Multiple Image Stitching Technique Research Based On The Feature Points

Image stitching is that these pictures which have the overlapping part between them andis a series of the same scenes with each other is spliced into a wide view angle, highresolution panorama, which requires the greatest degree close to the original image, thedistortion is as small as possible, and no significant seam. With the development of digitalimage technology, high resolution, wide view image is more and more important in a numberof important areas, the panorama image stitching technology can get high-resolution, widefield of view. It can supply solutions to these important areas. So Image stitching technologyis widely used in the stitching of panoramic images, medical image analysis, seabed imageanalysis, remote sensing image analysis, military intelligence analysis and other importantfields.This paper studies the basic frame and key technology of the image mosaic, whichconsists of two key techniques which are image registration and image fusion. according tothe characteristics of the seabed image, and at the same time, integrated the advantages anddisadvantages of each kind of key technology of each method and their applicability,proposed a image registration method which is based on SIFT and a image fusion methodwhich is based on the fade in-out.The purpose of image registration is to get the panorama which is wide view. In thispaper, first of all, extracting SIFT feature points and describing them; Secondly, preliminarymatching using the Euclidean distance ratio; Finally, further matching by random samplingconsistency algorithm and at the same time, calculate the transform matrix between images,then realize the registration between images. Finally, the algorithm of this paper is introducedto the four kinds of position and two actual scene(the seabed image and microscopy image)for registration, the correct rate of image registration reached100%. Image fusion is to get the seamless and high resolution panorama. In this paper, in viewof the current existing shortcomings of the fusion, two aspects are improved:(1)the interestedregion of improvement, for the issues of image overlap area which is much big, this papertake half of the overlapping area as the interested region, improve the processing speed;(2)improving the traditional calculating weight method based on the fade in-out, with gaussmodel calculating the weight method instead of the traditional calculating weight methods,improve to the effect of the cleaning seam. This paper first calculated the interested regionbetween images, then using the gauss model calculating weight, finally fusion and removingthe seam in the interested region using weight. This algorithm can improve the imageinformation entropy when the speed is doubled.The experiments of this paper is implemented on the Windows XP operating system andthe VS2005compiler. The experiments show that the algorithm can not only get imagestitching which is simple positional relationship between images, but also achieve microscopeimages and the seabed image which are the actual changing scene image quickly andaccurately.