This question, is not a real question, but it is a topic for those who want to perform some researches in future. I will try to briefly share my experiences (although it is not vast) here:

I did some researches on retinal vessels segmentation 4 years ago for my master thesis. The objective was "vessel extraction" using light weight processing (without applying machine learning methods) using matched filter. The starting point was a paper written by S. Chaudhuri in 1889 [1]:

well, let's start from the basics: what a matched filter does, is searching inside the data to find a pattern. So if you give it the pattern of a vessel (bell ring shape of the cross section of a vessel), and search the retinal image, pixels containing the pattern can be extracted.

The solution is not easy as appears in the beginning, since selecting a pattern is not such easy! Hence vessels have different diameter, so if you select a wide pattern, narrow vessels are ignored and vice versa!

To solve the trade-off, there are some solutions; selecting different patterns (a wide and then a narrow pattern as done in [2]) or selecting a median pattern (the work done in [3]).

OK, but how to make a pattern? That's easy. If you take a look to the cross section of a vessel in a grey scale retinal image, you see that the brightness starts to decrease until the middle of the vessel, and then increases again until the border of that vessel! That means you can estimate the shape with a ring-bell shape function! Thanks K.F. Gaus to introduce Gaussian function, so that you can use it to create bell ring shape patterns. However some other functions also can be used instead. As an examples, in [3], it was shown that Cauchy function presents better patterns for detecting vessels.

Note that since vessels are found in different directions on the retina, you should rotate the template while you are looking for a vessel in a pixel! It means that for every pixel that you are searching, you should rotate pattern to make sure you are not going to lose any vessels.

You may need to use some other image processing techniques to enhance the quality of image after performing the matched filter.

That's it, if you have any question, comment here.

[1] https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CCIQFjAA&url=http%3A%2F%2Fwww.researchgate.net%2Fpublication%2F3220187_Detection_of_blood_vessels_in_retinal_images_using_two-dimensionalmatched_filters%2Flinks%2F00b7d52146f1c7a9a5000000&ei=PvV2VI35I4_7au7EgvgN&usg=AFQjCNEuO47FOKKAeZT27P-CtsWJmRQ-Uw&sig2=icgpaa3uFj4d_qVigPgbzg&bvm=bv.80642063,d.bGQ&cad=rja

[2] http://www.siue.edu/~sumbaug/RetinalProjectPapers/An%20improved%20matched%20filter%20for%20blood%20vessel%20detection%20of%20digital%20retinal%20images.pdf

[3] http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3967450/