This is the code for circle detection in images using inscribed triangles. Circle detection is a critical issue in pattern recognition and image analysis. Conventional methods such as Hough transform, suffer from cluttered backgrounds and concentric circles. We present a novel method for fast circle detection using inscribed triangles. The proposed algorithm is more robust against cluttered backgrounds, noise, and occlusion.

The code was implemented with VS 2019, OpenCV 3.4.7, and Eigen3.

To test images for your own data. Run the 'test.cpp' in the './src' directory.
It allows you to specify the input file path:
cv::String path = "E:/Code/patterns/Images1/";
and the output path for the detected results:
cv::String dst = "E:/Code/patterns/result/";
Here, you need to create two directories, ie, 'Images1' and 'result'. If there are corresponding ground truths (GT), then you can further add the GT path:
cv::String GT = "E:/Code/patterns/GT/";

Four real-world datasets for circle detection: Dataset Mini, Dataset GH, Dataset PCB, and Dataset MY, are provided. Dataset Mini contains 10 images which are used as a benchmark by several works. Dataset GH contains 258 gray real-world images. Dataset PCB contains 100 industrial printed circuit board images, which are also grayscale. Dataset MY contains 111 colorful real-world images. We also provide ground truths for each dataset.

Due to the complexity of real-world images, we cannot hope a set of fixed parameters to get the best results for each image. To customize your purpose, we provide some suggestions:

• The inlier ratio threshold 'T_inlier', the larger the more strict. Hence, to get more circles, you can slightly tune it down.
• The sharp angle threshold 'sharp_angle'. To detect small circles, you can slightly tune it up
• The other parameters are usually fixed.

If you find our work useful in your research, please cite our paper:
@article{zhao2021occlusion, title={An occlusion-resistant circle detector using inscribed triangles}, author={Zhao, Mingyang and Jia, Xiaohong and Yan, Dong-Ming}, journal={Pattern Recognition}, volume={109}, pages={107588}, year={2021}, publisher={Elsevier} }