A post-doctoral position of duration extensible up to 2 years will be opened on machine learning and multimodal image indexing within the COGIT and MATIS teams, LaSTIG lab at IGN and Imagine team, LIRIS lab at École Centrale de Lyon.
The french research project ALEGORIA aims to index, geolocalize and showcase iconographic institutional collections describing the French territory at different times, in a 3D, space-time, immersive and interactive visualization environment. These iconographic resources represent the territory from different points of view (aerial or terrestrial pictures), different orientations, under different lighting conditions, at different dates from the interwar period to the present day, etc. They are thus difficult to index and geolocalize with state-of-the-art approaches. However, they are described by metadata that can be linked to reference datasets: these important sources of knowledge could be leveraged to address these issues.
Many works in the field of multimedia information research have shown the interest of image search by multimodal matching combining visual content and textual metadata. The textual information associated with the images can be used both to filter the most relevant images and weight the importance of some visual content descriptors according to the type of expected visual content. This post-doctorate work follows the intuition that it would be possible to customize the choice and the weighting of the image descriptors to be implemented for image content matching and annotation by producing a visual context specific to each image based on external sources of knowledge.
The images collections to be indexed describe the French territory. They are thus very likely to represent geographic features, like buildings, roads, watercourses, forests, etc. A priori knowledge of their presence on the images, of their feature type, their spatial distribution or their resolution could be used to guide the strategy of images matching. The objective of this post-doc is thus to propose new approaches for defining and taking into account a visual context in a multimodal cross-domain image matching strategy.
The successful candidate will work on the ALEGORIA’s work-package on multimodal information harnessingfor image indexing. Please visit here for further details: https://docs.google.com/document/d/1_6MudJ3EjiXEMkc7UQ26_4okUmT-pUJuhyP5_Wcm1h8/edit?usp=sharing
Applications should include a curriculum vitae, a cover letter focusing on this post-doc position, 2 reference letters with the contact information of the references (all grouped together in the same PDF file). Applications and letters should be sent via electronic mail to:
Application deadline: September 1st, 2018
Start date: before end 2018
Post-doctorate duration: 2 years
Location: Depending on the candidate's profile, the position may be located either at the LaSTIG (in Saint-Mandé, Paris neighborhood, Metro line 1, direction Château de Vincennes) or at the Imagine Team (Lyon neighborhood).
(c) GdR 720 ISIS - CNRS - 2011-2018.