Non-rigid image registration using local histogram-based features

This paper proposes a new non-rigid image registration method based on the formulation of the Demons algorithm. The proposed method utilizes combined geometric moments of local histogram to form new feature images. It greatly improves the accuracy of the original Demons algorithm, which is easy to get trapped at local minima during optimization. The local histogram-based features are rotation invariant and can capture sufficient spatial image information. During the registration process, local histogram-based feature images are built to substitute the original intensity images. This can reduce the possibility of being trapped at local solutions, and consequently improve the registration accuracy. The experimental results on both the synthetic image and real MRI image show that the proposed method can achieve higher accuracy than the Demons algorithm especially when the images are noisy.