Image Stitching via Convolutional Neural Network
Xinguo He, Lin He, Xinyi Li · 2021 7th International Conference on Computer and Communications (ICCC) · 2021
Image stitching aims at generating a panorama from a sequence of images with overlapping areas. Many approaches have been dedicated to deal with the task. However, they usually require handcrafted features and multiple fulfilling steps, which may incur high time cost and intermediate interference. Recently, deep networks have been applied in this area, but mainly focused on specific steps of image stitching, such as feature extraction and image alignment, not enabling an end-to-end stitching. In this paper, we propose a novel convolutional neural network (CNN) based method which can perform image stitching in an end-to-end fashion. Qualitative and quantitative results from experiments on synthetic image datasets verify the excellent performance of the proposed method.