You are on page 1of 7

Automated Detection of the Magnitude of Gaze

Correction for Strabismus Patients


$

P.Manju Bharghavi, ^Basireddy Krishnaja, #Ennawar Prajwala, ~Sangem Manisha


$^#~

Department of Biomedical Engineering, GRIET, Hyderabad


$
manjubhargavi@griet.ac.in
basireddykrishnaja@gmail.com
#
ennawar.prajwala@gmail.com
~
manisha.sangem@gmail.com

Abstract-- Strabismus is an eye defect which, if left


untreated, can result in inability to attain binocular vision or
blindness in the affected eye. Early detection is critical for
successful treatment. Current methods for detection involve
going to an eye specialist or going to a lab with specialized
equipment and lighting setup. These options can be
expensive and thus aren't always available to all people. The
purpose of this project was to determine the feasibility of a
software-only solution that could be deployed to photo
studios as a low-cost early screening mechanism. The
eventual goal is to be able to detect signs of strabismus from
a single high resolution portrait photograph of an infant.
Several existing computer vision algorithms for detecting
various eye features were investigated and the ones that
would handle the requirements best for implementation
were chosen. So far a good degree of detection accuracy
(roughly 80%) has been achieved and currently the eye
detection and corner detection routines are being fine-tuned
to improve detection accuracy even more. MATLAB
software is used as a tool to achieve the goal of this project.
MATLAB software is one of the most powerful software
using worldwide and it also consist a lot of image processing
library which can be use for object detection. For this
project, the program code is written using this software and
it shall be efficient to detect the objects, eliminate the
background image and detects the object.
Keywords
squint eye, gaze correction, strabismus, MATLAB,
Hough transform, eye recognition.

1.

INTRODUCTION

The medical name for Squint (or crossed-eye or lazy eye) is


Strabismus which means misalignment of eyes. The incidence of
squint is approximately 1.5% to 2% of population worldwide.
Strabismus is an eye defect characterized by a lack of
coordination between the eyes, causing the eyes to look in
twodifferent directions and thus be unable to fixate on a single
image. If children with strabismus are left untreated, the brain
may learn to ignore the input from the weaker eye.
Consequently, the children may suffer permanent loss of vision
in the ignored eye. New born babies often appear cross eyed but
this does not mean that they have strabismus.

The Strabismus detection algorithm consists of four phases. In


the first phase, the locations of cornea (i.e. centres and diameter)
of both eyes are detected for normal eyes. In the second phase,
we are finding the centre and diameter for normal eyes looking
in different directions (i.e. right, left, up, down directions). In the
third phase, we are finding the centre and diameter for normal
eyes moving horizontally with slight alignment. In the fourth
phase, we are finding the centre and diameter for strabismus
patients
Vladimir Vezhnevets & Anna Degtiareva[1] proposed a novel
robust technique, which combines stability and accuracy.
Suppose that rough eye regions are known. First, iris center and
radius is detected by looking for a circle separating dark irisand
bright sclera (the eye white). The iris center and radius detection
is performed in images red- channel, which emphasizes the iris
border. This is due to the fact, that iris usually exhibits low
values of red (both for dark and light eyes), while the
surrounding pixels (sclera and skin) have significantly higher
red values. Then, upper eyelid points are found using on the
observation that eye border pixels are significantly darker than
surrounding skin and sclera. The detected eye boundary points
are filtered to remove outliers and a polynomial curve is fitted to
the remaining boundary points. In order to reduce the effect of
image noise, the image is preprocessed with median filtering and
1D horizontal low-pass filtering. Finally, lower lid is estimated
from the known iris and eye corners.

WORK

The proposed method consists of a two dimensional Hough


transformation for detecting circle of unknown radius. The
circular Hough transform rst generates two dimensional
parameter space (xc,yc) using the gradient of grayscale. The
gradient is calculated from the facial image and draws a line
segment to gradient direction from each edge point in the
images. The radius of circle r is determined for each local
maximum in the (xc; yc) space. The advantage of this approach is
that the circle with unknown radius can be detected more
efciently than ordinal three dimensional Hough transform

because it works in two dimensional parameter space. The eye


detection step of proposed method rstly detects possible eye
center by the circular Hough transform. Then it extracts
histogram of gradient from rectangular window centered at each
eye center. Likelihood of eye of the extracted feature vector is
evaluated and pairs of eyes satisfying predened conditions are
generated and ordered by sum of the likelihood of both eyes
Hough Transform
The Hough Transform is a method for finding shapes in an
image. The Circle Hough has been implemented here, which is
used to find circles within an image.
Theory:
The HT can be described as a transformation of a point in a 2
dimensional region to a parameter space, dependent on the shape
of the objects to be identified. The basic functionality of the HT
is to detect straight lines. A straight line in the x,y-plane is
described by:
y = m*x + b
(1)
This line is represented in the Cartesian coordinate system by its
parameters b and m where m is the slope and b is the intercept.
Due to the fact that perpendicular lines to the x-axis can give
unbounded values for parameters m and b (b and m rises to
infinity), lines are parameterized in terms of theta and r such
that:
r= x*cos () + y*sin (), for [0, p]

Flowchart

Start

(2)
Load the cropped

image

Display the image

Figure 1: Hough transform for r and


Where r is the distance between the line and the origin, is the
angle. Thus, given x and y, every line passing through point (x,
y) can uniquely be represented by (, r). Both and r have finite
sizes. The distance r will have the maximum value of two times
the diagonal of the image.
An edge detector is commonly used to provide a set of points
that represents the boundaries in the image space. Equation (1)
corresponds to a sinusoid curve in the (r, ) plane unique to that
point. If several points are located on the same line, they
produce sinusoids crossing at the parameters for that line.

