Feature Points Selection for Rectangle Panorama Stitching

EasyChair Preprint 665

Abstract

Panoramic images can provide users with large view scene, which is widely used in various fields. Image stitching can combine images of adjacent views with small horizon field into a single image with large horizon. Currently stitching method can provide a rectangle panorama by cropped method to view or print. However, this method can occur shape distortion, and information loss. In this paper, we propose a novel shape-optimization method based on feature selection for rectangle panorama. First, to avoid local distortion in overlapping region, the matched feature points is selected by feature cluster analysis. And then, we combine MDLT and derivative warping method to reduce global shape distortion. At last, a quadratic mesh grid optimization model, which is based on boundary constraint and mesh lines constraint, is established to generate rectangle panorama image. Experimental results show that the proposed method can effectively stitch different view image to generate rectangle panorama without ghost, shape distortion and information loss.

Keyphrases: Rectangle Panorama, feature selection, image stitching, image warping

@booklet{EasyChair:665,
  author    = {Weiqing Yan and Shuigen Wang and Guanghui Yue and Jindong Xu and Xiangrong Tong and Laihua Wang},
  title     = {Feature Points Selection for Rectangle Panorama Stitching},
  doi       = {10.29007/ctsv},
  howpublished = {EasyChair Preprint 665},
  year      = {EasyChair, 2018}}