Large parallax image matching based on Graph Matching

In the security video surveillance system, image splicing technology faces difficulty of achieving large parallax and small overlap area. This paper summarizes traditional stitching analyzes that matching algorithm is key to solve problem stitching. Aiming at problems methods with low accuracy cannot adapt non-rigid transformatio, a model based on proposed in this for tasks. The which combines similarity global structure local can not only divide feature point structural level but also pair each other. At same time, framework be trained end-to-end. training samples composed simulation data sets are adopted train designed tested verifyed by test sample three kinds mixed data. Compared PCA-GM, adapts well achieve 93.1% 91.9% set respectively, initially scene. However, further research aiming 74.5% will needed.