Pattern Recognition 40 (2007) 1432 – 1450 www.elsevier.com/locate/pr Development and evaluation of fast branch-and-bound algorithm for feature matching based on line segments Lik-Kwan Shark a , ∗ , Andrey A. Kurekin b , Bogdan J. Matuszewski a a ADSIP Research Centre, Department of Technology, University of Central Lancashire, Preston PR1 2HE, UK b Department of Computer Science, Cardiff University, Cardiff CF24 3AA, UK Received 25 July 2005; received in revised form 28 September 2006; accepted 30 October 2006 Abstract By extending the previously proposed geometric branch-and-bound algorithm with bounded alignment for point pattern matching, the paper presents the development and evaluation of a new and fast algorithm for image registration based on line segments. Using synthetically generated data sets with randomly distributed line segments and hard test cases with highly symmetric line patterns, as well as real remote sensing images, the developed algorithm is shown to be computationally fast, highly robust, capable of handling severely corrupted data sets with considerable line segment position errors as well as significant fragmented and spurious line segments in the images to be matched.  2006 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved. Keywords: Feature matching; Hausdorff distance; Line segment matching; Branch and bound; Bounded alignment 1. Introduction Image registration can be achieved either directly by com- parison of the similarities between images via correlation or indirectly by matching of geometric features (such as con- trol points, corners, line segments, and arcs) extracted from images [1]. Compared with the direct methods, the indirect approaches are less sensitive to background noise, image clutter, intensity changes and contrast variations, as well as offer lower computational load. This paper focuses on geo- metric matching of line segments, which requires search of the correspondence between two sets of line segments and estimation of the relative position, orientation and scale be- tween them. Possible applications of line segment matching include industrial inspection to fuse non-destructive testing images of different modalities [2], high lossless data com- pression by structure decomposition [3], robot navigation to compare features in the images with the known landmarks ∗ Corresponding author. Tel.: +44 1772 893253; fax: +44 1772 892915. E-mail address: lshark@uclan.ac.uk (L.-K. Shark). 0031-3203/$30.00  2006 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.patcog.2006.10.022 [4,5], multimedia to retrieve images from a database via their geometric contents [6,7], and remote sensing to match images acquired with digital maps [8,9]. A number of approaches can be applied to match line segments extracted from an image with line features con- tained in its corresponding model. One approach is based on geometric hashing [10], whereby invariant geometric features extracted from an image are used as indexing keys to search possible matches in a precompiled hash ta- ble containing encoded geometric features of the model. Drawbacks of this approach lie in its sensitivity to image noise and inherent quantization errors introduced in con- struction of the hash table. Furthermore, it has been shown to have high implementation complexity and consuming large memory space [7]. Another approach for line segment feature matching is based on a combination of simulated annealing and local search [11], whereby the former is a randomly guided method used to find a good correspon- dence between two sets of line segments and the latter is used to find the best match by trial and error in its locality. The ability to match fragmented, noisy and cluttered data is one of the advantages of the local search algorithm [12].