Study Of Wood CT Image Registration

Image registration is a registration and superposition process which uses two or more images for the same scene obtained at different time, by different sensors(imaging equipment), or under different conditions (weather, illumination, camera position and angle, etc.). The main method of image registration can be divided into three categories: gray level image-based registration method; registration based on image characteristics, such as edge feature and corner point feature; image registration based on frequency domain, such as the method based on fast Fourier transform. The current image registration techniques are used in computer vision, medical diagnosis and adjuvant therapy, remote sensing images, 3D modeling, image mosaic and so on.This paper analyzes the three common methods of image registration and use them to registrate wood CT images. First, uses the method based on mutual information to registrate, and improves traditional PV interpolation algorithm and genetic algorithm, the improved algorithm can effectively suppress the generation of local extremum, while solves the premature problems of the genetic algorithm; then we discuss the image registration method based on frequency domain transform, and improve image registration algorithm using a coarse to fine the pyramid hierarchical registration strategy which is based on the traditional Fourier transform, this algorithm effectively improves the accuracy of image registration, image registration accuracy enables sub-pixel level; Finally the method based on the Harris feature point image registration is discussed. We propose a triangle algorithm to determine relationship between the feature points, and according to the relationship of the feature points we can obtain the parameters for the wood CT image registration. This algorithm can greatelly reduce the calculation, so registration is faster but more affected by noise.