Title: QuadSIFT: unwrapping planar quadrilaterals to enhance feature matching
Authors: Filax, Marco
Gonschorek, Tim
Ortmeier, Frank
Citation: WSCG '2017: short communications proceedings: The 25th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision 2016 in co-operation with EUROGRAPHICS: University of West Bohemia, Plzen, Czech RepublicMay 29 - June 2 2017, p. 7-15.
Issue Date: 2017
Publisher: Václav Skala - UNION Agency
Document type: konferenční příspěvek
URI: wscg.zcu.cz/WSCG2017/!!_CSRN-2702.pdf
ISBN: 978-80-86943-45-9
ISSN: 2464-4617
Keywords: projektivní transformace;perspektivní zkreslení;detekce funkcí;SIFT funkce
Keywords in different language: projective transformation;perspective distortion;feature detection;SIFT feature
Abstract: Feature matching is one of the fundamental issues in computer vision. The established methods, however, do not provide reliable results, especially for extreme viewpoint changes. Different approaches have been proposed to lower this hurdle, e. g., by randomly sampling different viewpoints to obtain better results. However, these methods are computationally intensive. In this paper, we propose an algorithm to enhance image matching under the assumption that an image, taken in man-made environments, typically contains planar, rectangular objects. We use line segments to identify image patches and compute a homography which unwraps the perspective distortion for each patch. The unwrapped image patches are used to detect, describe and match SIFT features. We evaluate our results on a series of slanted views of a magazine and augmented reality markers. Our results demonstrate, that the proposed algorithm performs well for strong perspective distortions.
Rights: © Václav Skala - UNION Agency
Appears in Collections:WSCG '2017: Short Papers Proceedings

Files in This Item:
File Description SizeFormat 
Filax.pdfPlný text7,93 MBAdobe PDFView/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/29729

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.