Centres and radius are


found using Hough
Transform

Display the output


outputcircles

Stop

Figure:3 Coordinates of the eyes looking in forward direction

Phase-1
The first phase in the classification process is to take the picture
of normal person eyes, looking in forward direction and finding
the diameter of cornea using imdistline function, which is shown
in figure 2

Phase-2
Apply the same above procedure to the another image in which
the eyes are looking towards left side. This is shown in the
figure 4

Figure: 2 Normal person eyes looking in forward direction with


imdistline
Next applying the viscircles command to the image then the
circles are drawn on the cornea. The center and radius are also
shown in the figure 3

Figure: 4 Coordinates of eyes looking towards left side and


center, radius.
Phase-4
Pre-operative
The fourth phase in the classification process is to take the
picture of a squint eye person, looking in forward direction and
finding the diameter of cornea using imdistline function, this is
shown in figure 5

Figure 6 Coordinates of eyes looking in upward direction and


center, radius.
Post-operative
Apply the same above procedure to the another image in which
the eyes are looking in forward direction. This is shown in the
figure 7

Figure 5 Coordinates of eyes looking in forward direction and


center, radius
Apply the same above procedure to the another image in which
the eyes are looking in upward direction. This is shown in the
figure 6

Figure 7 Coordinates of eyes looking in forward direction and


center, radius
Apply the same above procedure to the another image in which
the eyes are looking in left side direction. This is shown in the
figure 8

upright

up

Upleft

Center(left)

429

209.3

493.1

185.8

521.2

201.6

Center(right)

136.6

217.5

176.1

193.1

184.2

197.4

Radius(left)

35.5

39.1

36.9

Radius(right)

33.4

41.5

40.9

right

centre

left

Center(left)

401.1

222.2

471.1

219.9

524.9

214.3

Center(right)

75.3

240.9

158.8

221.7

215.1

221.9

Radius(left)

39.5

34.4

41.4

Radius(right)

39.5

33.4

38.6

Table 2 post-operative

3 CONCLUSION
Figure 8 Coordinates of eyes looking in upward direction and
center, radius

Pre-operative
upright

up

Center(left)

420.9

223.8

460.8

Center(right)

151.3

234

171.3

Upleft
220.
5
225.
3

506.3

220.8

211.4

213.1

Radius(left)

35.5

36

32.4

Radius(right)

32.5

35.4

39.9

right

left

Radius(left)

34.2

centre
227.
492.7
5
232.
172.2
6
38.7

Radius(right)

38.5

35.2

Center(left)

441.6

230.8

Center(right)

130.7

235.1

Table 1 pre-operative
Post-operative

525

238.3

197.4

231.8

32.3

This project has presented a Strabismus screening system, in


which Hough Transform segmentation stage is based on
accuracy and higher efficiency rate. Hough was found to be the
better method for the segmentation process. So the entire
process was carried out using Hough Transform. Properly
detecting the inner and outer boundaries of cornea texture is
important for all Strabismus screening systems. An automatic
segmentation algorithm was presented, which would localize the
cornea region from an eye image and isolate eyelid, eyelash and
reflection areas. Threshold was also employed for isolating
eyelashes and reflections of the image. Next, using Hough
Transform the segmented iris region was normalized to
eliminate the present dimensional inconsistencies between iris
regions. The results of the work are very favourable in that they
are reasonably accurate across a wide range of test cases and run
acceptably fast. This gives reason for optimism about the
success of the overall project. It seems that a pure software
solution for the detection of strabismus is quite possible, and
will merely require additional effort to produce the accuracy and
consistency needed to make

4 REFERENCES

35.6
[1] Robust and accurate eye contour extraction Vladimir
Vezhnevets, Anna Degtiareva 2003 In Graphicon.
http://citeseer.ist.psu.edu/viewdoc/summary?
doi=10.1.1.189.2314
[2] Object Detection using Circular Hough Transform. Introduction
to the Hough Transform by Fatoumata Dembele.
http://lockdog.ru/files/pdf_lib/uav_ip/Object%20Detection
%20using%20Circular%20Hough%20Transform.pdf

[3] International Journal of Electronics Communication and


Computer Technology (IJECCT) Volume 2 Issue 3 (May 2012)
A Study on Human Eye Detection in a Color Image Based on
Harris Corner Detector and Sobel Edge Operator ISSN: 22497838
[4] FENG, G. C., AND YUEN, P. C. 2001. Multi-cues eye detection
on gray intensity image. Pattern Recognition 34, 5, 10331046.
[5] FISCHLER, M. A., AND BOLLE, R. C. 1981. Random sample
consensus: A paradigm for model fitting with applications to
image analysis and automated cartography. Communication of
the Association of Computer Machinery 24, 5, 381395.
[6] KAMPMANN, M., AND ZHANG, L. 1998. Estimation of eye,
eyebrow and nose features in videophone sequences. In
International Workshop on Very Low Bitrate Video Coding
(VLBV 98), Urbana, USA, 101104.

[7] LAM, K.-M., AND YAN, H. 1996. Locating and extracting the
covered eye in human face images. Pattern Recognition 29, 5,
771779.
[8] RADEVA, P., AND MARTI, E. 1995. Facial features
segmentation by model-based snakes. In International
Conference on Computing Analysis and Image Processing,
Prague.
[9] STEWART, C. V. 1999. Robust parameter estimation in
computer vision. SIAM Review 41, 3, 513537.
[10] TIAN, Y., KANADE, T., AND COHN, J. 2000. Dual-state
parametric eye tracking. In Proceedings of the 4th IEEE
International Conference on Automatic Face and Gesture
Recognition (FG00), 110115.

You might also like