application of facial recognition in biometric security kyle ferris

15
Application of Facial Recognition in Biometric Security Kyle Ferris

Upload: vega

Post on 15-Jan-2016

28 views

Category:

Documents


0 download

DESCRIPTION

Application of Facial Recognition in Biometric Security Kyle Ferris. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Application of Facial Recognition in Biometric Security Kyle Ferris

Application of Facial Recognition in Biometric Security

Kyle Ferris

Page 2: Application of Facial Recognition in Biometric Security Kyle Ferris

Introduction

The purpose of this project is to be able to create a "key" for any person who wishes to use the program. An image of the client's face will be

taken and used as the base biometric key. The program should be able to recognize the client and authorize him or her, while denying access

to those not recognized.

Page 3: Application of Facial Recognition in Biometric Security Kyle Ferris

Input Method

A webcam will take a short video of the user The frames of this video will be converted to

image files To acquire test images, plain

photos are taken and used

as input

Page 4: Application of Facial Recognition in Biometric Security Kyle Ferris

Example Input

Page 5: Application of Facial Recognition in Biometric Security Kyle Ferris

PCA

There are many different approaches to this type of problem. One of the most effective is Principal Component Analysis (PCA). PCA involves identifying the principal features of a face (aka, eyes, mouth, jaw structure) and comparing these features for different images

Page 6: Application of Facial Recognition in Biometric Security Kyle Ferris

Algorithms

Conversion Converts a .ppm format file, which is a

color image, into a specially formatted .pgm format.

Pixel Subtraction: Creates a third image which shows the

obvious differences between two very similar images, with no analysis

Page 7: Application of Facial Recognition in Biometric Security Kyle Ferris

Algorithms (continued)

Mean Pixel Area Comparison (MPAC): Averages the pixel intensity in a specified

area and compares that value to the average value in the same area on a second image.

Interference Evaluation: Evaluates how much interference or

errant objects are in the image and tells you how reliable the results are.

Page 8: Application of Facial Recognition in Biometric Security Kyle Ferris

Algorithms (continued)

• Principal Components Labeling

– Evaluates an image and determines groups of pixels which are stored as components

– Uses aspects of edge detection and floodfill

Page 9: Application of Facial Recognition in Biometric Security Kyle Ferris

Example Output

• The “average difference between images” is the example of the Mean Pixel Area Comparison function

• You can see the interference level here as well, and an evaluation from the program

Page 10: Application of Facial Recognition in Biometric Security Kyle Ferris

Incorporating Principal Component Analysis

By default, the program runs by comparing 3x3 blocks of pixels.

For more ideal results, instead of evaluating the similarity of 3x3 blocks, identify the primary components of interest in the images, then compare across those.

This is known as Principal Component Analysis.

Page 11: Application of Facial Recognition in Biometric Security Kyle Ferris

Connected Components Labeling

• Also known as CCL, this is a method to identify which pixels in an image are connected.

• Algorithm returns groups of pixels in the form of arrays of their location.

Page 12: Application of Facial Recognition in Biometric Security Kyle Ferris

Aspects of CCL

Binary filtering method Filter the image into black and white pixels

only based on some arbitrary threshold

Page 13: Application of Facial Recognition in Biometric Security Kyle Ferris

Aspects of CCL (continued)

From this black and white image, run a recursive method that finds groups of connected pixels

Store these groups as arrays of pixel locations

Page 14: Application of Facial Recognition in Biometric Security Kyle Ferris

Incorporating CCL with MPAC

When the MPAC method is run on these groups, it gives better results than when run on just square blocks.

This overall process is Principal Component Analysis.

Page 15: Application of Facial Recognition in Biometric Security Kyle Ferris

Results/Expectations

Will be able to be implemented into any security system with fairly basic equipment.

Able to accurately recognize the user 90% of the time

No false positives- very important to avoid giving authorization to those unauthorized