Large-parallax Image Stitching Based on MAGSAC++ and Image Content
Aolong Qin, Lianhe Shao, Xihan Wang, Xudong Yang, Quanli Gao, Nan Shi, Qianqian Yan, Tongtong Luo · 2023
The purpose of image stitching is to generate high-resolution panoramic images that are seamless, distortion-free, and free of artifacts while minimizing computational costs. In this paper, we propose a stitching algorithm for large-parallax image based on MAGSAC++ and image content. Specifically, this paper decomposes the problem into three steps: image registration, seam cutting, and image fusion. Firstly, we employ an image registration algorithm based on MAGSAC++ to align the input images and obtain two aligned images in the same coordinate system. Next, we propose a pixel-based energy function that integrates color difference, gradient difference, and texture complexity information. We also introduce the LBP operator to better represent the underlying texture information of the images. Based on this energy function, we search for the perceptually optimal seams located in contiguous regions with high similarity. Finally, we use the Laplace pyramid fusion algorithm to eliminate visible seams and achieve natural color transitions. Experimental results demonstrate that the proposed method achieves seamless stitching of large-parallax images effectively and efficiently, outperforming other classical stitching algorithms.