pilu-1997 - A Direct Method for S tereo Correspondence...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
A Direct Method for Stereo Correspondence based on Singular Value Decomposition Maurizio Pilu Digital Media Department Hewlett-Packard Research Labs Filton Road, Stoke Gifford Bristol BS12 6QZ (UK) email: [email protected] hpl. hp.com Abstract ~ This paper proposes an new algorithm for match- kg point features across pairs of images. Despite the ell-known combinatorial complexity of the problem, his work shows that an acceptably good solution can e obtained directly by singular value decomposition of n appropriate correspondence strength matrix. The qpproach draws from the method proposed in [a] but, qesides suggesting its usefulness for stereo matching, zp thzs work a correlation-weighted proximity function ib used as correspondence strength to specifically cater fbr real images. i 4 Introduction I , The problem of feature correspondence across two & more images is well known to be of crucial impor- tbnce for many images analysis tasks. Reliable inter- i age feature correspondence - and its closely related p t oblem of image registration - is needed, just to cite a1 few, by structure-from-stereo approaches, motion aflalysis and tracking, image mosaicing, object pose a' d self-motion estimation. Recently, there has been a boost of interest in the cwespondence estimation problem due to the devel- o ment of the Fundamental Matrix theory [3] and its t ij emendous practical implications in the analysis of ubcalibrated stereo pairs and image sequences. If the fdndamental matrix is known, reliable and fast feature c rrespondence can be obtained in general situations. However, in order for the fundamental matrix to be c mputed one needs a good initial set of feature cor- r 1 spondences (either lines, points or both [ll]). ~ There are two schools of thought for solving the fea- ture correspondence problem. In the first one, features a+ detected in one image and then correspondences fot each of them are sought for in the second image, gqnerally via multi-scale techniques. In the second P 106316919/97 $10.00 0 1997 IEEE 26 1 approach, which the present work addresses, features are detected independently in both images and then matched up usually by relaxation (see, e.g., the clas- sic [SI). Incidentally, recent state-of-the-art work on the fundamental matrix estimation [14, 111 follows this latter avenue for achieving initial correspondences. This paper proposes an new neat and simple al- gorithm for achieving feature correspondence across pairs of images. Despite the well-known combinato- rial complexity of the problem, this work shows that an acceptably good solutions can be obtained directly by singular value decomposition of an appropriate cor- respondence strength matrix. The approach is largely inspired by the clever algo- rithm proposed by Scott and Longuet-Higgins for find- ing corresponding features in planar point patterns [8], which has been inexplicably overlooked in this area.
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.
  • Spring '11
  • Li
  • Singular value decomposition, correspondence strength, correspondence strength function, S. Ullman, correspondence strength matrix

{[ snackBarMessage ]}

Page1 / 6

pilu-1997 - A Direct Method for S tereo Correspondence...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online