Section: New Results

IKONA/MAESTRO sofware

Participants : Vera Bakic, Laurent Joyeux, Sofiene Mouine, Souheil Selmi.

This year, IKONA has been extended in the context of Pl@ntNet, Glocal and I-SEARCH projects.

• For the Pl@ntNet project, along the continuing improvements and optimizations in the MAESTRO software, a number of new features were added:

  new options for interest points distribution and filtering with the segmented image,

  a new shape context descriptor (corresponding to [11] ),

  various combinations of descriptors in one vector per interest point or region,

  for regions: extraction of sub-images, EOH and Fourier descriptors,

  more options for update of calculated signatures, new score type (used in ImageCLEF2012) and decision rules (adaptive Knn calculation based on individual plant information) for statistical tools.

  In addition, a number of new web services and functionality were developed/updated and deployed:

  the addition of new databases (Vignes, Musa), while some with the organ annotation (Photoflora, Girod), and the update of the multi search views for the new datasets;

  the development of new services allowing to return of botanical information in several formats (csv, xml, JSON...);

  the update of the indexation system so that it can crawl images from different sources (internal or external sources like CEL web service, which uses html or identiplante web service (in JSON format);

  the development of html pages to annotate a set of images by organ type;

  the development of an applet demonstration of leaf architecture (this applet interfaces a library developed by a PhD student working on this project);

  the development of an application "Pl@ntNet Identify" for android platform and its interfacing with the existing web services.

• For the Glocal project, functionalities such as fraud detection and similarity search were integrated in the mock up of user interface and in the final demonstration of the project. In addition, a dynamic indexation system of images from AFP (Agence France-Presse) was implemented as well as the similarity web services working on this dynamic dataset.

• For the I-SEARCH project, an integration of the video mining component in I-SEARCH platform was done. The component extracts visual objects that are the most recurrent from a set of images, or in a video.