Wavelet Transform and Image Retrieval: Preliminary experiments

In this paper a novel technique is proposed for image retrieval in large repositories, which relies on both colour and texture features. The technique is a two-pass strategy applied on the wavelet transformed image: in a first step simple colour features are taken into account; subsequently, the search is refined via texture features. Experiments for tuning the system parameters and evaluating the proposed algorithm are reported. The results are also discussed in terms of Receiving Operating Characteristic (ROC) curves.