Robust image registration based on local standard deviation and image intensity

Image registration is required whenever images taken at different times or from different sensors need to be compared or to be integrated. The transformation is computed by minimizing a suitable cost function, that formulates the dissimilarity between the transformed image and the template image. Typically, image intensity is used in cost functions to measure difference between images such as L₂-norm. However, these cost functions may perform poorly at important locations due to variation of contrast and/or brightness between the two images. Other well-known approaches, feature based approaches, completely neglect the image intensity, consequently, the registration result may be very poor. In this paper, we propose to use muti-features in the cost function consisting of image intensity and local standard deviations. This cost function is robust against global as well as local brightness variations. It is tested by medical image as well as computer graphic images. Based on the simulation results, the proposed algorithm outperforms two state-of-the-art algorithms. Moreover, it is suitable for high resolution and low resolution images.