Interactive Exploration for Image Retrieval.

Matthieu Cord (1) , Jérome Fournier (1) , Philippe-Henri Gosselin (2) , Sylvie Philipp-Foliguet (1)

1 ETIS - UMR 8051 - Equipes Traitement de l'Information et Systèmes
2 MIDI - Multimedia Indexation and Data Integration

In this paper, we present a new version of our content-based image retrieval system RETIN. It is based on adaptive quantization of the color space, together with new features aiming at representing the spatial relationship between colors. Color analysis is also extended to texture. Using these powerful indexes, an original interactive retrieval strategy is introduced. The process is based on two steps for handling the retrieval of very large image categories. First, a controlled exploration method of the database is presented. Second, a relevance feedback method based on statistical learning is proposed. All the steps are evaluated by experiments on a generalist database.

Domains: Machine Learning [stat.ML]

Matthieu Cord, Jérome Fournier, Philippe-Henri Gosselin, Sylvie Philipp-Foliguet. Interactive Exploration for Image Retrieval.. EURASIP Journal on Advances in Signal Processing, 2006, 14, pp.2173-2186. ⟨hal-00520284⟩