Globally Optimal Multimodal Rigid Registration: An Analytic Solution using Edge Information

Abstract

Current multimodal registration methods almost always rely on local gradient-descent type optimization strategies. Such registration methods often converge to an incorrect local optimum, especially when the initial misregistration is large. There are monomodal image registration methods that employ global optimization techniques. This paper introduces the use of these global optimization methods for multimodal image registration. The goal is to robustly bring the images into close enough registration that a local optimization method could fine-tune the solution. The method proposed here is based on edge information extracted from the images. Positive results from a modest set of test cases suggests that this approach is promising.

Robust Multimodal Registration Using Local Phase-Coherence Representations

“…A popular alternative to entropy-based methods are feature-based methods that utilize intensity gradients [11][12][13]. These methods attempt to capture structural information based on spatial variations in image intensity.…”

“…Local optimization schemes for monomodal cost functions are less likely to get trapped in local minima. Furthermore, methods exist to decouple the motion parameters and allow for efficient exhaustive evaluation of some cost functions [13]. The main problem with these methods (particularly those based on gradient magnitude) is that they are highly sensitive to image non-homogeneity.…”

An Adaptive Monte Carlo Approach to Phase-Based Multimodal Image Registration

“…A method to alleviate this problem is to evaluate all feasible solutions in the solution space. While methods exist to perform such exhaustive solution evaluation efficiently for low-dimensional solution spaces on a pixel level [18], it is intractable to evaluateT in such a manner for high-dimensional solution spaces or on a subpixel level from a computational perspective. For example, to exhaustively evaluate solutions from the solution space of all possible integer 2-D translations and rotations for two 256 × 256 images would require the evaluation of over 23 million solution candidates.…”

“…Therefore, such methods attempt to find image correspondences in an indirect manner by finding correspondences between extracted features that exist in a common feature space. Features used in such methods include intensity gradient information [17], [18], local frequency information [19]- [21], and shape properties [23]. There are several important advantages to the use of feature-based methods.…”

“…Second, featurebased methods allow more computationally efficient objective functions, such as sum of squared distances and cross correlation to be utilized. In particular, efficient techniques for evaluating such objective functions exhaustively for all possible translations and rotations on a pixel level have been proposed [18].…”