Global parametric image alignment via high-order approximation

TLDR: This paper improves both the convergence range and rate using a second order Taylor expansion and shows first order methods to be a degenerate case of the proposed scheme, and extends the scheme using a symmetrical formulation which further improves the convergence properties.

About: This article is published in Computer Vision and Image Understanding.The article was published on 2008-03-01 and is currently open access. It has received 12 citations till now. The article focuses on the topics: Motion estimation & Affine transformation.

Figures

Figure 3: Registration results for theLena image. The image was rotated and translated by(a) (θ = 5◦, δx = 20, δy = 20). (b) (θ = 10◦, δx = 20, δy = 20). A zero initial estimate of the motion was used. The SO3GM converged significantly faster than the SGM.The other algorithms diverged.

Figure 2: Rotation and translation estimation results for the Airfield image. The image was rotated and translated by(a) (θ = 5◦, δx = 20, δy = 20) . (b) (θ = 10◦, δx = 35, δy = 35). A zero initial estimate of the motion was used. The second order formulations converged significantly faster than the first order formulations. The SO3GM algorithm converged twice as fast as the O3GM algorithm.

Figure 5: Registration results for affine motion. The initial estimate of the motion was given by marked by the red X. The SO3GM converged 4 times faster than the GM and twice as fast as the O3GM and SGM.

Figure 9: Estimating affine motion of noisy images. The O3GM and SO3GM allow better registration of noisy images, compared to the regular GM. The initial estimate is the same as in Fig. 5 andσ is the standard deviation of the added noise.