Image splicing method based on grid deformation

Abstract

The invention discloses an image splicing method based on grid deformation, relating to the technical field of image splicing. The method includes the following steps of S1 respectively extracting SIFT characteristics of two space adjacent images, and matching the characteristics; S2 performing iterative computation by means of an RANSAC algorithm and screening characteristic point pairs; S3 estimating the global projective matrix and performing global projective transformation; S4 by means of Delaunay triangulation, constructing a dense triangular grid for an overlapping region and constructing a sparse grid for a non-overlapping region; S5 establishing the grid deformation energy function and optimizing the global projective registration; and S6 fusing the images to produce a wide-view-field image. The dense triangular grid represents the overlapping region and the sparse triangular grid represents the non-overlapping region of the image, so that compared with the APAP method, the whole image uses dense and same matrix grids, the algorithm efficiency is greatly improved. Meanwhile, the grid deformation can make full use of the parallel and high-speed operation performance of the CPU, and the splicing speed can be greatly improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, relates to the technical field of image splicing, and in particular relates to a real-time splicing method for images based on grid optimization. Background technique [0002] Image stitching is a technology that uses multiple cameras to acquire images from different angles and stitches them to form a wide-view image. Image stitching is mainly divided into three main steps: image preprocessing, graphic registration, and image fusion. [0003] In order to reduce splicing ghosting and blurring caused by parallax in splicing, the existing related algorithms can be summarized as the Spatially-Varying Warping method, that is, the local parameter transformation model is used in the registration stage instead of the global transformation model to apply to different depths. One scene. [0004] Zaragoza et al. proposed the As-Projective-As-Possible (APAP) method, which divides the image into mult...

AI Technical Summary

Problems solved by technology

The APAP method introduces the concept of grid, and the registration accuracy is greatly improved, but the shape is not maintained well, and the calculation complexity is relatively high, and the real-time performance is not good.

[0005] From the perspective of shape correction, the Shape-Preserving-Half-Projective (SPHP) method divides the image into three major areas, from overlapping areas to non-overlapping areas, and gradually transitions from projective transformation to similar transformation, which reduces shape distortion, but The registration accuracy is not high, and there are obvious ghosting and blurring in the overlapping area

Images