A Novel Approach for Accessing Partially Indexed Image Corpora
Abstract
This paper addresses the issue of efficient retrieval from image corpora in which only a little proportion is thematically indexed. We propose a hybrid approach integrating thematic querying/search with content-based retrieval. We show how a preliminary double clustering of image corpus exploited by an adapted retrieval process constitutes an answer to the pursued objective. The retrieval process takes advantage of user-system interaction via relevance feedback mechanism whose results are integrated in a {\it virtual image}. Some experimental results are provided and discussed to demonstrate the effectiveness of this work.