Improvement Of Shape-preserving Half-projective Warps For Image Stitching

Image stitching is one of the key topics in the field of computer vision,and this technology has been widely used in remote sensing images,medical imaging,virtual reality and other fields.SPHP is a spacial combination of a projection transformation and a similarity transformation.It provides good alignment accuracy as projection warps while preserving the perspective of individual image as similarity warps.This paper proposes improves SPHP image stitching algorithm.The main research results are as follows:(1)Improved RANSAC algorithm in SPHP.SPHP uses the robust RANSAC to filter coarse matching point pairs in the semi-projection transform region.Since the RANSAC algorithm has unknown number of iterations,in order to obtain a good registration result,it usually needs to pay a large computational cost.If there is a large amount of incorrect data in the data set,it will also affect the purification accuracy.In order to solve these problems,a pre-possessing is performed when the coarse matching point pairs are obtained,and the double-sided matching is used to filter out the feature pairs which are obviously wrong.After the improvement,it is found that the matching point pairs input into the RANSAC algorithm are significantly reduced,the quality of the matching points is improved,the precision of the extracted matching points pairs is obviously improved,and the operation speed is slightlyimproved.(2)Adaptive luminance algorithm based on feature point pairs.The two images involved in the stitching may be affected by the light intensity and brightness,which leads to the unnatural transition of stitching images using SPHP.For such images,it is transformed into HSV color space first,and the feature point pairs are weighted to collect the color attribute values,and then the weight of adjustment is calculated to adjust the image brightness.After the improvement,the two images involved in the stitching can be well adjusted to the same brightness,and the stitching image transitions naturally.(3)Improved fusion method in SPHP.SPHP often produces "ghosting" when fused,resulting in impaired image clarity.There are two reasons for its appearance.One is that the pair of feature points cannot be perfectly overlapped after the image transformation,the other is that there are differences in the content of the overlapping area.For the "ghost" phenomenon,the Poisson fusion method is introduced.Firstly,the mask is used to extract the overlapping area without "ghost",and the SPHP original result combined with the area is used to construct the coefficient matrix and constraint equation,thus the improved result is obtained.After the improvement,the image distortion in the “ghost”area is restored and the image sharpness is significantly improved.This paper designs simulation experiments to verify.Theexperimental results show that the improved SPHP image stitching algorithm proposed in this paper has significantly improved the stitching quality compared with the original algorithm.