The IIDF model is the model I use for multimodal indexing of images.
The model is composed of several parts, and organized in a similar way to MPEG-7:
IIDF.xsd
metadata.xsd
keywordBasedAnnotation.xsd
globalModality0.xsd: images are indexed in a image retrieval engine, and QBE is used to get nearest neighbors. This modality represents the nearest neighbor using a specific image retrieval engine.
flickr.xsd: images from flickr. These images have an
URL from where they come from, eventually a flickr tag query, and flickr tags.
The very first version dedicated to test only:
paths to referenced images are now either relative either absolute if it's a local path. Using relative paths could be usefull when I directly copy images from the web (i.e. using Flickr for example): for instance ./test-dataformer/set0-flickr is the root of an IIDF, and ./test-dataformer/set0-flickr/images is the file under which are located the images.