This interface demonstrates content-based retrieval from the IRMA database (a subset of 10.000 images). Query images can be selected from the local hard disc and transferred into the system. The query response is calculated immediately within only a few seconds. All user interaction is protocolled to allow to step back to any past state of the system, as well as boolean combinations of those (so called Extended Query Refinement). However, this demo models only the first of all seven processing steps of the IRMA approach, e.g., image categorization based on global features. More specifically, each image is represented by 1024 feature values.

This interface allows the evaluation of content-based image retrieval on several databases. In contrast to other online demos, the similarity of the query image to images in the database is calculated at runtime. Currently neither indices nor pre-computed distances are kept in memory to speed up the query. This may be slower but allows easy switching between different experiments, i.e. feature sets, distance measures and parameter settings. Please keep in mind, that this demo covers only the first of the seven processing steps of the IRMA approach, namely the automatic categorization of images based on global image features.