Key Technology Of Wide Angle Image Mosaic Research And Implementation

Restricted by the small angle of view,common camera can only shoot a local part of the wide target scene.In order to get a wider view under the condition of maintaining high space resolution, image mosaic technology is required.Image mosaic is the art of stitching some overlapped pictures into a bigger one after registration and fusion,which has wider visual angle,higher resolution and whole information of original image sequence.During these two or three decades,image mosaic technology has become more and more advanced,and it plays a big role in wide range area exploration, microscopic image analysis,virtual reality,panoramic monitor,video compression and other fields.Most of existing study of image mosaic focus on expanding horizontal angle of view and neglect the vertical direction.In this paper,based on the optical lens detection project and foreign object detection project,the study object are the image sequence obtained by a 2D scanning of microscope and common camera.The theories and methods of multi-row and multi-column image registration based on phase correlation and SURF are studied in depth.This paper focuses on the basic processes and key technology of image mosaic,and accomplishes the following research work:(1)This paper studies the basic principle of phase correlation.Due to the periodicity of Fourier transform,it's error-prone to calculate translation parameters directly by the returned value of pulse signal position.Hence,it's more accurate to calculate the cross correlation coefficient of four possible position separated,and get the final translation parameter through the position corresponding to the maximum.And this paper proposes using the maximum likelihood method to estimate the range of image overlapping area,in order to eliminate the wrong translation parameter.(2)Aiming at the global accumulated errors caused by local image registration error of multi-row and multi-column microscopy images,this paper uses a global registration method based on translation transform.This method first uses the whole translation parameters and the global coordinate of each image establishing a error energy function,then minimizes the error energy function to eliminate global accumulated errors.(3)Aiming at the problem that the panoramic image of microscopy images with high resolution is always too large to be loaded or displayed by common picture viewer,this paper proposes an image block management method based on the theory of tiered and blocking storage hierarchical pyramid,which reduces the memory requirement,and realizes real-time browsing.(4)Registration method based on feature is studied,and make the comparative analysis of the classical operator SIFT and SURF. Establishing K-d tree with whole feature vectors and accomplishing feature matching through BBF queries algorithm.And confirms the adjacent images of each image by building a probability model, then finds the largest connected component of image sequence. Finally, using the L-M algorithm to achieve bundle adjustment to eliminate the global accumulated errors.