Local feature selection for urban image retrieval

Nicolas Hascoet (1) , Titus Zaharia (2, 1, 3, 4, 5)

1 TSP - ARTEMIS - Département Advanced Research And Techniques For Multidimensional Imaging Systems
2 ARMEDIA-SAMOVAR - ARMEDIA
3 SAMOVAR - Services répartis, Architectures, MOdélisation, Validation, Administration des Réseaux
4 MAP5 - UMR 8145 - Mathématiques Appliquées Paris 5
5 CNRS - Centre National de la Recherche Scientifique

In this paper, we propose an improved image retrieval method, dedicated to images of buildings/landmarks from urban environments. Locally detected key points are binary labelled as building or no-building using a SVM-based classifier. Thereafter, only key points labelled as building are retained. In this way, the data in the database vocabulary is reduced to only the relevant one and solely the relevant features, effectively describing the targeted buildings are considered. The experimental results, carried out on the Paris6k and Oxford5k data sets show significant improvement in terms of retrieval precision

Domaines: Traitement du signal et de l'image [eess.SP]

Nicolas Hascoet, Titus Zaharia. Local feature selection for urban image retrieval. ISSCS 2017 : International Symposium on Signals, Circuits and Systems , Jul 2017, Iasi, Romania. pp.1 - 4, ⟨10.1109/ISSCS.2017.8034887⟩. ⟨hal-01687246⟩