Information Theoretic Rotationwise Robust Binary Descriptor Learning
In this paper, we propose a new data-driven approach for binary descriptor selection. In order to draw a clear analysis of common designs, we present a general information-theoretic selection paradigm. It encompasses several standard binary descriptor con
- PDF / 661,587 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 58 Downloads / 224 Views
3
Groupe de Recherche en Informatique, Image, Automatique et Instrumentation de Caen Normandie Univ, UNICAEN, ENSICAEN, CNRS, GREYC, 14000 Caen, France {loic.simon,luc.brun}@ensicaen.fr 2 44screens, Paris, France [email protected] Computer Vision Center Dep. Inform` atica, Universitat Aut` onoma de Barcelona, 08193 Bellaterra (Barcelona), Spain {josep,felipe}@cvc.uab.es
Abstract. In this paper, we propose a new data-driven approach for binary descriptor selection. In order to draw a clear analysis of common designs, we present a general information-theoretic selection paradigm. It encompasses several standard binary descriptor construction schemes, including a recent state-of-the-art one named BOLD. We pursue the same endeavor to increase the stability of the produced descriptors with respect to rotations. To achieve this goal, we have designed a novel offline selection criterion which is better adapted to the online matching procedure. The effectiveness of our approach is demonstrated on two standard datasets, where our descriptor is compared to BOLD and to several classical descriptors. In particular, it emerges that our approach can reproduce equivalent if not better performance as BOLD while relying on twice shorter descriptors. Such an improvement can be influential for real-time applications.
1
Introduction
Since the advent of SIFT [12], extracting local descriptors has become a common practice in order to assess the similarity of image regions. Applications of local descriptors have been considerable, such as image stitching to build panoramas [5], context-based image retrieval, visual odometry or multi-view 3D reconstruction [15]. As a result of its success, this line of research has greatly impacted our everyday behaviour, be it by our use of efficient exemplar based image search engine, or the pervasive introduction of computer vision in mobile devices. Due to this important economical and societal repercussions, the design of ever improving descriptors has drawn a strong interest [4,14]. One of the main enhancements relates to data-driven construction schemes, where a typical database of image correspondences is leveraged to learn an efficient descriptor [8,21]. c Springer International Publishing AG 2016 A. Robles-Kelly et al. (Eds.): S+SSPR 2016, LNCS 10029, pp. 368–378, 2016. DOI: 10.1007/978-3-319-49055-7 33
Information Theoretic Rotationwise Robust Binary Descriptor Learning
369
In particular, recent approaches based on deep learning techniques [25] have shown a strong improvement on the state of the art. However, some kind of “no free lunch” principle applies in that quest. Depending on the targeted application, the desired properties of the descriptor may differ significantly, leading to several trade-offs and design principles. Among others, the following questions are recurrent. Is the computational complexity of paramount importance? Does accuracy matter more than completeness? What class of invariance is required? For instance, in context-based image retrieval, a query image is proposed to a s
Data Loading...
