Local Binary Patterns and its application in Face Recognition

1
Local Binary Patterns and its application in Face
Recognition

• Fan Lu
• CS 766 Computer Vision
• Class Project

2
References

• T. Ojala et al, A comparative study of texture
  measures with classification based on feature
  distributions, Pattern Recognition, vol. 29, pp.
  51-59, 1996.
• T. Ojala et al, Multiresolution gray-scale and
  rotation invariant texture classification with
  local binary patterns, IEEE Transaction on
  Pattern Analysis and Machine Intelligence, vol.
  24, pp. 971-981, 2002.
• Timo Ahonen et al, Face Recognition with Local
  Binary Patterns, ECCV 2004, LNCS 3021, pp.
  469-481, 2004.

3
Local Binary Patterns 3

• The original defined using 33 neighborhoods
• Extensions
• Circular neighborhood with different radiuses
  using interpolation
• Introducing the concept of uniform patterns with
  at most two bitwise transitions, e.g. 0000000 and
  1000011

4
Constructing the LBP histogram 3

• Notation LBPP,Ru2
• (P,R) neighborhood
• u2 stands for using only uniform patterns and
  labeling all other patterns with a single label.
• Hi?x,yIfl(x,y)i, i0,,2P-1
• I is the indicator function
• 2P-1 is only when we count all patterns
• We can also construct H for different regions
  indexed by j.

5
Dissimilarity Measures 3
6
Face Recognition with LBP 3
7
Questions of (my) interests

• How to estimate the optimal or tune the
  parameters, e.g. (P,R), u2, wj?
• How to separate regions for face or non-face
  images?
• Which dissimilarity measure (maybe a modified/new
  one) will perform generally better or just for a
  specific task?
• More efficient computation for image database.

8
A study plan under modification

• Acquire face image sets of moderate size from the
  web
• Try face classification/clustering using
  different combination of parameters, separating
  schemes and dissimilarity measures, check the
  average performance or just see if dissimilarity
  matrix makes sense
• Test the optimal choices on new image