The Research Of Image Registration Methods Based On Wavelet Transformation

As a rising and having abroad application prospect subject, remote sensing is the only method which could offer the global dynamic observation data. With new type sensors welling up continually, different spatial resolution and different temporal resolution images could be obtained from different sensors. These images offer much cursory or precision observation data. At present, it is hard to registry these images efficiently, and direct influence much data's using efficiency. So, how to registration these images efficiently and fusing these existing resource is a very significant research topic.Firstly, the paper introduces the characteristics of remote sensing imaging, common using satellites' fundamental state and the significance of remote sensing images' registration. The paper not only gives image registration's definition and routine procedure, but also analyses the present status and existing problems in image registration domain. For ensuring image registration's precision, it needs to apply geometric correction to images with distortion. The paper introduces several common used methods, and validates polynomial correction. Image registration methods are commonly classified into two types: the methods based on grey level information and the methods based on feature information. The paper compares several traditional image registration methods and summarizes their relative merits. Based on the researches of wavelet transformation's basic theory, Mallat algorithm and image wavelet transformation's frequency characteristic, the paper presents two image registration methods basing on wavelet transformation: a registration approach using wavelet transform to extract edge feature points and a registration approach based on wavelet transformation and mutual information. The methods refers many other knowledge such as edge detection, edge correlation, affine transformation, local matching and iterated algorithm et al, all these are described and researched in the paper. Finally, several experiments are carried to registry some types of images. The result testified that these two approaches have high accuracy and high reliability. This also indicates the validity and application value of the research result.