An Image Registration Method With The Edge Characters And Mutual Information

Image registration, which is the first issue should be settled in image data amalgamationand applied in the aspects of remote sensing image processing,medical image analysis and soon, is an significant task in the fields of computer vision and mode identification. It will existrelative move,rotation and different proportion zoom between the same image data gained atdifferent time using different sensor and imaging mode, so we must settle the imageregistration first of all.We seek a kind of space transformation for the space coherence achieving bycorresponding point of the same objectives which represents the two images. The result ofregistration should make all the points or practical applied points at least in both imagesachieve matching. This is the course of image registration.At present, there are two methods of image registration, one is based on pixel (grey scale)and the other based on the character of image. Recently, the image registration method basedon mutual information which bears high registration precision and can achieve sub-pixel levelwithout segmentation predisposal, has universal application. But there are also many limits inthis method, including false matching brought by lots of extremum arising in mutualinformation calculation and great influences brought by grey scale level of both images to theresult of registration. Because Mutual information makes use of grey scale of image only butneglect other characters, so that it is sensitive to image yawn.This paper is based on the largest mutual information by using image registration methodbased on gray level and character. We examine the edge of image for getting edge image andcalculate the edge mutual information and edge correlation deviation of it after calculatingmutual information of original image, then define a kind of new measure function to directseeking better variable parameter. Through the test, we find that sensitivity to yawp in newmethod is lower than that in traditional mutual information way, shape of BI Curve inregistration function is excellent with a sharp peak value. The new method is easy forselecting the best registration parameter and possesses exact registration result.Image registration relates to a multi-parameter optimizing problem in itself. For the localextremum, this paper optimizes the registration function using mixing arithmetic of PSOand Powell.