Face Description with Local Binary Patterns: Application to Face Recognition

1. Face Description with Local Binary Patterns: Application to Face Recognition TimoAhonen, AbdenourHadid and MattiPietikainen

2. Overview • Motivation • Local Binary Pattern Methodology • Application to Face recognition

3. Motivation • 2-D surface texture is a valuable cue in machine vision • To develop leading-edge methodology for 2-D texture analysis; • To create basis for new applications of machine vision. • Guiding principles • Computational simplicity for real-time operation • Invariance wrt. illumination changes • Invariance wrt. spatial rotation of objects

4. Description of Local image texture • Texture at gcis modeled using a local neighborhood of radius R, which is sampled at P (8 in the example) points: • Let’s define texture T as the joint distribution of the gray levels gc and gp (p=0,…,P-1): • T = t(gc,g0 ,…,gP-1 )

5. Description of Local image texture (cont.) • Without losing information, we can subtract gcfrom gp : T = t(gc, g0-gc,…, gP-1-gc) • Assuming that gcis independent of gp-gc, we can factorize above: T ~ t(gc) t( g0-gc ,…, gP-1-gc) • t(gc) describes the overall luminance of the image, which is unrelated to local image texture, hence we ignore it: T ~ t( g0-gc ,…, gP-1-gc) Above expression is invariant wrt. Gray scale shifts

6. LBP: Local Binary Pattern • Invariance wrt. to any monotonic transformation of the gray scale is achieved by considering the signs of the difference: T ~ t( s(g0-gc),…, s(gP-1-gc)) Where • Above is transformed into a unique P-bit pattern code by assigning binomial coefficient 2p to each s(gp-gc):

7. LBP: Local Binary Pattern (cont.)

8. LBP: Local Binary Pattern (cont.) • LBPP,Rencodes simple binary microstructures into P-bit number: • LBPP,Rprovides less information than signed difference p8 but: • invariant wrt. To any monotonic transformation of the gray scale • Vector quantization not needed • Computational simplicity

9. LBP: Example • Local Binary Pattern (LBP) is a texture descriptor which codifies local primitives (such as curved edges, spots, flat areas, etc.) into a feature histogram.

10. Rotation invariant LBP

11. Rotation invariant LBP (cont.)

12. Uniform Pattern Heuristic hypothesis • Certain local binary patterns are fundamental properties of texture, providing a vast majority, sometimes overall 90%, of all 3x3 patterns in the observed textures: • Define the concept of ‘uniform’ patterns, which have a limited number of spatial transitions • Use only uniform patterns • Exclude ‘nonuniform’ patterns of high angular frequency (they provide statistically unreliable information)

13. Uniform Pattern (cont.)

14. Uniform Pattern (cont.)

15. Face Description with LBP

17. Dissimilarity Measures

18. Face Recognition with LBP

19. Experimental Results