Main  ◊  Intro  ◊  Active Contours  ◊  Locating Eyes  ◊  Hough Transform  ◊  Neural Networks  ◊  Network Training  ◊  Applet Intro  ◊  The Applet  ◊  Links  ◊  Home Page

Locating the Eyes

Finding the eyes within the faces proved to be a difficult problem.   Because the iris of the eye is usually a perfect circle I choose to use the Circular Hough Transform.   Please read the next section detailing the transform.

Steps for locating eyes within an image

1. Using the contour of the face that was found earlier... restrict the search to a box in the upper part.  The Hough transform is computationally demanding so this is necessary

2. Compute the gradient of the image using the Sobel Transform

3. Guess at a reasonable size for the eye 1/20th to 1/30th the width of the face works well

4. Take the Hough transform of the area for many different sized circles

5. Find the best pair of circles

The last step proved to be the most difficult.  The iris of the eye is partially occluded by the face and eyelid so often the higher gradient between the face and eyebrow would throw it off.  I found it necessary to incorporate some color information into the transform by increasing the gradient value for white areas on the face (assuming the eyeball is generally white)