Abstract
The success of content-based retrieval systems stands or falls with the quality of the utilized similarity model. In the case of having no additional keywords or annotations provided with the multimedia data, the hard task is to guarantee the highest possible retrieval precision using only content-based retrieval techniques.
In this paper we push the visual image search a step further by testing effective combination of two orthogonal approaches - the MPEG-7 global visual descriptors and the feature signatures equipped by the Signature Quadratic Form Distance. We investigate various ways of descriptor combinations and evaluate the overall effectiveness of the search on three different image collections.
Moreover, we introduce a new image collection, TWIC, designed as a larger realistic image collection providing ground truth. In all the experiments, the combination of descriptors proved its superior performance on all tested collections.
Furthermore, we propose a re-ranking variant guaranteeing efficient yet effective image retrieval.