Research On Fast Panorama Mosaic Algorithm Based On Cylindrical Surface Mapping

Panoramic image mosaic technique performs image processing through enhancement,registration and fusion,and combines sequence images(video)with a certain degree of coincidence and taken at same place and time into a large panoramic image with all the input image information,which can be applied to a variety of military and civil applications.To obtain a high quality panoramic image without distortion and meet real-time requirement is one of research hotspots in field of image stitching.Panoramic image based on cylindrical surface mapping model has the advantages of uniform image quality,high detail and reality,and researches of panoramic image mosaic algorithm based on cylindrical surface mapping are performed to solve the problems of distortion,ghosting and seam during stitching,and improve efficiency in this dissertation.The main research work are listed as follows:Firstly,to solve the problem of focal length estimation in cylindrical panoramic mosaic and improve the efficiency of features matching in image mosaic,a fast feature points matching algorithm based on prediction is presented to meet real-time requirements,and a focal length correction algorithm based on homography is proposed.High-quality images are first selected as the images to be stitch using a gradient-based image sharpness detection algorithm,Harris corners are then extracted and HOG(Histogram of Oriented Gradient)descriptor is calculated.A fast feature points matching algorithm based on prediction is performed.A simplified focal length estimation algorithm based on pure rotational motion is used to estimate the initial focal length,and a focal length correction algorithm based on homography is used to obtain a more accurate focal length value.Finally,the planar image is projected onto the cylindrical plane and the panoramic image is synthesized.Stitching experimental results demonstrate that,the proposed method can synthesize high quality panoramic image rapidly and have little image distortion.Secondly,to solve the problem of ghosting and stitching seam in the fusion image caused by moving object or the inaccurate registration area in stitching process,an improved optimal seam algorithm based on difference image weighting is proposed and multi-pattern partitioning algorithm based on multi-resolution image fusion algorithm is taken to eliminate stitching seam problem.The overlapping region of fusion image is first divided into transition regions and seam search regions;Differential image weighted optimal seam algorithm is then performed in the seam search region;Multiresolution image fusion algorithm is performed in overlap region,and weighted average image fusion algorithm is used to eliminate the seam line in transition regions.Experimental results demonstrate that,the improved optimal seam algorithm can effectively avoid the seam line pass through the region where the moving object exists and the region where the registration error exists,and synthesize high-quality panoramic images.Finally,the main work and further research of the thesis is summarized.