Research On Image Registration Based On Support Vector Machines

Image registration is one of the basic tasks of image processing. It is the process of overlaying two or more images of the same scene taken in different conditions (at different times, from different viewpoints, or by different sensors).Generally speaking, it geometrically aligns two images--the reference and sensed images. Image registration is the base of image processing and application (object recognition, change detection and 3-D image restoration), and the result of image registration can directly effect the further processing work.Now, there are many image registration methods, but neither of them can be fit for all images. So, it is important to find a suitable way for the target image. SVM (Support Vector Machines) is the youngest and the most applied part of SLT (Statistical Learning Theory). Because of its excellent performance in small sample, nonlinear, and high dimension pattern recognization, and its generality, here, SVM has been applied into image registration, and the main points are as follow:First, the basic theory of SVM has been introduced systematically, and then the arithmetics of SVM classification and regression have been described.Second, the conception and flow chart of image registration has been introduced simply, then the principle of geometric transform, the measurement of comparability, resampling & interpolation technology and the common image registration methods have been mainly analysed.Third, the SVC (Support Vector Classification) has been used in image preprocessing, and a newmethod--the two-pass median filter based on SVM for image denosing has been presented. As ourexperiments turned out, this method can preserve more image details like edges while effectively reducing impulse noise, and has good stability.Fially, the respective advantages and disadvantages of different methods of transformation model...