Parallax-tolerant Image Stitching
Fan Zhang and Feng Liu
Department of Computer Science, Portland State University
Input Images
Our result
Parallax handling is a challenging task for image stitching. This paper presents a local stitching method to handle parallax based on the observation that input images do not need to be perfectly aligned over the whole overlapping region for stitching. Instead, they only need to be aligned in away that there exists a local region where they can be seam-lessly blended together. We adopt a hybrid alignment model that combines homography and content-preserving warping to provide flexibility for handling parallax and avoiding objectionable local distortion. We then develop an efficient randomized algorithm to search for a homography, which,combined with content-preserving warping, allows for optimal stitching. We predict how well a homography enables plausible stitching by finding a plausible seam and using the seam cost as the quality metric. We develop a seam finding method that estimates a plausible seam from only roughly aligned images by considering both geometric alignment and image content. We then pre-align input images using the optimal homography and further use content-preserving warping to locally refine the alignment. We finally compose aligned images together using a standard seam-cutting algorithm and a multi-band blending algorithm. Our experiments show that our method can effectively stitch imageswith large parallax that are difficult for existing methods.
Fan Zhang and  Feng Liu. Parallax-tolerant Image Stitching. 
IEEE CVPR 2014, Columbus, OH, June 2014. (oral, acceptance rate 5.75%)
Supplementary material
Related project
Fan Zhang and  Feng Liu. Casual Stereoscopic Panorama Stitching. 
IEEE CVPR 2015, Boston, MA, June 2016.
Project website

Feng Liu
, Yu-hen Hu and Michael Gleicher. Discovering Panoramas in Web Videos
ACM Multimedia 2008, Vancouver, Canada, October 2008. pp. 329-338.
Project website
We experimented with our parallax-tolerant image stitching technique on images from the APAP (as-projective-as-possible stitching) dataset and our own stitching dataset. You can download the stitching dataset here.

Note, we used the code shared by the APAP authors with the default setting to produce the APAP results. For your convenience, in this website we share the APAP code that we downloaded when preparing our publication. You can also download the code directly from the APAP website, which may contain new versions with improved performance.
Detailed Figures
Example 1
Example 2
Example 3
Failure Examples
Example 1
Example 2
Example 3