panorama

features

• able to handle 360 panorama.
• Random sequence of images input is welcomed.
• use color blending and smoothing to make the image more continuous.

how to run

Prerequisite: matlab 2014b or higher
images sets are already in ./imgs

• If you want to see results directly, go to ./results folder
• If you want to test all images sets with only one click,run RunAllDatasets.m.(10 image sets, about 1 minute)
• If you want to specify the image folder, run main.m with path to images folder as argument as follow:
  main('./imgs/redrock');
  Note that, this currently support image sets in imgs folder. If you use your own image set, you need to set focus length and other parameters in main.m.

######details of my algorithms are shown below:

360 panorama

• [x] mapping image to cylindrical coordinate
  warp.m

recognize panorama(random inputs)

I select two random sequence images set:family_house, and west_campus1
They are already shuffled. You can see them in imgs folder.
Or you can run shuffle.bash to shuffle them again.
As described in Brown's paper, I use $N_inlier>k*N_pairs+b$ to compute whether a pair of images match or not
k,b are const. Set to 5.9 and 0.22 respectively.
See recognizing panorama for details

imorder.m

merging and blending

• [x] Alpha
• [ ] Pyramid
• [x] Noblend

merge.m

transformation

• [x] homography transformation.
• [x] translation transformation.( This is more robust)

computeTrans.m

matching

• [x] RANSAC
• [ ] exposure matching

RANSAC.m

global adjustment

• [x] end to end adjustment(comput shift and subtract shift/n to each image)
• [ ] bundle adjustment(difficult way)

create.m

getting features

• [x] use SIFT features(using VLFeat library, professor allowed)
• [x] SURF features, (SIFT is better)

getSIFTFeatures.m, getMatches.m

resize

• [x] I resize image larger than 400 pixel in width

main.m

References

A nice tutorial on panorama I find useful.