Circular Hough

Transform (CHT) was largely used in eyes detection, the circle is actually simpler to represent in

parameter space, compared to the line, since the parameters of the circle can be directly transfer to the

parameter space.

The CHT can be formulated as a convolution whose binary

mask coefficients are set on the circle boundary and are zero elsewhere.

For the CHT calculation, a separate circle filter

can be used for each radius of circle to be detected. This forms the familiar

3-dimensional parameter space, usually associated with the CHT, where two

dimensions represent the position

of the circle centre (a, b), and the

third its radius r see (1).

III.3. Circle

Hough Transform Modifications

Modifications to the CHT was implemented in order to maximize algorithm convergence,

reduce complexity and even increase speed which is essential in real time

applications by.a single accumulator space for multiple radii, the use of edge

orientation, and implementation as convolution operators. Two patches resulting

from pre-processing step represent regions of interest we apply modified CHT on

each one to search circles representing irises (Left, Right). Centre of circle

obtained is taken as being centre eye. If it fails to find the circle the mean

of the patch become Centre eye.

EVALUATION

Bio_ID database was chosen to test

our algorithm

Presence of Eyelids and eyelashes

make edge discontinuous and diverge CHT detector reason why we decide to

overcome this limitation by adding second detector which work parallelly to CHT

detector.

MSER detection

Maximally Stable Extremal Region

(MSER) seems to be evident for eye localisation because of its nature, invariant

to continuous geometric transformations and affine intensity changes and are

detected at several scales. MSER are further considered as the fastest interest

point detection method This algorithm was proposed by Matas et al. 21. It

starts from intensity function of an image and end by Maximally Stable Extremal

Regions, where each region is defined by an extremal property of the intensity

function within the region to the values on its outer boundary. Elliptical

frames are attached to the MSERs by fitting ellipses to the regions. Those

regions descriptors are kept as features.

We apply MSER algorithm to regions

R1 and R2 previously found, in each region MSER gave us one unique region

representing eye region by ellipse and centre by ellipse centre following steps

shown in (Figure3).

The first step consiste of

performing a luminance thresholding on each Eye Region starting from black to

white, Regions are extracted from connected components called “Extremal

Regions”, then compute local minimum of

the relative growth of its square that represent “Maximally Stable” regions

each Maximal Stable Extremal Region are approximated by ellipses. Eye Region is

resuting from tests on these Ellipse descriptors. We introduce algorithm to

choose best candidate represent true eye centre feature. Best Feauture is

choosen by (1) where n=number of ellipses found, Maxvariation is defined by

tests