Retina pattern recognition towards human identification

A paper prepared from my undergraduate thesis work has been indexed in IEEE Explore recently. Find the abstract at IEEE explore. It was basically on human identification based on retina blood vessel orientation pattern recognition. We proposed a way to detect the pattern of the blood vessels followed by identifying the intersection points of those vessels. Blood vessel orientation of different human beings are different and hence the intersecting points of these vessels give a unique pattern to recognize. However, to compare between two patterns, the variance of the intersecting points are measured. If the measured variance is under a threshold value which is empirically determined, we assume it as a match. Steps are pictorially described below:

1. Human Retina:

Green-Scale version of a colorful retina image

2. Segmentized blood vessels:

Binary image. Detected blood vessel portions are black

3. Intersecting points detection (Binary Image):

Binary image. Intersection points are determined

Here are some snapshots of the result found from our experiment:

		Sample 1 of
		Person 1	Person 2
Sample 2 of	Person 1	81.72%	12.90%
	Person 2	08.70%	84.78%

Its mentionable that before starting all the above steps we needed to compensate the rotational and translational effect between two sample of retina image. We inspired for the work from this paper published in ICECE 2004 hosted by BUET. In this work I paired with my thesis partner Amran Siddiqui (Lecturer, Stamford University, Bangladesh)

Thank you.