Hamming Embedding and Weak Geometry Consistency for Large Scale Image Search - extended version
نویسندگان
چکیده
This technical report presents and extends a recent paper we have proposed for large scale image search. State-of-the-art methods build on the bagof-features image representation. We first analyze bag-of-features in the framework of approximate nearest neighbor search. This shows the sub-optimality of such a representation for matching descriptors and leads us to derive a more precise representation based on 1) Hamming embedding (HE) and 2) weak geometric consistency constraints (WGC). HE provides binary signatures that refine the matching based on visual words. WGC filters matching descriptors that are not consistent in terms of angle and scale. HE and WGC are integrated within an inverted file and are efficiently exploited for all images, even in the case of very large datasets. Experiments performed on a dataset of one million of images show a significant improvement due to the binary signature and the weak geometric consistency constraints, as well as their efficiency. Estimation of the full geometric transformation, i.e., a re-ranking step on a short list of images, is complementary to our weak geometric consistency constraints and allows to further improve the accuracy. Key-words: image retrieval, nearest neighbor search, image matching, geometrical transform, large image databases ∗ [email protected] in ria -0 05 48 65 1, v er si on 1 20 D ec 2 01 0 Recherche d’image par Hamming Embedding et Contraintes Géometriques faibles Résumé : Ce rapport technique reprend et étend un article récent sur la recherche d’images dans des grandes bases. Les méthodes de l’état de l’art reposent sur une représentation des images par sac de mots. Nous exprimons la mise en correspondance de ces descripteurs dans le contexte de la recherche approximative de plus proches voisins. Nous montrons que cette représentation est sous-optimale. Ceci nous amène à définir une représentation plus précise, basée sur 1) l’immersion dans un espace de Hamming (HE) et 2) des contraintes géométriques faibles (WGC). Le HE ajoute aux descripteurs une signature binaire qui permet d’affiner leur mise en correspondance. Le WGC filtre les correspondances de points dont les caractéristiques d’angle et d’échelle ne sont pas cohérentes. HE et WGC sont intégrés dans une structure de fichier inversé et appliqués à toutes les images, même pour de très grandes bases. Des expériences sur un million d’images montrent que la signature binaire et la contrainte géométrique faible améliorent significativement la précision, sans allongement des temps de calcul. Le réordonnancement des meilleures images par l’estimation d’une transformation géométrique complète est complémentaire avec notre WGC, et améliore encore la précision. Mots-clés : recherche d’image, recherche de plus proches voisins, appariemment d’image, transformation géométriques, grandes bases d’image in ria -0 05 48 65 1, v er si on 1 20 D ec 2 01 0 HE and WGC for large scale image search 3
منابع مشابه
Hamming Embedding and Weak Geometric Consistency for Large Scale Image Search
This paper improves recent methods for large scale image search. State-of-the-art methods build on the bag-of-features image representation. We, first, analyze bag-of-features in the framework of approximate nearest neighbor search. This shows the sub-optimality of such a representation for matching descriptors and leads us to derive a more precise representation based on 1) Hamming embedding (...
متن کاملRecent Advances in Large Scale Image Search
This paper introduces recent methods for large scale image search. State-of-the-art methods build on the bag-of-features image representation. We first analyze bag-of-features in the framework of approximate nearest neighbor search. This shows the sub-optimality of such a representation for matching descriptors and leads us to derive a more precise representation based on 1) Hamming embedding (...
متن کاملLarge Scale Image Search
We address the problem of large scale image search, for which many recent methods use a bag-of-features image representation. We shows the sub-optimality of such a representation for matching descriptors and derive a more precise representation based on 1) Hamming embedding (HE) and 2) weak geometric consistency constraints (WGC). HE provides binary signatures that refine the matching based on ...
متن کاملINRIA-LEAR'S Video Copy Detection System
Fig. 1 illustrates the video copyright detection system we have developed for the TRECVID 2008 evaluation campaign. The components of this system are detailed in Section 2. Most of them are derived from the state-of-the-art image search engine introduced in [2]. It builds upon the bag-of-features image search system proposed in [3], and provides a more precise representation by adding 1) a Hamm...
متن کاملMultiple Dictionaries for Bag Ofwords Large Scale Image Search
We explore building better dictionaries of visual words in Bag of Words (BoW) systems for searching large scale image collections. We propose a novel way to use more visual words in the BoW algorithm that significantly increases the recognition performance: combining multiple dictionaries built from different subsets of the features. We report improvement in performance by 40-45% over the basel...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2008