superpixel_image_stitching

In this paper, the image is spliced based on super pixel.

Introduction

Super - pixel - based fast image stitching code implementation. We take the superpixel as the alignment unit and separately calculate the mapping transformation matrix of each superpixel. With the same alignment effect as APAP, we can greatly reduce the time of image stitching.

Requirements

Our program runs on windows8.1. Compile and run with Visual Studio 2015. You need four external dependencies to help you compile and run the code. The four external dependency libraries are:opencv3.3.0、Eigen3、 TinyVLib and vlfeat-0.9.20-bin.

Compile

You can refer to the "readme.docx" documentation for instructions. The detailed compilation process is described in this document.

Stitching Process

We illustrate the experimental process with the scene of railway tracks. The image stitching process is divided into three steps.

step1

The image to be spliced is divided into super pixels. The following figure is the result of super-pixel division of the image.

step2

We calculate the mapping matrix for each superpixel.

step3

We map and transform each super pixel to get a stitching image. The picture below shows the result of stitching.

Here we take a look at the difference between our experiment and the AutoStitching. Here's the AutoStitching result.

The results of the two images were compared. We can see that our method has the same effect as APAP. Our method can eliminate the problem of ghosting and dislocation in the process of image stitching.

Result

Here we tested the stitching time of ten groups of pictures. Meanwhile, the splicing time of APAP was compared. Here are the results.

Conclusion

First of all, our algorithm has the same alignment ability as APAP algorithm. Secondly, our algorithm greatly improves the speed of image stitching. Compared with the running speed of APAP algorithm, the running time of our program is almost improved by an order of magnitude.