large-scale image collections. We are investigating a query-free retrieval approach, which relies solely on an iterative relevance feedback mechanism driven by user subjective perception of image similarities.
Most of the image retrieval approaches require an initial query before offering relevance feedback tools. The motivation for a query-free retrieval approach relies on the observation that formulating a query might not be the most optimal way of initializing a searching session. User retrieval needs are often difficult to describe in terms of keywords and relevant images may be easily filtered out.
The aim of this project is to increase the retrieval capabilities of the query-free approach by modeling the user similarity judgments beyond the relevance feedback information given during a single searching session. A particular goal is to investigate the possibilities to extract useful indexing information from the user logs. We aim to identify a solution that is able to support personalized similarity metrics eventually.