cpsc 425: computer visionlsigal/425_2018w2/lecture23b.pdf · grouping in human vision slide credit:...
TRANSCRIPT
![Page 1: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/1.jpg)
Lecture 23: Clustering
CPSC 425: Computer Vision
40
![Page 2: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/2.jpg)
Grouping in Human Vision
Humans routinely group features that belong together when looking at a scene. What are some cues that we use for grouping?
41
![Page 3: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/3.jpg)
Humans routinely group features that belong together when looking at a scene. What are some cues that we use for grouping? — Similarity — Symmetry — Common Fate — Proximity — ...
42
Grouping in Human Vision
![Page 4: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/4.jpg)
43
A. Kanizsa triangle B. Tse’s volumetric worm C. Idesawa’s spiky sphere D. Tse’s “sea monster”
Figure credit: Steve Lehar
Grouping in Human Vision
![Page 5: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/5.jpg)
44
Grouping in Human Vision
Slide credit: Kristen Grauman
![Page 6: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/6.jpg)
45
Grouping in Human Vision
Slide credit: Kristen Grauman
![Page 7: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/7.jpg)
Clustering
It is often useful to be able to group together image regions with similar appearance (e.g. roughly coherent colour or texture) — image compression — approximate nearest neighbour search — base unit for higher-level recognition tasks — moving object detection in video sequences — video summarization
46
![Page 8: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/8.jpg)
Clustering
Clustering is a set of techniques to try to find components that belong together (i.e., components that form clusters). — Unsupervised learning (access to data, but no labels)
Two basic clustering approaches are — agglomerative clustering — divisive clustering
47
![Page 9: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/9.jpg)
Agglomerative Clustering
Each data point starts as a separate cluster. Clusters are recursively merged.
Algorithm: Make each point a separate cluster Until the clustering is satisfactory Merge the two clusters with the smallest inter-cluster distance end
48
![Page 10: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/10.jpg)
Agglomerative Clustering
49
![Page 11: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/11.jpg)
Agglomerative Clustering
50
![Page 12: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/12.jpg)
Agglomerative Clustering
51
![Page 13: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/13.jpg)
Divisive Clustering
The entire data set starts as a single cluster. Clusters are recursively split.
Algorithm: Construct a single cluster containing all points Until the clustering is satisfactory Split the cluster that yields the two components with the largest inter-cluster distance end
52
![Page 14: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/14.jpg)
53
Divisive Clustering
![Page 15: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/15.jpg)
54
Divisive Clustering
![Page 16: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/16.jpg)
55
Divisive Clustering
![Page 17: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/17.jpg)
56
Divisive Clustering
![Page 18: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/18.jpg)
Inter-Cluster Distance
How can we define the cluster distance between two clusters and in agglomerative and divisive clustering? Some common options:
the distance between the closest members of and
– single-link clustering
the distance between the farthest members of and a member of
– complete-link clustering
57
min d(a, b), a 2 C1, b 2 C2
max d(a, b), a 2 C1, b 2 C2
C1 C2
C1 C2
C1 C2
![Page 19: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/19.jpg)
58
How can we define the cluster distance between two clusters and in agglomerative and divisive clustering? Some common options:
an average of distances between members of and
– group average clustering
C1 C2
1
|C1||C2|X
a2C1
X
b2C2
d(a, b)
C1 C2
Inter-Cluster Distance
![Page 20: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/20.jpg)
Dendrogram
The algorithms described generate a hierarchy of clusters
59
Forsyth & Ponce (2nd ed.) Figure 9.15
![Page 21: CPSC 425: Computer Visionlsigal/425_2018W2/Lecture23b.pdf · Grouping in Human Vision Slide credit: Kristen Grauman. 45 Grouping in Human Vision Slide credit: Kristen Grauman. Clustering](https://reader034.vdocuments.us/reader034/viewer/2022042404/5f1bffa192cc18723b221b95/html5/thumbnails/21.jpg)
Dendrogram
The algorithms described generate a hierarchy of clusters, which can be visualized with a dendrogram.
60
Forsyth & Ponce (2nd ed.) Figure 9.15