Research On Image Stitching Algorithm

Image mosaic technology is a technique of stitching two or more narrow view images with overlapping regions into a wide view angle and high resolution image.It has been widely used in medical image processing,remote sensing image processing,virtual reality and so on.Image registration and image fusion are two key research contents in image stitching.The traditional registration method requires process of shooting a picture to meet the ideal condition that the photographing device is located in a fixed point in three-dimensional space.However,ideal condition is often hard to reach,which results in a certain parallax between images.To solve this problem,a new registration algorithm is proposed in this paper and can precisely registration large parallax images.At present,the processing algorithms for large disparity images need to adjust complex parameters.Based on the proposed algorithm,this paper makes a slight improvement and designs a splicing error to measure the quality of stitching.This parameter drives the proposed algorithm to automatically optimize and realize automatic registration of large parallax images.Image fusion is the key research problem of image mosaic.The image may have some brightness or color difference due to the factors such as shooting time,illumination,equipment and so on.The images with color difference will have obvious artificial traces after stitching.To solve this problem,an improved optimal stitching algorithm based on HSV(Hue,Saturation,Value)space is proposed in this paper.The algorithm can effectively eliminate the stitching marks caused by color difference.The proposed method is verified by experiments,and the effect is obvious compared with the traditional algorithm.The main research contents are as follows:1.Firstly,the reason why the large disparity image is difficult to be registered is analyzed,and a hypothesis is proposed that large parallax image includes a plurality of display sub-plane and the feature points are densely distributed in each plane.Then,the hypothesis is verified experimentally.The proposed algorithm is based on the distribution of feature points,and the sub-plane is segmented by clustering algorithm,and then the local registration of the sub-plane is carried out.First,the K-means algorithm is used to group the matched feature points.The overlapping regions are re-clustered taking the clustering centers of each group feature point as new clustering centers to obtain the dominant sub-plane.Then,the projection parameters of each sub-plane are solved through the feature points of them,and the projection matrix of non-overlapping region is assigned by the nearest neighbor principle.Experimental results show that this method can accurately register large disparity images.2.On the basis of the above algorithm,a mosaic error parameter is designed to measure the quality of image mosaic.By this parameter we can find the optimal number of groups of feature points,that is,the number of the image contains the sub-planes.Hierarchical clustering algorithm has the characteristics of clustering first and then grouping.It can achieve one time clustering,multiple grouping.In the process of searching the optimal number of groups,compared with the K-means algorithm which each group needs random given center and re-clustering,it can enhance the stability of the similarity relation between the feature points,and thus improve the convergence and efficiency of the algorithm.Experimental results show that this method can effectively register large disparity images and can realize automatic registration.3.By analyzing the traditional best stitching algorithm,we find that it can't accurately find the optimal stitching line because it can't accurately measure the color difference between pixels.In view of this problem,firstly the overlapping regions between images are obtained by image transformation in this paper.Because the HSV color space can more accurately describe the color,brightness and other information,then the overlap region is converted to HSV color space.Then,the best splicing path is found by the energy function designed in this paper.The energy function takes into account the horizontal color and structure difference of the single image,and the longitudinal color and structure difference between the two images.Finally,a region is delineated around the optimal stitching path,and the two images are smoothed by the simplified pixel correction method.Experimental results show that the algorithm can effectively process images with certain color differences,and achieve seamless stitching effect.