Detect veins in a leaf (deep learning, image processing)

Short description
-------------------------
I need a service which would take a photo of a cigar as input and return an image showing the main veins on the cigar leaf. In the sample images it is shown: the original photo of a cigar, an image with veins located and the last image with just veins.

The service should be developed as a microservice which I can deploy to any major platform (such as AWS). It is prefferably written in Java. It should be able to return the result in a matter of seconds. There should be no manual work as part of the service.

Some details
------------------------
Cigars are wrapped in a tobacco leaf which has veins visible on it. Veins are of different thickness. For this service its important that the service locates the main vein and the ones connected to it. Plus there is always the edge of the leaf itself which should be detected.

The most important part is that the system always detect the same veins on the same cigar even when the photo is taken in different light. But it is assumed that the photos are of good quality.

I have cca 100 close up photos of cigars for learning purposes. I could provide more if needed and if it helps getting the result.

Important notes
------------------------

The result should be a clean black and white picture with only veins showing as clean lines as you can see in the sample images here.

This is not just about edge detection - it is about locating veins and turning them into clean lines. The resulting image should only show veins and no other pixels should be on it. The veins have to be lines made of black pixels.

Only the main thick veins (main veins and connecting ones) shuold be located, the small ones that are hardly visible are not needed for this stage.