Slides from Dr Shahera Hossain Pixel Neighborhoodbased Feature The most important for texture analysis is to describe the spatial behavior of intensity values in any given neighborhood ID: 461592
Download Presentation The PPT/PDF document "LBP & More" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
LBP & More
Slides
from Dr.
Shahera
HossainSlide2
Pixel Neighborhood-based Feature
The most important for texture analysis is to
describe
the spatial behavior of intensity values in any given neighborhood.
Different
methodologies have been proposed.
Local binary pattern (LBP) is one of the most-widely used approach – mainly for face recognition.
LBP is used for texture analysis too. Slide3
Local Binary Pattern (LBP)
Binary
threshold function
is,
For each PIXEL of an image, a BINARY CODE is produced
to make a new matrix with the new value (binary to decimal value).
where
, neighborhood pixels in each block is thresholdedby its center pixel value p sampling points (e.g., p = 0, 1, …, 7 for a 3x3 cell, where P = 8)r radius (for 3x3 cell, it is 1).
Slide4
Component-wise multiplication
Computation of Local Binary Pattern
1
2
4
128
8
64
32
16020128
?
0
0
0
16
0101Nc0001
3
7
2
8
4
1
235
∑
146
Neighborhood
of a gray-scale image
Binary code for >
N
c
Representation
Sum
LBP
Example of how the LBP operator works
topSlide5
Computation of LBP
Code/Weight
:
1 x 2
7
1 x 2
6
1 x 25
1 x 240 x 23
0x 220x 211 x 20 = 128= 64= 32= 16= 0= 0= 0= 1
1 x 2
7
1 x 2
6
1 x 251 x 240 x 230x 22
0x 211 x 20 = 128= 64= 32= 16= 0= 0= 0= 10
(LSB
)
Binary Pattern:
1
(MSB)
1
110001(LSB)LBP:1 + 0 + 0 + 0 + 16 + 32 + 64 + 128 = 241Slide6
A New Method
: DCLBP
A
new method called
diagonal-crisscross local binary pattern (DCLBP) for texture
representation is proposed recently.Basic concept: An image feature should take diagonal pixel variations as well as horizontal and vertical (crisscross) pixel variations in the neighborhood, so that it can perform well even in the cases of rotations in images.
Two
other variants of the LBP, considering rotational feature & mean/median of the neighbor pixels are also proposed
These are compared with LBP, median-LBP, interpolation-LBP, number-LBP, neighborhood-intensity-LBP.Slide7
Diagonal-Crisscross
Local Binary Pattern
(DCLBP
)
(a)
(b)
(c)(d)
(LSB)
N
0 – N4N1 – N5N2 – N6N3 – N7topSlide8
3x3
image
patch
Start
from N
0 pixel
positionGet differences for front-diagonal values (N0 – N4), in vertical direction (N1 – N
5), for back-diagonal
direction (N2 – N6), and in horizontal direction (N
3 – N7)Multiply each difference with 21, 23, 25 and 27 sequentially to compute a new valueTake the mean value of newly-computed value and the central pixel value Diagonal-Crisscross Local Binary Pattern (DCLBP)…1357Slide9
where
,
Sampling-point
set is,
Cardinality
of the set is
,
Difference
parameter
is,
The binary threshold function
is,
Diagonal-Crisscross Local Binary Pattern…
p
0 = 21 p1 = 23 p2 = 25 p3 = 27 Slide10
Here, the cardinality,
that demonstrates that it has 4 different-possible values as ‘2
1
, 2
3
, 25 and 27’. when
, we get = which, covers the front-
diagonal difference of
. Similarly,when
covers the vertical difference of ;when covers the back-diagonal difference of ;when covers the horizontal difference of Through
this manner, we get the new central pixel value for each patch.
Diagonal-Crisscross Local Binary Pattern…Slide11
Median-RILBP
where
,
symbolizes ‘rotational-invariant’ nature;
is pattern (e.g., 8 for a 3x3 patch);
is
radius (1 for 3x3 patch);
Rho (
)
is the rotational function where it circularly does bitwise right-shift operation. is 8 when if . so 8 times rotationThe bit-shift is done 8 times if . The concept here is to rotate the neighbors.Finally, compute the value that the neighbor chain may represent. Slide12
Mean-RILBP
where
,
symbolizes ‘rotational-invariant’ nature;
is pattern (e.g., 8 for a 3x3 patch);
is radius (1 for 3x3 patch);
Rho (
) is the rotational function where it circularly does bitwise right-shift operation.
is 8 when if
. so 8 times rotationThe bit-shift is done 8 times if . The concept here is to rotate the neighbors.Finally, compute the MEAN value. Slide13
Experiment: Dataset 1
Figure:
Sample images for each class of USC-SIPI Rotated Textures dataset.
Figure:
Seven rotated images for
bark
class (USC-SIPI Rotated Textures dataset).Slide14
USC-SIPI
(i.e., University of Southern California – Signal and Image Processing Institute))
Rotated
Textures
dataset
Each image is digitized at seven different rotation angles: 0, 30, 60, 90, 120, 150, and 200 degree
for each of the 13 different imagesThese images are taken from the Brodatz database which is the most widely-used dataset
Experiment: Dataset 1 – USC-SIPI DB
USC-SIPI DatabaseTotal classes13
No. of images per class7Rotations7Image size512x512 pixelsTotal images91Slide15
Experiment: Dataset
2
Ten different classes for KTH-TIPS dataset
(sandpaper
, crumpled aluminum foil,
styrofoam
, sponge, corduroy, linen, cotton, brown bread, orange peel and cracker B) Slide16
Experiment: Dataset
2 – KTH-TIPS Database
KTH-TIPS
(TIPS stands for ‘Textures under varying Illumination, Pose and Scale’)
database.
KTH-TIPS Database
Total classes10No. of images per class81Rotations & variations81Image size200x200 pixelsTotal images
810Slide17
Classification
Lets
employ
K-nearest neighbor classifier (KNN) &
Multi-class support vector machine (SVM) (radial basis kernel function, gamma = 1). We consider 10-fold cross-validationSlide18
Recognition Results – 1
USC-SIPI
R
otated Textures
DB
Feature
KNN SVM
LBP
46.074.7
Interpolation_LBP46.180.2Median_LBP30.880.2MedianR_LBP84.693.4Number_LBP92.387.9NI_LBP23.165.9
Mean-RILBP
100
95.6
Median-RILBP
92.396.7DCLBP10090.1
USC-SIPI DatabaseTotal classes13No. of images per class7Rotations7Image size512x512Total images91Slide19
Recognition
Results – 2
KTH-TIPS DB
Feature
KNN
SVM
LBP
8053.7Interpolation_LBP
6053.1Median_LBP7046.4MedianR_LBP7028Number_LBP9035.2NI_LBP7055.7Mean-RILBP
70
23.4
Median-RILBP
70
29.6DCLBP9072.5
KTH-TIPS DatabaseTotal classes10No. of images per class81Rotations & variations81Image size200x200Total images810topSlide20
Thank you very much for your kind attention