Method of searching for parameterized contours for comparing irises
A method for detecting outlines for iris comparison comprises a step of selecting N candidate outlines of circular form by applying a circle search technique to an image of edges of an iris. It also comprises a step of optimizing the form and the position of the N candidate outlines, the optimized candidate outlines being determined by using parametric models, a set of parameters being determined for each candidate outline by minimizing a quantity of energy E(C). The method also comprises a step of selecting the best optimized candidate outline.
1. A method for detecting outlines for iris comparison, comprising:
selecting N candidate outlines of circular form by applying a circle search technique to an image of edges of an iris,
optimizing the form and the position of the N candidate outlines, the optimized candidate outlines being determined by using parametric models, a set of parameters being determined for each candidate outline by minimizing a quantity of energy E(C),
selecting the best optimized candidate outline
wherein the quantity of energy E(C) associated with a candidate outline C is equal to the sum of the edge energy E edge (C) and of the region energy E region (C) of said outline.
2. The method for detecting outlines as claimed in claim 1 , in which the circle search technique used to select N candidate outlines of circular form is a circular Hough transform.
3. The method for detecting outlines as claimed in claim 1 , in which the circle search technique used to select N candidate outlines of circular form is the Random Sample Consensus method.
4. The method for detecting outlines as claimed in claim 1 , in which the quantity of energy E(C) is minimized by using the gradient descent method.
5. The method for detecting outlines as claimed in claim 1 , in which the energy of edges is determined by using the following expression:
E
edge
(
C
)
=
∫
0
2
π
▽
→
[
I
(
X
→
p
(
θ
)
)
]
·
n
→
θ
ⅆ
θ
in which:
{right arrow over (n)} θ represents the normal unitary vector outgoing at the angular parameter point θ;
{right arrow over (∇)}[ ] represents the gradient function;
{right arrow over (X)} p (θ) represents a column vector whose elements are the coordinates x(θ), y(θ) of a point of the image;
I({right arrow over (X)} p (θ)) represents the intensity of the image at the point associated with the vector {right arrow over (X)} p (θ).
6. The method for detecting outlines as claimed in claim 1 , in which the region energy E region (C) is determined by using the following expression:
E
region
(
C
)
=
-
∫
Rin
log
P
in
(
I
(
X
)
)
ⅆ
X
-
∫
Rout
log
P
out
(
I
(
X
)
)
ⅆ
X
in which:
R in represents the region of the digital image of the eye inside the outline;
R out represents the region of the digital image of the eye outside the outline;
P in ( ) and P out ( ) correspond to the probabilities for a pixel of intensity I(X) of the image to be located inside or outside the outline.
7. The method for detecting outlines as claimed in claim 5 , in which the parametric model is a circle model, the set of parameters associated with this parametric model then comprising 3 parameters.
8. The method for detecting outlines as claimed in claim 5 , in which the parametric model is an ellipse model, the set of parameters associated with this parametric model then comprising 5 parameters.
9. The method for detecting outlines as claimed in claim 1 , in which the best outline out of the N optimized candidate outlines is the one for which the energy E(C) has the lowest value.
10. A method for detecting inside and outside outlines of the iris of an eye, wherein the method as claimed in claim 1 is applied a first time to detect the inside outline and a second time to detect the outside outline.