Robust Binary Feature Using the Intensity Order
Binary features have received much attention with regard to memory and computational efficiency with the emerging demands in the mobile and embedded vision systems fields. In this context, we present a robust binary feature using the intensity order. By a
- PDF / 3,073,191 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 18 Downloads / 246 Views
roduction
Finding correspondences between images is a fundamental step in many computer vision methods, such as object recognition, image retrieval, and widebaseline stereo. The key component of a correspondence search is to extract invariant image features, and many computer vision researchers have focused on extracting invariant image features based on their importance. The main concerns with regard to invariant features are localization accuracy, invariance to geometric and photometric deformations, and distinctiveness to be correctly matched against a large number of features. SIFT [1] and SURF [2] are known as the best known and most widely used methods among all various image features. They find scale-invariant distinctive image regions and represent local regions using feature vectors which are invariant to rotation and illumination changes. The discriminative power of SIFT and SURF has been validated in many computer vision techniques, and variants of these methods are widely used for robust image representation. Two other important factors pertaining to invariant features are time and space efficiency levels when detecting, matching, and storing features. Recently, Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-16865-4 37) contains supplementary material, which is available to authorized users. Y. Choi and C. Park—The first and the second authors provided equal contributions to this work. c Springer International Publishing Switzerland 2015 D. Cremers et al. (Eds.): ACCV 2014, Part I, LNCS 9003, pp. 569–584, 2015. DOI: 10.1007/978-3-319-16865-4 37
570
Y. Choi et al.
demand has increased for such efficient image features, as mobile and embedded vision systems are emerging for visual searches [3] and for direct 2D to 3D matching [4,5]. Also, for mobile visual search applications, the amount of data sent over the network needs to be as small as possible so as to reduce latency and lower costs. Several binary features, such as BRIEF [6], ORB [7], BRISK [8] have been developed to describe local image regions with small binary strings which can be matched much faster with the Hamming distance compared to SIFT. However, despite the effort and advances in this area, SIFT has remained the best option for various deformation tasks apart from non-geometric transforms [9]. In this paper, we aim to extract binary features with a method that can achieve matching performance levels comparable to those of SIFT and SURF with even less storage than that required for existing binary features. We apply FAST-like binary tests [10] to reject non-feature regions quickly and present an efficient approximation of the Determinant of the Hessian for robust feature detection with high repeatability. Motivated by earlier work [11], we employ ordinal descriptions of local image measurements for robust representations of feature regions. The ordinal description encodes the rank order of each measurement and is therefore invariant to monotonic deformations of the measurements. Also, an ordinal de
Data Loading...
