Image Registration Based on Sum of Square Difference Cost Function

There are numerous applications for Image Registration (IR). The main purpose of the IR is to find a map between two different situation images. In this way, the chief objective is to find this map to reconstruct the target image as best as possible. Needless to say, the IR task is an optimization problem. As the optimization method, although the evolutionary ones are somehow more effective in escaping the local minima, their speed is not emulated the mathematical ones at all. In this paper, we employed a mathematical framework based on the Newton method. This framework is suitable for any efficient cost function. Here, we used the Sum of Square Difference (SSD). We also provided an effective strategy in order to avoid sticking in the local minima. As one of the fundamental drawback of mathematical based optimization methods, local minima have been managed properly by our new strategy. Scale and Rotation – as two variables of the problem- have been treated solely in a different iteration. In another word, in every iteration, one of this variable is actually as the only variable of the problem as the other one considered as constant. Results indicate the effectivities of our model.

Keywords:

Image registration , Sum of Square Difference , Root Mean Square Error , Mutual Information