Master Assignment
Generating feature vectors from fingerprint minutiae using Deep Learning
Type: Master EE
Period: T.B.D.
If you are interested please contact :
Description:
In biometrics, fingerprints are traditionally described using minutiae, the locations and directions of the endings or bifurcations of a ridge in the fingerprint pattern. Minutiae comparators generate a similarity measure for two sets of minutiae extracted from two fingers by performing 2-dimensional geometric comparisons on those two unordered sets of minutiae.
Although comparing fingerprints using minutiae leads to high accuracy, it also has a number of drawbacks. Firstly, minutiae comparison is a complicated and therefore relative slow process which makes it unsuited to quickly search for fingerprints in large databases. Secondly, the complicated nature of the minutiae comparison process makes it extremely hard the implement it efficiently in many Privacy-by-Design approaches such as homomorphic encryption.
Other modalities often represent a biometric as a high-dimensional feature vector. This feature vector representation allows for an efficient biometric comparison process by calculating a distance between two feature vectors. This simpler process also makes it feasible to implement it in several Privacy-by-Design methods.
Goal of the research:
Crafting high quality feature vectors from minutiae information requires significant effort of highly specialized scientists and engineers in an iterative and time-consuming process. The goal of this research is to develop an approach based on Deep Learning for deriving feature vector representations of minutiae sets that lead to good accuracy. Part of the work is concerned with choosing the best Deep Learning approach, minutiae input representation, training the chosen network(s) and assessing the accuracy of the generated feature vectors.
Profile of the student:
Some background in biometrics and experience in Deep Learning.