Research On The Image Mosaic Algorithms Based On Corner Detection

With the development of technology，the images get from the ordinary photographicequipment cannot satisfy people’s requirements of wide field of vision and high definition ofimages,so the irreplaceable advantages of image mosaic technology in many areas such asimage processing and motion recognition have become increasingly prominent. Because ofpeople’s requirements of real-time and influence of objective factors such as illumination ingetting images, the study on the algorithm having both faster registration speed and higherregistration accuracy becomes research priorities of image mosaic area.First, this thesis analyzes the developments of image stitching algorithm in the domesticand international, introduce a variety of currently available image stitching algorithm learnthe theory of image stitching algorithm based on corner detection, analyzes the advantageand disadvantages of some kinds of currently corner detection operators. This thesis based onactual, puts improve the accuracy of image mosaic algorithm as the core, considers the speedof image mosaic, improved the the image mosaic algorithm based on corner detection. Themain contents are as follows:(1) By researching the advantages of B-spline function, combined B-spline functions andHarris corner detection algorithm, replace the original algorithm Gaussian window, proposedan improved Harris corner dectection algorithm based on B-spline. This novel algorithm cangreatly reduce detection deviation cased by difficult to determine the size of Gaussianwindow in Harris corner detection algorithm.It can solves the problem of redundancy andimprecise positioning in Harris corner detection algorithm. Experiments show that thisalgorithm can keep the advantages of Harris corner detection,it can also improve theaccurancy of corner detection.(2) In image registration, correlation algorithm can achieve sub-pixel accuracy, but ithave the disadvantages of large computation, image rotation sensitive and so on,so thisthesis combine Harris corner and Rotate Invariability Phase Only Correlation,proposed anovel image registration algorithm based on Harris and Rotate Invariability Phase OnlyCorrelation. Experiments show this algorithm can solve the problem of image rotationsensitivity in phase correlation algorithm. By using Harris corner in the algorithm, it reducesthe boundary effects of phase correlation algorithm by using Fourier transform,it can improvethe speed of image registration,and ensure the accurancy of image registration.Then in thispaper, it uses the image fusion algorithm based on laplace pyramid which belongs toMulti-resolution analysis to fuse images,it has a better result.(3) In image fusion,this thesis use the image fusion algorithm based on Laplacianpyramid which is a multi-resolution analysis algorithm Essentially。This algorithm first decomposes the image in many different layers,then fuses image in these different layers,finally reconstruct the image and get the final fused image.The experiments show that thisalgorithm has a better fusion effect.(4) Participated in the project of "design of interactive panoramic roaming platformbased on image registration and fusion ". In the project, first take a group photos of360degrees to a scene, and then use the improved algorithm to mosaic the image to get apanorama, finally make the virtual system by using the obtained panoramas.