Research On Technologies Of Image Mosaic Based On Improved SIFT Algorithm

The image mosaic technology has been one of the hotspots in areas of digital image processing and computer vision.It has been widely applied to various fields of national defense and national economy.The main task of this technology is to mosaic a collection of partially-overlapped images into a high-resolution panoramic image with wide range of view.Through nichetargeting research,this paper analyzed deeply various key technologies involved in the process of image mosaic and designed a set of automatic stitching technologies of image fragments based on ASIFT.The work done and the results obtained in this paper are listed as follows:1.Introduces the the concept of the image mosaic and its technology process,and analysis the potential applications and overseas and domestic research status of this technology.States the practical significance of studying automatic stitching technologies of image fragments based on ASIFT,and determined the main research contents of the paper.2.Analysis the principles and characteristics of the ASIFT feature extraction algorithm.Based on the anti-affine invariance of ASIFT algorithm and the speed advantage of SURF algorithm,used SURF algorithm to extract and match features and then to improve ASIFT algorithm when ASIFT algorithm was simulating affine images.Described the similarity measure of the image feature points matching,search strategies,eliminating false match points and other key technologies.Through experimental verification,under the circumstances that the tilt angle is comparatively large and affine deformation is serious,if both matching effects and matching speed should be taken into account,the feature matching performance of the improved algorithm based on ASIFT is optimal.3.Analysis the calculation problem of the image mosaic model parameter.Demonstrated the camera image-forming model and the image mosaic quasi-geometrical registration model according to the principle of the camera image-forming.Based on that method,presented the solution of the image global registration parameter matrix.Through the impact of multi-slice image mosaic error accumulation on mosaic effect,analyzed the validity of sparse LM algorithm adjustment optimization methods.4.For the phenomena of overlap region gray level "Jump" and blurring the image that were caused by difference in radiation quantity and geometrical registration errors during image mosaic process,analyzed and compared several image fusion methods.The experiments verified that multi-resolution fusion algorithm could achieve smooth transition in overlapped areas and generate a clear seamless panorama.