ASIFT Python implementation.

This repo is still in active development.

• opencv_contrib_python
• numpy

• CLI: python asift.py --img1 [PATH_TO_FIRST_IMAGE] --img2 [PATH_TO_SECOND_IMAGE] --detector [DETECTOR_NAME]
• Invoke the asift_main() method from asift.py

Example - Using SIFT detector with FLANN algorithm to match two sample images:

python asift.py --img1 sample/adam1.png --img2 sample/adam2.png --detector 'sift-flann'

Please refer to inline docs for more info.

• macOS 11.4 Big Sur
• macOS 12.2 Monterey
• Windows 11 Pro version 22H2

Tested with Python 3.7, 3.9 and 3.10.

While the program should run on most modern platforms, considering the time complexity of ASIFT algorithm, we recommend using a 6 core or better CPU for better image matching speed.

Requirements regarding GPU will be added when we complete development of relating modules.

The team is currently working out ways to accelerate the program effectively. We may release a CUDA enabled version in the future.

Please notice that the time bottleneck in image matching is descriptor matching (which GPU acceleration may yield limited performance improvement) , rather than feature extraction. Hence, GPU acceleration won't be the team's primary focus.

Original ASIFT algorithm was put forward by JM Morel, please refer to:

Morel, Jean-Michel, and Guoshen Yu. "ASIFT: A new framework for fully affine invariant image comparison." SIAM journal on imaging sciences 2.2 (2009): 438-469.

The project also use code from the Python OpenCV sample of ASIFT, please refer to:

https://github.com/opencv/opencv/blob/master/samples/python/asift.py

Should you use code from this repo for research purposes, please use the following citation:

@article{zhou2022mars,
  title={Mars Rover Localization Based on A2G Obstacle Distribution Pattern Matching},
  author={Zhou, Lang and Zhang, Zhitai and Wang, Hongliang},
  journal={arXiv preprint arXiv:2210.03398},
  year={2022}
}