Content-Based Image Retrieval
Click picture to get a higher resolution screen shot of our system Introduction/Problem Statement The amount of research dedicated to content-based image retrieval (CBIR) has increased immensely over the last decade. In order to avoid the expense and limitations of manual text annotations, there is considerable interest in efficient database access by perceptual and other automatically extractable attributes. From a computer vision point of view the main challenge of CBIR remains the topic of active research: the semantic gap between the user and the retrieval system. Most current retrieval systems only rely on low-level image features such as color and texture whereas users rather think in terms of concepts. |
 |
Approach With the goal to bring the retrieval process closer to the user, we propose a framework called vocabulary-supported image retrieval. The idea hereby is to give the user the possibility to describe an image in a ``human way'' by a set of vocabulary and the size of the image area that should contain the particular concept (e.g. ``Find images with x% of sky and y% of water.''). In order to enable the use of vocabulary, the system provides a set of image detectors. These detectors may either be implemented specifically or can be learned. For the retrieval, the detectors work locally and return the probability that a certain segment contains the concept being searched for. Depending on this output, a general decision whether the whole image satisfies the query can be made. Additionally, vocabulary-supported image retrieval enables a performance prediction prior to the search. Depending on the result, the internal parameters of this prefiltering stage can be set so as to maximize precision or recall. |
 |
Project Partners The research is funded by the COGVIS project. The psychophysical experiments are collaborative work with Dr. Adrian Schwaninger.Publications
Vocabulary-Supported Image Retrieval Bernt Schiele and Julia Vogel, In 1st DELOS Workshop on Information Seeking, Searching and Querying in Digital Libraries, Zurich, Switzerland, December 2000. Contact |
 |