stukkie d'r beijx

broadbeansromanceΤεχνίτη Νοημοσύνη και Ρομποτική

18 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

67 εμφανίσεις










Kees van Dijk S0213101

Assignment.3








Tim Witteveen S0210838

Multibiometrics
.

1.
Scatterplot
.


See
Error! Reference source not
found.
.

for the scatterplot of
face
and fingerprint scores for genuine
and impostor comparisons
.









2.
EERs for each of the methods


Method

Equal Error Rate
(%)

Individual Modalities

Face Modality

12
.
41

Fingerprint Modality

22
.
36




Multi‐Modal

Face+Fingerprint (Sum rule, No normalization)

8
.
47

Face+Fingerprint (Product rule, No normalization)

16
.
91

Face+Fingerprint (Sum rule, Minmax normalization)

8
.
93

Face+Fingerprint (Product rule, Minmax normalization)

10
.
21

Face+Fingerprint (Sum rule, Z‐score normalization)

8
.
18

Face+Fingerprint (Product rule, Z‐score normalization)

22
.
75




Multi‐presentation

Face+Face (Mean)

6
.
56

Fingerprint+Fingerprint (Mean)

6
.
74

Face+Face (Maximum)

6
.
03

Fingerprint+Fingerprint (Maximum)

6
.
51


Figu
r
e

1
, Scatter plot of face and fingerprint scores. Scores are not normalized.










Kees van Dijk S0213101

Assignment.3








Tim Witteveen S0210838


3.
Roc Curves





Commentary on performances of different methods:

Individual modalities
: Here we see that the face recognition has
a lower EER and thus a better
performance compared with fingerprint. A reason for this difference could be the system used for
capture of the specific biometric. Or the method used to create the feature vector from the captured
biometric.

Multi modalities:


Here we see that the sum method scores quite good in comparison
with the product method. The
product method scores low even compared with the individual modalities of the face recognition.
Only when the Z score normalization is used, the EER of the multi

modality case scores better than
the individual modality.

A reason for this change is that Z score normalization removes the negative numbers in the
fingerprint similarity matrix. This does not make much difference when you take the sum of both
similarit
y matrixes. But when you take a product of a similarity matrix containing negative numbers
the resulting matrix will also contain negative numbers. These negative numbers will mostly be
below the threshold and thus accepted, even if it is an imposter. This

explains the poor performance
of the product method, when no normalization or the min max normalization is used.

Another reason why the sum method scores better than the product method can be derived from
the scatterplot.





Multi presentation modalities
:

Here we see that the face +face multi presentation case scores best. This result can be derived from
the scores of the individual cases where the EER of the face is much lower than the EER of the
fingerprint. When you use 2 independent samples the best res
ult will be given by the method with
the lowest EER, being the face recognition method.

The reason why you should use the max method is because you want the 2 closest matches. In the
case of face recognition this could be caused by the position of the hea
d.