Research On Technology Of Image Stitching Based-on Computer Vision

With the development of digital technology,imaging equipment is becoming more and more popular and the demand for image information is also getting higher.Ordinary cameras can't meet the requirements of high-resolution and wide-angle and wide-angle cameras are expensive,complex and not universal.Therefore,image stitching has been widely concerned.Image stitching is mainly divided into image registration and image blending.Image registration is to find the corresponding relationship between images and images from different perspectives can be projected onto the same plane.In the process of image blending,the overlapping areas of images are aligned,merged and optimized to form a high-quality panoramic image.The existing image stitching methods have different advantages and application scenarios.For scenes with large visual difference or moving target,the existing image stitching methods often have the problems of foreground fracture,double shadow ghost and so on.The stitching quality is low.In order to improve the quality of stitching and meet the visual requirements of human eyes,this paper proposes an effective image stitching method,designs and builds a panoramic image stitching system.The specific research contents are as follows:(1)Propose an image registration method based on improved SURF.At the stage of describing the feature points,the dimension of feature point descriptor is reduced by diagonal dimensionality reduction.At the stage of matching the feature points,several steps are added to delete the wrong matching points,including the nearest neighbor distance ratio method,one-to-many deletion method and angle deletion method.In this paper,a threshold comparison experiment is designed to determine the appropriate threshold range and the proposed algorithm is compared with common registration algorithm.The results show that the proposed algorithm can improve the accuracy and speed of image registration by 5% ? 20%.(2)Design an image stitching method based on improved seam cutting method.The traditional optimal suture energy criterion is improved.In addition to the intensity of color difference and structure difference,the similarity of pixels on both sides of the stitching image is taken into account.Image texture features are introduced to find the most suitable area for image blending which makes the panoramic image transition more natural.In this paper,a comparative experiment of stitching methods is carried out.The results show that the proposed method can eliminate stitching traces and ghosting.The panoramic image contains more information and is more clear and complete.(3)Build a panoramic image stitching system.On the one hand,the system can match and mosaic two images through improved SURF algorithm and improved suture algorithm.On the other hand,the system provides traditional SURF and asymptotic algorithms for users to compare the effect.In addition to image blending,the system also includes brightness adjustment,image evaluation and other modules to meet more requirements for image processing.Furthermore,the system provides a simple,clear and user-friendly interactive interface.