dbarac / sift-cpp

A C++ SIFT implementation (Scale invariant feature transform)

sift-cpp

Introduction

This is a C++ implementation of SIFT, a feature detection algorithm.

Libraries used

stb_image and stb_image_write for loading and saving images. (included in this repo)

Usage example

Find keypoints, match features in two images and save the result:

#include <vector>
#include "image.hpp"
#include "sift.hpp"

int main()
{
    Image img("./../imgs/book_rotated.jpg");
    Image img2("./../imgs/book_in_scene.jpg");
    img = rgb_to_grayscale(img);
    img2 = rgb_to_grayscale(img2);
    std::vector<sift::Keypoint> kps1 = sift::find_keypoints_and_descriptors(img);
    std::vector<sift::Keypoint> kps2 = sift::find_keypoints_and_descriptors(img2);
    std::vector<std::pair<int, int>> matches = sift::find_keypoint_matches(kps1, kps2);
    Image book_matches = sift::draw_matches(img, img2, kps1, kps2, matches);
    book_matches.save("book_matches.jpg");
    return 0;
}

Result:

Build and run the examples

Build

$ mkdir build/ && cd build && cmake .. && make

The executables will be in sift-cpp/bin/.

Run

Find image keypoints, draw them and save the result:

$ cd bin/ && ./find_keypoints ../imgs/book_rotated.jpg

Input images can be .jpg or .png. Result image is saved as result.jpg

Find keypoints in two images and match them, draw matches and save the result:

$ cd bin/ && ./match_features ../imgs/book_rotated.jpg ../imgs/book_in_scene.jpg

Result image is saved as result.jpg

Useful links

• SIFT paper
• Anatomy of the SIFT method
• Blog post about a Python SIFT implementation

sift-cpp's Issues

I need a permission.

Hello,

Thank you for the awesome work. I have translated this SIFT implementation to dlang to include it in DCV (an open-source CV library). I am one of the maintainers. Of course, all the credit will be given to @dbarac by explicitly mentioning in the source file and a reference to this repo.

Thanks in advance
Ferhat

https://github.com/libmir/dcv

Comparison report

Thanks for this great project. Did you compare this implementation of SIFT with the OpenCV version? In terms of output, computation time and other performance parameters.