Non-rigid Registration Method Research And Application Of Weather Maps

Image registration is an important topic in the field of computer vision and pattern recognition, having extensive practical value in remote sensing, medical imaging, multi-sensor fusion-based target identification and other research. Meteorological cloud's non-rigid registration is the effective means of non-rigid distortion correction, the matching results will directly affect the quality of the fused image of the weather effects.Based on the reason of non-rigid image distortion meteorological, this paper studied the non-rigid medical image registration algorithm. Based on gray-scale Demons algorithm and based on B-splines FFD model of non-rigid registration algorithm was used to achieve the non-rigid registration of meteorological cloud image, and introduced a certain of improvements on related algorithm.As time-sharing imaging of the meteorological cloud image has the gray-scale's continuity, this paper used the gray-scale Demons algorithm to finish the non-rigid registration studies on meteorological cloud image. Based on the inadequacy of Demons algorithm and through the study of literatures, this paper has been given an improved algorithm for the Demons algorithm. The algorithm uses gradient grayscale of reference image and the floating image, as well as the gray gradient of mutual information to determine the movement of each pixel in the floating image. A simulation experiment proved that the improved algorithm in this paper has more accurate than the original registration of the Demons algorithm.As meteorological cloud image has local non-rigid deformation characteristics, this paper used base on the B-splines FFD model to simulate the cloud's distortions, and multi-level multi-resolution methods and adaptive mesh refinement method to improve the matching speed and accuracy. In order to maintain the image topology, this paper introduced the constraints to restrict the movement of the scope of the control points of the grid and optimized the cost function by using the improved gradient descent method. After a series of image registration experiment, from subjective and objective evaluation of the parameters, it's proved that a series of optimization methods introduced by this paper can reduce the registration time, increase accuracy of registration, and have a certain degree of enforceability.