Research On Image Stitching Technology Based On SIFT

Image mosaic is an increasingly popular research area among image processingand computer vision.The purpose of image mosaic is to merge two or more imagesthat have overlap part of the same scene into a high-resolution panoramicimages.Image mosaic can solve the contradiction problem between the filed of visionand the resolution.Image mosaic has a very wide range of applications in many fileds,such as medicine image, virtual reality technology, video compression, video retrieval,remote sensing technology and so on. In general,Image mosaic has steps asfollows:image acquisition,image registration and image fusion,etc. image registrationdecide while image mosaic have succeeded among these steps.Base on the study ofbasic principle and methodology of image registration,we find image registrationbased region need to take advantage of all the pixel information.The method isheavily depended on the gray information of the image,so it is very sensitive to thechange of gray.Image registration based on the frequency required large ratio ofoverlap between the registrating images,and the calculation of the method is toolarge.But image registration based on the characteristics is not sensitive to the changesof light, and they are not easily affected by factors such as rotation, zooming and soon.In addition, the number of the feature point is far less than the number of thepixels,greatly reduces the computation of the matching process.This have helped toimprove the matching speed.So this paper realize image registration based on thecharacteristics.This paper first introduces the the classification of image mosaic technology,and the application field and current studies, thus has show the wide applicationprospect of image mosaic. This article simply introduces the basic knowledge ofimage matching, including as follows: First, the camera model and the three methodsto get images; Second, suppressing the noise of the image, Then comparing the effectof the filters that recude the noise through the experiment and decide to use themedian filter to denoising at this paper; Third, we introduced briefly a few classicalapproaches about pixel-level image fusion. This paper lays emphasis on the research of the SIFT feature point detectionand matching technology, and analyze the performance of every step of the SIFTfeature point detection through the experiment which draws the conclusion that SIFTfeature points registration algorithm spend more time on the process of detecting anddescribing the feature points. According to the problem that detecte SIFT featurepoints is more time-consuming, this paper put forward reduce the number of octaveand interval when we constructe the gaussian pyramid in the traditional algorithm.Weimplement the improvement algorithm and traditional algorithm of image mosaicthrough the experimental simulation. By the comparison of the experimental results,we find that the improved algorithm increase speed efficiently of the imageregistration and the image mosaic and it don’t reduce the accuracy of the imageregistration and the image mosaic at the same time. The traditional algorithm is onlyfocus on the overlapping area but ignore other area of mosaic image, and this makethe brightness of mosaic image diaaccord. According to this problem, this paper putforward a new algorithm to distribute the brightness of the whole image. Theexperimental results show that the improved algorithm can make the brightness to theagreement of the whole image, so that it can make people’s image visual experiencemore satisfy.