This is the project design of course Digital Image Processing (2017-2018, Fall) in Tsinghua University presented by Dr.Shengjin Wang.

1. OpenCV 3.4.1 for python
2. OpenCV-Contrib 3.4.1
3. numpy 1.8.0rc1 or higher

SIFT(Scale-invariant Feature Transform) is a computer vision algorithm presented by Dr.David Lowe in 1999. It can extract feature points of a image, and has been proved very efficient in Image Mosaic.

RANSAC can be used to fit linear relation. Different from The Least Square Method, RANSAC is especially efficient for strong noisy data.

See Report.pdf for more detail.

Put all the images you want to mosaic in ./Pics/. The image files should be named in ascending order. The algorithm will mosaic them from left to right according to the index of file names.

For example:

ROOT/
└----Pics/
     |----table1.JPG
     |----table2.JPG
     |----table3.JPG
     |----table4.JPG
     └----table5.JPG

Run:

python SIFT_Project.py

or:

python SIFT_Project.py --img_dir=YOUR_IMG_DIR --result_dir=YOUR_DIR --result_name=YOUR_NAME --format=YOUR_FORMAT

where:

--img_dir: The path of your images set. Default:'./Pics/'
--result_dir: Where the mosaic result should be saved. Default:'./Result/'
--result_name: The file name of the mosaic result. Default:'result.jpg'
--format: The file format of your images set. It can contain various format. Default:'jpg png'

The result image will be placed where you specify or in ./Result/ by default.