A Novel Fractional Order Derivate Based Log-demons with Driving Force for High Accurate Image Registration

Abstract

Image registration methods based on Thirion's demons method update displacement field by the image gradient obtained by integer order derivate. However, the fractional order derivate is superior to integral order derivate for computing image gradient under weak texture or smooth regions. To obtain high accurate image registration, we propose a new fractional order derivate based Log-Demons with driving force. We design a new fractional order derivate convolution mask based on Grünwald-Letnikov (GL) definition to get accurate image gradient. Then, we integrate fractional order derivate into Log-Demons with driving force. The experiments on synthetic and MRI brain images validate that the use of fractional order derivate to compute gradient not only improves the registration accuracy but also speeds up the registration process.