Eye Locator

1
Eye Locator

• Presented By
• Amanda Buczkowski

2
Project Description/Goals

• First Stage Eye Tracking System
• Given Picture of Person, Locate Face
• Locate Persons Eyes, Crop Eyes
• Identify Each Separate Eye and Crop Each Eye

3
Approach

• Used MATLAB
• Divide Task into 5 Different Steps
• Skin Tone Filter (Locate Face)
• Image Processing on Face
• Edge Detection (Zero-crossing)
• Intensity Filter
• Find Eyes
• Draw Box around Both Eyes and Crop Them
• Draw Box around Each Separate Eye and Crop

4
Skin Tone Filter

• Read Image
• RGB -gt YIQ
• Check Every Pixel For Skin Tone
• Skin Tone -gt White
• Non-Skin Tone -gt Black

5
Skin Tone Filter Reducing Noise

• Morphological Operation Open
• Small Size Filter Based on Area of Image

6
Cropping the Face

• Label Each Region
• Use Bounding Box
• Cycle Through Each Region
• Find Top and Bottom Boundaries
• Find Left and Right Boundaries
• Draw Box Around Face on Original Picture
• Crop Face Region

7
Cropping the Face
8
Problems With Skin Tone Filter

• Background
• Has Skin Tone Regions
• Most of Background is Skin Tone

9
More Background Examples
10
ProcessFace

• processFace Performs Several Image Processing
  Techniques on Face
• Cropped Face Passed as a Parameter to processFace
• RGB -gt Grayscale

11
Zerocrossing

• Perform Edge Detection
• If Thresh gt.2 -gt White
• Else -gt Black

12
Morph open operation

• Helps Eliminate Some Noise
• Separate Thinly Connected Regions

13
Small Size Filter

• Label Each Region
• Use Area Feature
• If Area of Region lt totalArea .0035, Region
  is Deleted

14
Intensity Filter

• Use Image Right After Edge Detection
• Draw Histogram
• Take Top 30 Based on Histogram Values

15
Deleting Large and Small Regions

• Label Each Region
• Use Area Feature
• Delete Regions More Than 3 of Total Area
• Delete Regions Less Than .3 of Total Area
• Percents Are Based on Only Having Face Region
• Experimental Percentages

16
Deleting Long Regions

• Regions Too Long Cannot be Eyes
• If Height of a Region gt 1/5 of Height of Image,
  Region is Deleted

17
Eye Location Algorithm

• Delete Regions
• Top 10 of Height
• Bottom 40 of Height
• Remaining Regions Which are Most Aligned
  Horizontally are Determined to be Eyes
• Zero or One Region
• Error Message is Printed
• Program Terminates
• Two Regions
• Test if There is Alignment
• If Yes, These are Eyes
• IF Not, Error Message, Program Exits

18
Eye Location Algorithm Continued

• Three or More
• Create Alignment Array
• Store Alignment of Every Set of Two Regions in
  Array
• Create Deletion Array
• Set all Values to One
• Set Max Alignment to Zero, Tie to Zero
• Cycle Through Alignment Array
• Find Max Alignment, With Location Numbers (Row,
  Column Numbers)
• See if a Tie for Max Alignment Occurs
• If There is a Tie
• Print Error Message
• Exit Program

19
Eye Location Algorithm Continued

• Set Locations with Max Alignment to 0 in Delete
  Array
• Delete all Regions that Have Value of 1 in Delete
  Array
• Remaining Regions Are Determined to be Eyes

20
Possible Eye Candiates

• Top 10 of Height Ignored
• Bottom 40 of Height Ignored
• Regions in These Locations Are Deleted

21
Eye Location Examples
Tie!
22
Analysis

• Algorithm Works About 50 of Time
• Lighting Issues
• Particular Types of Pictures Used
• Pictures With Little Background Interference
• Pictures of People Looking Forwards

23
Problem Nose Shadow
24
Confusing Hair with Eyes
TIE ERROR
25
Eye Locations on color picture

• Find Top and Bottom Pixel Heights
• Find Leftmost and Rightmost Pixel Values
• Pass Original Image, x, y, width, height Values
  to Function box2

26
Enlarged Eyes

• Crop Eyes
• Use x, y, width, height values already found
• Enlarge Eyes
• Factor of 2

27
Box drawn around each eye

• Label Each Region
• Use Bounding Box
• Find x, y, width, height of Each Bounding Box
• Draw Box Around Eye Each Using These Numbers

28
Eyes

• Crop Each Eye
• Enlarge by Factor of 2

29
Special Thanks

• Intelligent Human-Computer Interaction Laboratory
• Dr. Bouchaffra
• Jun Tan
• Suvarna
• Intelligent Information Engineering Laboratory
• Shuo Feng
• Rishi
• Aiyesha