily their algorithms. In particular the journal asks from the authors a thorough literary description of their algorithm with its pseudo-code and all parameter specifications. The referees are asked to check that the code in in agreement with this specification. In that way algorithms are reproducible in three different ways:
a) by executing them remotely on the journal’s server;
b) by downloading the public code and using it;
c) by reprogramming it easily following the detailed instruction given by the pseudo-code. Each online paper at IPOL has its experimental public archive, where all experiments made remotely by anonymous readers are stored. Currently (0ctober 2015), the archive contains some 150000 experiments.