-
Write a unit test to check accuracy of each algorithm
-
Make it require minimum no of tweaks in configurable parameters
-
(Optional) Use canny edge detection to binarize image. Compare results before and after.
-
Use hough transforms to find all radiis
-
Convert to normal form
-
Filter out radiis with deviation > threshold. How to decide threshold? No tuning required there
-
Discard all circles lying outside median. Smallest circles are also the most frequent for bacterial colonies """
-
Calculate mode radii of circles. Try plotting histogram of values """
-
Write unit tests. Assert that accuracy for method is greater than 85% in each case """
-
Write this version with unit tests first """
-
Read papers on automatically counting bacterial colonies. They should give you an idea of state of the art implementation """