Nicole Vincent

Researchers in the DIA field have proposed different interesting techniques for the extraction of elements of contents inside images at different levels: lexical, syntactical, semantical. Besides, different knowledge representation techniques have been involved for the analysis of relations between them. From these observations, it seems very interesting to propose a comparative study showing the increasing similarity between CV and DIA problems and discussing the recent trends that have been or can be proposed in a short future to try to deal more efficiently with semantic and user interaction in these fields (computer vision and robust reading). By selecting and explaining interesting recent works from the CV and DIA communities, this paper tries to bring answers to the following questions: Can we do more than automatic features selection with CNN based architectures? Can we reach a more semantical point of view on the data? What are the solutions to define adaptable semantic models that can be easily learned or user-defined? How can we get more generic systems by combining machine learning & user interaction?

Satellite Image Time Series (SITS) provide valuable information for the automatic mapping of our territories. In this article we focus on the analysis of urban land covers from SITS, trying to evaluate the density of artificialized areas. We hypothesize that such areas do not evolve significantly through time (over the interval of a year) compared to other nonartificialized areas (e.g., agricultural crops, vegetation). The proposed approach is based on a spatio-temporal characteristic measuring the temporal stability of a zone, extracted using the Run Length Encoding method. Preliminary results obtained on a series of 41 SENTINEL-2 images highlight the ability of our approach to discriminate different urban land-cover classes (e.g., artificial areas, high density vs. low density housing areas).

Image change detection has been extensively tackled in the literature in various domains, and in particular for remote sensing purposes. Indeed, very high resolution geospatial images are nowadays ubiquitous and can be used to update existing 2D and 3D geographical databases. Such databases can be projected into the image space, by a rasterization step. Therefore, they provide 2D label maps that can be subsequently compared with classifications resulting for geospatial image processing. In this paper, we propose a classificationbased method to detect changes between label maps created from 2D land-cover databases and an more recent orthoimage, without any prior assumptions about the databases composition. Our supervised method is based both on an efficient training set selection and a hierarchical decision process, that follows the structure of topographical databases. This allows to take into account the intrinsic heterogeneity of the objects and themes composing a database while limiting false detection rates, a standard limitation of existing approaches. The designed framework is successfully applied on very high resolution images of Pl´eiades sensor and two distinct national land-cover databases.

In this paper we introduce the numerical tools that have been developed in the context of the Graphem project, in order to automate or leverage several steps in the study of medieval writing samples. We first describe various kinds of features that have been extracted from the samples, and then present two graphical tools to compare writing samples according to the features that have been extracted.