Image Registration Based On Maximum Mutual Information And Phase Correlation Theory

Image registration, which is a fundamental problem in image processing, is a process to match two or more images of the same scene taken at different times, from different viewpoints, or by different sensors.Image registration has been widely used in the military, remote sensing, medicine, computer vision and other fields.Image registration techniques can be divided into manu-registration and auto-registration. Auto-registration, which is also the ultimate development goal of the image registration technology, do not need anybody to interfere in. The current method of auto-registration can be divided into four major categories: intensity based, transform domain based, template based and feature based registration. The mutual information based image registration can be used in almost any different modes image, whose accuracy is generally higher than segmentation-based approach, which does not need to do feature extraction and other pre-processing, and which needs less man-machine interaction. However, it has larger computational, lower operating efficiency, and has a shortage of"local maximum"introduced by the absence of spatial information and the interpolation error. To solve this two points,this paper first use the histogram equalization instead of traditional histogram on the calculation of mutual information, which is good to inhibit the"local maximum". And then,the phase correcting based on Fourier transform theory is used as a pre-registration, which can reduce the iteration scope of Powell optimization without affecting or even improving the accuracy of image registration and avoiding the local maximum.