OntoImage'2008 - 2nd International "Language Resources for Content-Based Image Retrieval" Workshop

There are two computer-aided methods for accessing images:

* indexing textual descriptions (done mostly by public retrieval engines)
o assigned manually on community sharing sites, e.g. Flickr;
o extracted from the text of ...

There are two computer-aided methods for accessing images:

* indexing textual descriptions (done mostly by public retrieval engines)
o assigned manually on community sharing sites, e.g. Flickr;
o extracted from the text of the web page associated with the image , e.g. Google
o extracting textual information inside the image using OCR techniques
* indexing low-level visual features (colour, texture, etc.) of the images (called Content Based Image Retrieval - CBIR)

Combining these two methods is currently a hot topic since it involves developing hybrid approaches using automatic image tagging from text along with low level visual features from an image to produce text metadata. One of the main limitations within state-of-the-art CBIR resides in the mismatch (the "Semantic Gap" in image analysis) between low-level visual descriptors and the human-understood semantics of picture similarity. There are problems of vocabulary at all levels in the description of an image. There is a mismatch between the text found around an image, between the keywords that people assign to an image, and between the objects that can be recognized in an image. How can available language resources (terminology, thesauri, ontologies, etc.) help? How much must they be altered to reduce the semantic gap?

» more

Are you the conference organizer?

Report a Problem With This Conference