Large-Scale, Real-World Face Recognition in Movie Trailers

Large-Scale, Real-World Face Recognition in Movie Trailers

Week 2-3Alan Wright

(Facial Recog. pictures taken from Enrique Gortez)

Preliminary Steps

• Extract Facial Tracks- Working on MATLAB code now

• Worked on detecting blurry images, no solid results.

• Extract the features from the facial tracks.• Build framework to load and test data.• Begin with baseline testing (Sparse, min, meant,

etc)• Algorithm development…

Blur Detection

• Canny Edge Detection• Hough transform• Hough Lines• Find perpendicular line• Using that perpendicular line, get two parallel

lines on each side of the Hough line. • Choose 10 points on each side to find the

gradient.

Hough Lines

Using Perpendicular lines

Gradient Points

Good Edge

Mean

Inte

nsity

Pixels 1 - 20 (10 on each side of the Hough Line)

Bad Edge

Results

• Bad Hough Lines

• Dataset is not ideal for this algorithm, but works well on larger photos.

Facial Recognition

Linear Combination

+ x2 + x3

+ x4 + x5 + x6

+ x7 + x8 + x9

Test Image

= x1

Training Images

Linear Combination

y

Testing

A=

=

Training

x

Coefficients

Now in videos…

• We have:

Instead of:

Baseline

Sparse Representation-based Classification (SRC)

+ x2 + x3

+ x4 + x5 + x6

+ x7 + x8 + x9

Test Image

= x1

Training Images

+ 0 + 0 + 0

+ 0 + 0 + 0

SRCSparse

Linear

SRC

• Method– Impose sparsity on coefficient vector

• We want to minimize the coefficient sum to enforce sparsity.

(Wright09)

Minimize coef.

Possible Baseline Algorithms

• Sum up the coefficient vector and take: average, min, etc..

• SRC linear combination.• Then creating our own algorithm…

Related Papers Read

• “Face Tracking and Recognition with Visual Constraints in Real-World Videos”– Project Page

• “Large Scale Learning and Recognition of Faces in Web Videos”