Image Alignment Via Multi-model Geometric Fitting And Hierarchical Homography Estimation

With the rapid development of computer vision and artificial intelligence,image stitching technology has been widely used in many fields,such as target recognition,medical imaging,geological exploration,robot vision and so on.In recent years,with the deepening of scientific research,many image stitching algorithms have emerged,but most of the current image stitching algorithms are based on global single transformation or multi-level local transformation.If the target is a multi-plane image,the transformation will be inaccurate because of the mismatching point or the low precision of extracting feature points,resulting in image distortion or ghosting in overlapping areas.Therefore,we develop a multi planar image stitching technology.The main work and innovation of this paper are as follows:1.Multi-transform fitting mechanism for feature point classification and an image layering method are proposed to transform the multi-plane problem into a multi-layer problem.Firstly,the feature points in the image are extracted based on scale invariant feature transformation.Because the algorithm has some mismatches,the random sampling consistent algorithm is used to eliminate the mismatched feature points and improve the accuracy of feature point extraction.Classification of image feature points is completed.Finally,according to the classification results of feature points,the image is divided into basic layer and transition layer by using the boundary information between classes.2.Order to solve the stitching problem of image mosaic,a hierarchical transformation scheme is proposed,and different transformation algorithms are designed for the basic layer and the transition layer.The dynamic direct linear transformation method is used to transform the base layer.A multi-homography estimation algorithm based on energy optimization is proposed to improve the accuracy of single strain transformation.The proposed algorithm completes the image level division before the image transformation step,and can effectively solve the multi-plane problem,and the subsequent image hierarchical transformation method can better solve the stitching problem of image mosaic.Experimental results show that the proposed algorithm can seamlessly stitch multi-plane images,and the image results have a good visual effect.The experimental results are superior to many existing algorithms.