Dear Devs,

In the process of tracking objects under the microscope I was confronted with an image analysis problem, for which I could not find a satisfactory solution in איק standard scientific packages (nor in the literature survey I made).

At the time I started this, scikit-image did not include a circle Hough transform and OpenCV's was not good enough.

I ended up developing an algorithm myself, which can be regarded as an off-spring of the circle Hough transform.

I summarised the work in a manuscript titled:

"Robust and highly performant ring detection algorithm for 3d particle tracking using 2d microscope imaging" (a pre-print is available at arXiv:1310.1371).

NB, it was written minded at potential users and applications rather than computer scientists, which I am not.

I would like to make the code publicly available.

I believe it may be beneficial to the scikit-image community and users as a practical alternative to the current Standard Circle Hough transform.

I contacted Stéfan van der Walt, who expressed interest in the algorithm.

I would be grateful to have your feedback, any input would be helpful!

With kind thanks to all contributors,

Eldad

Eldad Afik

Physics of Complex Systems

Weizmann Institute of Science

Some technical notes:

A typical example of the images I need to analyse can be found in the manuscript arXiv:1310.1371 (Fig. 1a as well as S2a).

The results of the software I wrote exhibit a detection rate which exceeds 94% and have only 1% false-detection, providing sub-pixel accuracy.

The conceptual steps in making the algorithm robust are sketched in Fig. 2.

The main steps to make it efficient, both in execution-time as well as memory consumption, are outlines on page 8; more technical details in this respect can be found in the "Supporting Information" Text section.

The "methods" section of the manuscript provides some details regarding the benchmarking.