Research On Multi-view Image Stitching Based On Multi-plane Registration And Seam-line Fusion

Ordinary cameras have small field of view and information.By shooted the same scene from different perspectives with sensors,multi-view images with overlapping areas can be obtained.A series of narrow viewing angle images can be combined into a wide viewing angle image with high visual quality using image stitching technology.Aiming at the problems of multi-view image stitching algorithm,such as high fault detection rate of feature point,low registration accuracy,many splicing traces and serious fuzzy,we proposed a multi-view image stitching method based on multi-plane registration and seam-line fusion.Key work and innovations include:(1)Aiming at the problem that the existing multi-view image stitching algorithms have high fault detection rate of feature point,we proposed a feature points screening algorithm based on the clusters with their plane information.Firstly,the feature points in images are extracted and coarsely matched by the K-D tree algorithm based on the distance ratio.Secondly,the plane similarity between the feature points is computed according to the characteristic that the same plane should be corresponding to the same homography matrix.Finally,matched feature point pairs are classified into wrong and correct matched classes by agglomerative hierarchical clustering to screen feature points.(2)Aiming at the problem that the existing multi-view image stitching algorithms have low registration accuracy,we proposed a stitching method based on multi-plane registration.Firstly,the image is divided into grids with the same size,and the plane corresponding to each grid is computed according to the result of hierarchical clustering for the feature points.Secondly,the weight of the local homography matrix is defined by combining the plane and distance information of the grid and the feature points.Finally,the bundled adjustment algorithm were used to optimize the transformation parameters,and the pixels of all images are transformed into the reference coordinate system to achieve image registration.(3)Aiming at the problem that the existing multi-view image stitching algorithm has obvious stitching traces and the fusion results of the image with different focus are blurred,we proposed an optimal seam-line search algorithm based on focus score.Firstly,a salient sparse representation model was proposed to decompose the image.Secondly,the model was used to extract the focus of the overlapping region of the images,and constructed a focus score map.Finally,the energy function is established according to the focus score map,the luminance difference map and the edge difference map.We searched for the seam-line using graph-cut algorithm to achieve image fusion.Compared with the widely used image stitching method,the proposed method effectively reduce the fault detection rate of feature points,improved the registration accuracy,improves the splicing traces of overlapping areas and enhances the image clarity.