Informace o publikaci

Face Image Retrieval Revisited

Autoři

SEDMIDUBSKÝ Jan MÍČ Vladimír ZEZULA Pavel

Rok publikování 2015
Druh Článek ve sborníku
Konference Proceedings of 8th International Conference on Similarity Search and Applications (SISAP 2015), LNCS 9371
Fakulta / Pracoviště MU

Fakulta informatiky

Citace
Doi http://dx.doi.org/10.1007/978-3-319-25087-8_19
Obor Informatika
Klíčová slova face retrieval; face detection; effectiveness; efficiency; face matcher fusion
Popis The objective of face retrieval is to efficiently search an image database with detected faces and identify such faces that belong to the same person as a query face. Unlike most related papers, we concentrate on both retrieval effectiveness and efficiency. High retrieval effectiveness is achieved by proposing a new fusion approach which integrates existing state-of-the-art detection as well as matching methods. We further significantly improve a retrieval quality by employing the concept of multi-face queries along with optional relevance feedback. To be able to efficiently process queries on databases with millions of faces, we apply a specialized indexing algorithm. The proposed solutions are compared against four existing open-source and commercial technologies and experimentally evaluated on the standardized FERET dataset and on a real-life dataset of more than one million face images. The retrieval results demonstrate a significant gain in effectiveness and two-orders of magnitude more efficient query processing, with respect to a single technology executed sequentially.
Související projekty: