Sequence Of Infrared Image Automatic Registration Algorithm Research

The fundamental problem of the image registration is to find out a best image trans- formation model that is used to correct different forms of deformation s between images. I- mage deformations can be characterized that the imaging grayscale of the same surface feature was different or target features appeared relatively translation, rotation, scaling, and so on, and so we need to conduct image registration between images. With the dev- elopment of science and technology, digital image s as spatial information carrier are mor- e and more easy to access through various channels, and it will be an important topic ho- w we effectively extract the useful information in many mixed digital image s in the field of digital image processing. Therefore, in the processing of digital image, the image registr- ation techniques in the foundation position is worth for us to deeper research and discussAimed at mainly USES the way of registration technology between images by manu- al and semi-automatic registration whose running time is longer consuming and efficiency is lower in the existing field of image registration, the paper puts forward the improved c- orrelation coefficient automatic registration algorithm and SIFT automatic registration alg- orithm. Taking three groups of aerial infrared images with different heights and scene se- quences obtained from Yantai for example, we used the above two kinds of automatic r- egistration algorithms to correct the sequence images and eliminate the distortions to mat- ch the image background, and realize the encapsulation of registration algorithm through developing and designing convenient algorithm implements. Improved correlation coeff ic- ient automatic registration algorithm model mainly use the one combined the gray with ch- aracteristics, employ the method combined Forstner operator with correlation coefficient to obtain the feature points on the images after uniform grid matching, to achieve image matching and precise matching between operations. SIFT algorithm is the one that make matching feature points scale and rotation unchangeable through obtaining the chara- cteristic points in the scale space extraction. The registration algorithm includes main feature point detection, the algorithm generated features description and feature matching, and establishes the connection between the images.Compared with the traditional artificial registration algorithm, the improved correiat- ion coefficient automatic registration algorithm and SIFT automatic registration algorithm improve the processing speed and registration accuracy of registration algorithm. The experiments proved that the results and accuracy of the improved correlation coefficent automatic registration algorithm were influenced by the flying height, transform model and frame interval; For the sequence infrared image datas from high altitude, it was got steady registration results in a certain frame frequency interval, and the quadratic tran- sformation model is superior to the linear transformation one. SIFT automatic registration algorithm was more steady, and the one based on quadratic transformation model within a certain range has better stability.