Recognise and count newly implanted hair follicles


My parter does hair transplants and I thought it would be good to verify how many follicles were actually transplanted at the end of the surgery. I’m new to computer vision so am wondering if someone could point me in the right direction of an approach that could be used to first recognise the follicles in each image, then count them.

My thought was to use edge detection to show each follicle as an approximate circle, then count the circles.

I can probably manage preparing the image, reducing blown highlights to nearby colours so they don’t get picked up in the edge detection, but I’m not really sure what to do once I have the edges… or even if edge detection is the right approach, but I’m not sure where I would go next. Or even if this is the right approach?

If I go down the ML route, will I need to manually mark some images to be able to train the model?

Sorry if you’re squeamish. Here’s a zoomed in example of the hair follicles. The blue is there because sometimes the patient’s follicles are not as easy to see, so dye is placed on the area.


Anyone have any thoughts about this? I’m not really sure how to start!

It makes no sense to develop a program for a single run…

So, you should start by making a big print of the photographs and starting to do the count manually. Mark with a pen those that have been counted, as you go along.

If no exact number is needed and all areas are similar, pick a smaller are for counting.

She does three surgeries per week with up to 5000 transplants each, it definitely needs automating.