Working Group III / 3 - Image Analysis for Indexation and Image retrieval, 2008-2012
|
MissionsInformation retrieval from image databases has been the subject of intense interest over the past few years as the size of such collections has grown, but the proportion of this work devoted to satellite, aerial and street-view images, as opposed to generic image and multimedia databases for example, is quite small. Satellite and aerial image indexing presents several specificities. First, remote sensing databases are operated by specialists from widely varying fields. The needs of the user are thus precise and complex, yet very different from one application to another. Second, the nature of satellite, aerial and street-view images is very different from those in generic databases. Textural and spectral measures are very important for remote sensing image understanding, but are seldom used in generic databases. Other elements, such as buildings, bridges, human activity, and so on, are better characterized by their inter-relations than by their individual characteristics. Third, the high-resolution imagery that will form the databases of tomorrow, varies considerably in resolution. Methods to compensate for this variability are therefore essential, but again do not form part of existing information retrieval techniques. Another very important component of the satellite, aerial, and street-view image database processing is the huge amount of data. For retrieval purpose, this scalability aspect has to be carefully considered. For generic image databases, several indexing schemes have been designed to face this problem, but it is still a scientific lock with a lot of opened questions. Learning categories and semantic concepts from these data is also a very challenging problem. Recent advances have been done to fill the gap between the low-level features extracted from the data and the high-level semantic concepts that the final user attempts to manipulate. In particular, statistical learning framework allows to efficiently work on high dimensional data, complex categories or concepts and with on-line learning. Based on the above considerations, the objective of this working group is to promote works related to the use of low-level (textural and spectral) and high-level (structural) primitives for information retrieval from remote sensing databases containing images of widely varying resolutions, within a semantic learning framework based on working image retrieval systems. Terms of Reference
Technical Commission Officers
|
Olivier Tournaire) |
© ISPRS |