Semi-automatic Image Annotation

Image annotation is an open issue. Indeed, there is a "semantic gap" between the digital audiovisual documents composed of pixels and the words, human users express their image needs in. Moreover, we need to describe images as well when we annotate them (describe whet we see) as when we search them (describe what we want to see). In order to find images and in general to manage them, one needs to translate the words and the pixels composing the images into comparable artifacts.

Challenges and Contributions

We identified several scientific issues related to image annotation. Some of them tackle fully automatic annotation approaches, others include the human user in the process and try to accelerate, improve the methods. The main challanges we adress are:

• metadata and semantic resource based image annotation
• use trace based image annotation assistance
• social network based image annotation
• content based image annotation

Our main contributions are:

• keyword cleaning and term distance, based on semantico-statistic methods
• keypoint based image comparison optimization, using early keypoint pruning
• content and metadata based geo-localized image annotation enrichment methods
• use trace model to assist non trivial tasks (video clip composition, image and video search and exploring)

Contributors

present

• Elöd EGYED-ZSIGMOND
• Sylvie CALABRETTO
• Jean-Marie PINON
• Betül AYDIN

past

• Hatem MOUSSELLY-SERGIEH
• Reim DOUMAT
• Sonia LAJMI

Grants

• Startup co-funding (109Lab), regional grant, 2011
• Serious games regional projetc: VieTaVille, 2010-2012

Selected publications

You can find here a list of publications

Software

• Geo localized image annotation

A framework for geo localized image annotation using semantic resources and metadata combined with image content based features.

• Wanaclip

A trace based video clip creation assistant over the Wanaclip application developed by Webcastor. (work in collaboration with Webcastor and the Silex team.

• Armarius

A system for semi-automatic annotation of digitized ancient manuscripts.