Automatic Image Anonymizer Alex Brettingen James Esposito.

Automatic Image Anonymizer

Alex BrettingenJames Esposito

GoalsTake any input image and

remove, distort, or cover all human faces

Retain the original integrity of the input image

Step One: Detect Faces

Viola-Jones Object (Face) Detection FrameworkOutlined here – http://www.cs.cmu.edu/~efros/courses/LBMV07/Papers/viola-IJCV-01.pdf

Viola - Jones

Feature types and evaluation:sums of image pixels within rectangular areas four different types of features used in the

framework:

value of any given feature is equal to the sum of the pixels within white rectangles subtracted from the sum of the pixels within dark rectangles

Viola - JonesLearning Algorithm in a standard 24x24 pixel sub-window, there

are 162,336 possible features

the Viola – Jones Algorithm employs a variant of the learning algorithm ‘AdaBoost’ to both select the best features and to train classifiers that use them.

Viola - Jones

For this project, we used the Computer Vision Toolbox Matlab add-on to implement our Facial Detection (highly recommended)

http://www.mathworks.com/products/computer-vision/

How accurate is the Algorithm?

How accurate is the Algorithm?

IS IT A FACE?

12

3

4

5

12

3

4

5

Anonymizer

Now that we know that the algorithm is effective at detecting faces, we can find applications for it

One such application is protecting the identities of people in photographs

Anonymizer

We must alter the area of the photograph containing faces

Blurring, covering entirely, or replacing with another image are possible methods

Method 1: Gaussian Blur

Method 1: Gaussian Blur

Method 1: Gaussian Blur

Method 1: Gaussian Blur

Method 2: Black-out

Method 2: Black-out

Method 3: Image Replacement

Anony–mice-er

Method 3: Image Replacement

Purrrrrrfect Anonymization

Remaining workSmooth blur edgesTry a pixelation methodBlending Image Replacement

Questions?