BMVC 2011 http://dx.doi.org/10.5244/C.25.93 GAUGLITZ et al.: IMPROVING KEYPOINTORIENTATION ASSIGNMENT 1 Improving Keypoint Orientation Assignment Steffen Gauglitz sgauglitz@cs.ucsb.edu Matthew Turk mturk@cs.ucsb.edu Tobias Höllerer holl@cs.ucsb.edu Four Eyes Lab Department of Computer Science University of California, Santa Barbara Abstract Detection and description of local image features has proven to be a powerful para- digm for a variety of applications in computer vision. Often, this process includes an orientation assignment step to render the overall process invariant to in-plane rotation. In this paper, we review several different existing algorithms and propose two novel, effi- cient methods for orientation assignment. The first method exhibits a very good speed- performance trade-off; the second is capable of multiple orientations and performs com- parable to SIFT’s orientation assignment while being significantly cheaper. Additionally, we improve one of the existing orientation assignment methods by generalizing it. All algorithms are evaluated empirically under a variety of conditions and in combination with six keypoint detectors. 1 Introduction Keypoint detection, description and matching has proven to be a powerful paradigm for a variety of applications in computer vision, including image retrieval, object recognition, and visual tracking. In many frameworks, this paradigm includes an orientation assignment step [2, 3, 8, 9, 15] to make the overall process invariant to in-plane rotation. While this ap- proach seems to work well and is widely accepted, the orientation assignment is frequently presented as a mere “add-on” to a descriptor, and little work has been devoted to the orien- tation assignment algorithms themselves. In particular, we are not aware of any studies that evaluate and compare different orientation assignment algorithms directly. Our contributions in this paper include: • We propose two novel, very efficient algorithms for orientation assignment; one if a single dominant orientation is sought, and the second capable of detecting multiple dominant orientations. • We present a detailed quantitative evaluation and analysis of our two algorithms as well as four competing algorithms. In this process, we generalize one existing method (Taylor and Drummond [15]) and, by doing so, significantly improve its robustness. Our results entail observations about the orientation assignment problem in general as well as observations about individual algorithms. c  2011. The copyright of this document resides with its authors. It may be distributed unchanged freely in print or electronic forms.