Research On The Technique Of Geometric Calibration And Color Correction For Image Stitching

In the field of computer vision,image stitching has been more and more mature after years of development.And its applications involve aerospace,remote sensing mapping and many other areas.Nowadays,researches on image stitching mainly focus on the contents of image,including geometric information and color information.Geometric calibration is to unify the contents of the overlapping regions while preserving the completeness of geometric information of the original image.However,the existing methods improve the accuracy of geometric calibration,but also lead to the projective distortion problem.Therefore,it is necessary to find a way to improve the accuracy of geometric calibration while suppressing the deformation of the projection.For this problem,this thesis uses the projective transformation to perform image matching,and uses the similarity transformation as the shape optimization criterion to constrain the projective distortion.This hybrid method takes into account the advantages of both projective transformation and similarity transformation,which is to ensure the accuracy of image matching,while constraining the image projective deformation.At the same time,this thesis also improves the image splicing process according to the existing methods,and makes use of the automatic mesh division to make it more suitable for splicing multiple images.Color correction is mainly to adjust the difference of the color and brightness information between the images.Traditional image fusion methods are difficult to handle the stitching of multiple images with better performance.And color correction methods dealing with tedious calculation,there are not combined with the characteristics of image stitching.In order to deal with the problem of color difference in image stitching better,this thesis proposes a color correction method for image stitching based on its characteristics.Color regions can be located by using the feature information existing in the image stitching,and the matched color regions can be selected to obtain more accurate color information.Then,for the obtained color information,we consider the actual color range to calculate the color transformation model.By obtaining the color transformation model of each color channel,the color differences between the images can be rationally unified under the same color and brightness conditions.