Image Stitching Based on Feature Optimization of Bayesian Probability Model
Xiaoyuan Luo, Chongchao Wang, Shaobao Li · 2022 34th Chinese Control and Decision Conference (CCDC) · 2022
The goal of image stitching is to create a natural mosaic without ghosts, local distortions and artifacts. Towards this end, this paper presents a local feature optimization method based on Bayesian probability model for the image stitching. The spatially-varying algorithm is employed to perfectly align the entire overlapping area of images. A local feature optimization Bayesian probability model is then proposed to remove incorrect matching points in the images aiming to obtain a more accurate local homography estimation for robust alignment. Moreover, the strategy of combining local warping and similarity transformation is applied to reduce the distortion of non-overlapping regions. The salient features of the proposed method are that it is highly compatible with other image transformations and can effectively solve the inevitable ghosting problem of image stitching. Finally, experiments are conducted to demonstrate the effectiveness of the proposed algorithm.