Epipolar geometry estimation method based on maximizing image correlation

Abstract

AbstractThe authors proposed a method for estimation of epipolar geometry parameters based on correlation between two images. This method is characterized by the fact that, unlike conventional epipolar geometry estimation methods using correspondence of characteristic points between the images, it uses parametric block matching correlation with the results of block matching in the image approximated by a two‐dimensional function. This technique involves (1) performing block matching between the two images and obtaining the parametric block correlation function, and (2) minimizing the obtained parametric block correlation function by a robust estimation based on the epipolar geometry parameters, and estimating the epipolar geometry parameters. In the current study, the authors discussed a technique for estimating the epipolar geometry parameters by linear solution method, and then a technique for estimating the epipolar geometry parameters based on nonlinear solution method with those parameters set as initial values. Finally, it was shown by tests using real images that this method is suitable for estimating epipolar geometry parameters with highly correlated images. The tests confirmed that this technique is efficient for improving image quality in the course of image synthesis or improving encoding efficiency during image encoding. © 2003 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 86(3): 73–81, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjb.10137