Scribd
Upload
25 views

Week 1 Report: +different of Gaussian (Dog) + Sift

The document discusses image analysis techniques including Scale-Invariant Feature Transform (SIFT) to detect keypoints, bag-of-features to construct feature vectors from image patches, and Gaussian mixture models. It presents results of applying these methods using Matlab programs and the Vlfeat library to classify images of different sonar objects, and evaluates several point process models including Strauss, hardcore, and softcore processes fit to the keypoint locations.

Uploaded by

Pangolin Le
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
25 views

Week 1 Report: +different of Gaussian (Dog) + Sift

The document discusses image analysis techniques including Scale-Invariant Feature Transform (SIFT) to detect keypoints, bag-of-features to construct feature vectors from image patches, and Gaussian mixture models. It presents results of applying these methods using Matlab programs and the Vlfeat library to classify images of different sonar objects, and evaluates several point process models including Strauss, hardcore, and softcore processes fit to the keypoint locations.

Uploaded by

Pangolin Le
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Week 1 Report:

+Different of Gaussian (DoG) + SIFT:

Use to detect keypoint


Describe the features of keypoint.

Bag of features:
Detection and description of image patches
Assigning patch descriptors to a set of predetermined clusters (a vocabulary) with a vector quantization
algorithm
Constructing a bag of keypoints, which counts the number of patches as-signed to each cluster
Applying a multi-class classifier, treating the bag of keypoints as the feature vector, and thus determine
which category or categories to assign to
the image.

Matlab program: Use Vlfeat library:


SIFT detector (plot descriptor)

(a)

(b)

(c)
(d)
SIFT detector and descriptor: (a) Mud, (b) ripple, (c)posidonie, (d) sand_posi.

Bag of features Matlab results:


Demo program:
Also use Vlfleats library, dense SIFT descriptor, Support vector machine classifier ( linear SMV).
Use caltech101 data with 5 classes: Accordion, airplane, budha, ant and camera
15 training images, 15 data images.
Confusion matrix:
Confusion matrix (92.00 % accuracy)

accordion

airplanes

ant

budha

camera

accordion

airplanes

ant

budha

camera

GMM: KMeans intialization

GMM: Gaussian mixture - kmeans init

GMM: Gaussian mixture - random init

Gaussian Mixture model:


The Gaussian Mixture Model (GMM)

The log-likelihood of this model:

EM algorithm:

Repeat until Loglikelihood convergence.

Results: Training with 20 sona images.


2000

1800

1800

1600

1600

1400

1400

1200

1200

1000
1000

800
800

600

600

400

400

200

200
0

1800
1600
1400
1200
1000
800
600
400
200
0

(a)

(b)

(c)

1800

1400

1600
1200

1400
1000

1200

800

1000
800

600

600
400

400
200

200
0

(d)
(e)
Keypoint classification of images: (a) Sona1, (b) Sona2, (c) Sona3, (d) Sona4, (e) Sona5

Week 3: Strauss Point Process:

0
0

10

20

30

40

50

60

70

80

90

100

-2740
-2760
-2780
-2800
-2820
-2840
-2860
-2880
-2900
-2920
-2940

10

20

30

40

50

60

70

80

90

100

Gamma Predict

Maximum Pseudo Likelihood

Choose r=6, Gamma= 0.614

Condition Intensity function

Model diagnostics (raw residuals)


Diagnostics available:
four-panel plot
mark plot
smoothed residual field
x cumulative residuals
y cumulative residuals
sum of all residuals
sum of raw residuals in clipped window = -1.575e-11 8

Kest

Strauss r=6

StraussHar
dcore r=6

Softcore(ka
ppa=0.5)

-2150.018
-1765.172
-2219.694
-1728.662
-1695.617
-2107.925
-1107.001
-881.5504
-1053.122
-582.7899
-1018.985
-1213.966
-2220.523
-1691.061
-1900.423
3

-2149.969
-1764.972
-2218.69
-1728.614
-1695.396
-2107.925
-1106.953
-881.193
-1052.917
-582.7004
-1018.923
-1213.544
-2220.523
-1691.001
-1900.423
8

-2179.729
-1780.183
-2283.463
-1747.326
-1714.108
-2122.054
-1108.335
-882.559
-1055.874
-582.6918
-1011.754
-1217.783
-2261.285
-1696.457
-1915.615
0

