backlinks: /PaperReviews

**[SZGW05]** Siavash Zokai and George Wolberg. Image Registration Using Log-Polar Mappings for Recovery of Large-Scale Similarity and Projective Transformations. *IEEE Transactions on Image Processing*, 14(10):1422—1434, 10/2005.

- Most images of 3D scenes do not only differ by a perspective transformation. For that to be true, constraints must be placed on camera motion, or the objects photographed must be far away.
- Brown's framework for registration techniques:
**[LGB92]**L. G. Brown. A survey of image registration techniques.*ACM Comput. Surv.*, 24(4):325—376, 12/1992.

- Zitova and Flusser's overview of image registration methods:
**[BZJF03]**B. Zitova and J. Flusser. Image registration methods: A survey.*IVC*, 21(11):977—1000, 10/2003.

- Correlation ratio as similarity measure:
**[ARGM+98]**A. Roche, G. Malandain, X. Pennec and N. Ayache. The correlation ratio as a new similarity measure for multimodal image registration.*Proc. 1st Int. Conf. Medical Image Computing and Computer-Assisted Intervention*, pages 1115—1124. 10/1998.

- Hierarchical image registration. Mapping modeled as perspective transformation. Algorithm estimates parameters of perspective transformation. Parameters selected to minimise sum of squared differences. Computer iteratively in coarse-to-fine hierarchical framework, using modified Levenberg-Marquardt. Claims sub-pixel accuracy.
- Drawback of above algorithm: fails for large parameters (rotation, offset etc.). Introduce log-polar registration — gives good estimate with which to initialise optimisation-based method.
- Overviews of Levenberg-Marquard algorithm, Log-Polar transform, Fourier-Mellin transform and method for feature-based image registration.
- Feature based: RANSAC-algorithm for outlier rejection (http://en.wikipedia.org/wiki/RANSAC). These methods fail on non-textured images.
- Presents modified LMA, which does not require re-calculation of Hessian with each iteration.
- Mention Fant's resampling algorithm. Also see
**[GW90]**G. Wolberg. Digital Image Warping.*Los Alamitos, CA: IEEE Computer Soc.*, 1990.

- Describes global registration using log-polar transform. Their strikes me as being overly computationally expensive? Mentioned in their section on future work, that one could potentially only examine areas of high variance.
- Claim better success than SIFT with comparable computation times.

rainbow