Research On Sub-Pixel Image Registration Methods With Noise

Image registration, which is a fundamental problem in image processing, is a method to match two or more images in the same scene. Those images are taken at different sensors, different viewpoints, different times or different conditions. Image registration technique has been used in many fields, such as computer vision, medical diagnosis and treatment, remote sensing, image reconstruction and image fusion and so on.There are many image registration methods, and the transform domain-based approach is commonly used compared with other methods. The translation, rotation and scale of the image can be showed in Fourier transform domain based on Fourier transform properties. This thesis focuses on this method. The thesis is organized as follows:Firstly, the theory of image registration is presented, which is composed of the mathematical model, the space transformation model, the process and methods of image registration, and the image resample and the evaluation standard of registration algorithms.Secondly, the transform domain-based approach is analyzed. Traditional phase correlation algorithm can only process the parameter of translation. Fourier-Mellin algorithm, which based on phase correlation, can detection translation, rotating and scale parameters. And combined with edge features, the algorithm based on edge features and phase correlation is discussed.Thirdly, the sub-pixel motion is analyzed, and the local phase correlation method is presented compared with sub-pixel registration algorithms. The result shows that it has excellent robustness and anti-noise ability by experimental analysis.Finally, conclusions of the work and suggestions of further research are given.