Poisson
Sona01_01
Sona01_21
Sona01_40
Sona02_01
Sona02_21
Sona02_40
Sona03_01
Sona03_21
Sona03_40
Sona04_01
Sona04_21
Sona04_40
Sona05_01
Sona05_21
Sona05_40
Time of
maximum

-2181.348
-1780.236
-2285.435
-1747.328
-1714.287
-2128.879
-1110.386
-882.6423
-1055.875
-582.965
-1019.234
-1217.897
-2264.706
-1697.715
-1921.357
0

PairPiece
(r=5)
-2148.034
-1773.054
-2235.879
-1737.707
-1700.734
-2120.438
-1109.643
-882.2873
-1055.841
-582.4404
-1018.683
-1216.747
-2238.694
-1694.802
-1905.489
1

DiggleGratt
on
(0.05,0.2)
-2179.87
-1780.236
-2285.435
-1746.569
-1714.04
-2122.48
-1110.386
-882.5832
-1055.875
-582.9356
-1019.164
-1217.897
-2261.417
-1695.566
-1915.436

LennardJon
es
-2179.083
-1774.519
-2283.811
-1740.42
-1712.54
-2122.265
-1105.97
-880.6295
-1050.915
-33789152
-1011.711
-1217.039
-2261.026
-1693.013
-1915.371
4

P-value of homogenous StraussHard(r=roptimum) with no rbord (edge effect)


Image
Sona01_01
Sona01_21
Sona01_40
Sona02_01
Sona02_21
Sona02_40
Sona03_01
Sona3_21
Sona03_40
Sona04_01
Sona04_21
Sona04_40
Sona05_01
Sona05_21
Sona05_40

Optimum r
5.3
6.2
6.4
6.4
6.4
5.9
6.1
4.4
6.4
8.8
7
6
5.9
6.4
6.7

p-value
0.0003015077
0.008560914
7.128579e-08
0.001921654
0.002268519
0.01058475
0.2263023
0.3377286
0.1437486
0.3081564
0.3087066
0.2521565
3.354012e-05
0.04243896
0.000384445

P-value of inhomogenous StraussHard(r=roptimum) model with rbord=9.5


Image

Optimum r

p-value

Sona01_01
Sona01_21
Sona01_40
Sona01_15
Sona01_30
Sona02_01
Sona02_21
Sona02_40
Sona02_15
Sona02_30
Sona03_01
Sona3_21
Sona03_40
Sona03_15
Sona03_30
Sona04_01
Sona04_21
Sona04_40
Sona04_15
Sona04_30
Sona05_01
Sona05_21
Sona05_40
Sona05_15
Sona05_30

5.4
6.1
6.4
6.3
6.6
6.4
6.5
6.1
5.9
6.1
6.7
2.4
6.6
5.4
5.4
8.8
6.9
7
9.5
8.3
6.7
6.4
6.7
6.3
6.7

1.629268e-05
0.003474408
1.541811e-08
0.0001993449
6.290544e-07
5.674017e-05
5.357882e-06
0.001032961
1.368404e-08
3.0307e-06
0.006028275
0.1544988
0.05494958
6.30801e-09
0.1892866
0.06928745
0.3051947
0.03310093
0.03410365
0.005665654
1.09733e-06
0.009255479
2.782162e-06
0.0127437
6.612725e-06

sum of raw
residuals
-2.693e-08
-1.401e-09
-9.213e-09
-5.264e-11
-5.187e-10
-2.522e-08
-5.651e-08
-1.626e-11
-6.802e-08
3.0307e-06
-3.703e-10
-2.091e-07
-6.225e-07
2.119e-12
-5.623e-09
-2.091e-07
-4.323e-07
-2.656e-08
-4.198e-08
-5.691e-10
-9.549e-11
-5.651e-08
-1.626e-11
-6.802e-08
3.0307e-06

LennardJones
0.7953043
0.1619573
0.9114976
0.2343573
0.2132667
0.6585665
0.5456646
0.1234234
0.4565464
0.1619573
0.3838797
0.8038365
0.6917028
2.377465e-11
0.08558589
0.1877029
0.06520478
0.8916772
0.3112048
0.4009809
0.7435345
0.2342342
0.2343241
0.1132332
0.6546568

QQplot, Strauss model, Edge correction, Sona01_01

You might also like