Using Principal Component Analysis to Improve Accessibility

Ben Simondet

University of Minnesota - Morris

April 30, 2016

Outline

Introduction

  • What's the problem?

Background

  • What is PCA?
  • How does PCA work?

​Ways PCA Can Help

  • Facial Recognition
  • Emotion Recognition
  • Eye Tracking

​Conclusions

Introduction

Background

Methods

Conclusions

Introduction

A Normal Day

  • Knock on the door
  • What do you do?
  • What if that's not easy?

A Normal Day

  • What if you go to have a conversation with a friend?
  • What if you can't figure out what they're feeling?

What's the problem?

  • User Experiences
  • Limitations
  • Diversity

Introduction

Background

Methods

Conclusions

Background

Principal Component Analysis (PCA)

  • Tons of Applications
    • Statistics
    • Facial Recognition
    • Neuroscience
    • Increasing Accessibility?
  • Data Simplification
  • Predictor Models

PCA Algorithm

  1. Find principal components
  2. Evaluate eigenvalues
  3. Re-plot data

Finding Principle Components

Finding Principle Components

Finding Principle Components

First Principal Component

Finding Principle Components

Second Principal Component

Evaluate Eigenvalues

Large eigenvalues = More influence

Large eigenvalues = More influence

Small eigenvalues

= Less influence

Finding Principle Components

Finding Principle Components

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

x

Evaluate Eigenvalues

2-D to 1-D Graph

120-D (not possible as a physical graph)

to 2(or 3)-D Graph

Re-plot Data

Billions Month Eigenvalue
.66 11 2

X=11*2

Y=.66*2

(22,1.32)

Re-plot Data

Re-plot Data

Introduction

Background

Methods

Conclusions

Methods

Facial Recognition

Emotion Recognition

Eye Tracking

Method 1:

Facial Recognition

Facial Recognition

  • One of the most common uses for PCA
  • Human face has a set of features
  • Remote Door Access
    • Physical and Memory Assistance

Facial Recognition

  • Physical
    • Allow user to know who is at the door
    • Automatically let in specific people
    • Grant people door access without needing to go to the door
  • Memory
    • Ability to assign friendly nicknames to visitors to improve future memory

Application: Web Based Door Access

Facial Recognition

  • Very similar to base PCA algorithm
  • Image processing steps to accomplish

Algorithm Change

Facial Recognition

  • Eigenfaces
  • Mean Face

Algorithm Change

Facial Recognition

Emotion Recognition

Eye Tracking

Method 2:

Emotion Recognition

Emotion Recognition

  • Similar to face recognition
  • Specific parts of the face
  • Change faces to make emotions
  • Tons of applications
    • Augmented Reality - Identify emotions of those around the user
    • Emotion Training - Identify emotions in a flash card format

Emotion Recognition

Application: Assistance with Emotion Recognition

  • Adults and children with Autism Spectrum Disorder tend to struggle when identifying emotions of others (Rump, Giovannelli, 2011)
  • After some practice, this can improve
  • Augmented Reality
  • Practice

Emotion Recognition

Application: Assistance with Emotion Recognition

  • 94 points on each face
  • Same people make different emotions
  • Joy, Surprise, Disgust, Fear, Anger, Sorrow 

Emotion Recognition

Application: Assistance with Emotion Recognition

  • Currently existing face databases don't show emotion
  • Modifying emotion increases access
  • Also provides labeled data for flash cards

Facial Recognition

Emotion Recognition

Eye Tracking

Method 3:

Eye Tracking

Eye Tracking

  • Eyes move differently in those with ASD
  • Pupil movement can be plotted using PCA
  • Early detection and diagnosis of ASD (Mlouka and Martineau, 2014)
  • Replacement for mouse/keyboard

Eye Tracking

Application: Assistance with ASD Diagnosis

  • Measuring gaze patterns/blinks
  • Focus times
  • Reaction times

Eye Tracking

Application: Assistance with ASD Diagnosis

Introduction

Background

Methods

Conclusions

Conclusions

Conclusions

Current Obstacles

  • Lack of commercial solutions
  • Homemade/small-scale
  • Availability
  • Large data sets

Conclusions

What does the future hold?

  • Smart homes
  • Wider availability
  • More computing power
  • Endless possibilities!

References

  • "Synthesis of emotional expressions specific to facial structure" - Agarwal, Chatterjee, Mukherjee
  • "Face recognition using face-autocropping and facial feature points extraction" - Karmakar, Murthy
  • "Face recognition and facial expression identification using pca" - Meher, Maben
  • "Principal component analysis of eye-tracking data during visual perception of human faces in aduls and children with autism" - Mlouka, Martineau
  • "The development of emotion recognition in individuals with autism" - Rump, Giovannelli
  • "Web-based online embedded door access control and home security system based on face recognition" - Sahani, Nanda, Sahu, and Pattnaik

Any Questions?

Made with Slides.com