- #1
roldy
- 237
- 2
I'm working on a fingerprint recognition project and have run into a road block. As of now, I am using a 3 layer feed-forward neural network to find the minutiae in a fingerprint. If a minutiae is detected, it's location in the image (row, col) and the type (n types) of bifurcation is noted; these parameters will be known as properties. I do this for my target (database) set and for my input fingerprint. What I want to do now is to maybe use another neural network to compare the properties of the input fingerprint to the properties of the fingerprints in the database.
The problem here is that I basically have 3 property values. row can range from 1 to max row in target and col can range from 1 to max column in target. The type n, for the sake of explanation, is say 6. Is it possible to train a neural network to recognize any integer value for these three properties?
Another possibility that I was thinking of implementing was the use of an error function. I could come up with some sort of error function that's dependent on the input fingerprint and the target fingerprint properties. Basically the error function would return a 1 or 0 depending on if the value falls or exceeds a certain threshold. The problem with this is that I would have different threshold values.
Any thoughts or suggestions on the feasibility or possibility of such a neural network/error function? Your help is greatly appreciated.
The problem here is that I basically have 3 property values. row can range from 1 to max row in target and col can range from 1 to max column in target. The type n, for the sake of explanation, is say 6. Is it possible to train a neural network to recognize any integer value for these three properties?
Another possibility that I was thinking of implementing was the use of an error function. I could come up with some sort of error function that's dependent on the input fingerprint and the target fingerprint properties. Basically the error function would return a 1 or 0 depending on if the value falls or exceeds a certain threshold. The problem with this is that I would have different threshold values.
Any thoughts or suggestions on the feasibility or possibility of such a neural network/error function? Your help is greatly appreciated.