Computer Vision an der RWTH Aachen | Karteikarten & Zusammenfassungen

# Lernmaterialien für Computer Vision an der RWTH Aachen

Greife auf kostenlose Karteikarten, Zusammenfassungen, Übungsaufgaben und Altklausuren für deinen Computer Vision Kurs an der RWTH Aachen zu.

TESTE DEIN WISSEN

Canny edge detector (steps)

Lösung anzeigen
TESTE DEIN WISSEN
1.  filter image with derivative of gaussian
2. find magnitude and orientation of gradient
3. non-maximum suppression (find local maximum along multi-pixel edge)
4. linking and thresholding (hysteresis) (k_high threshold to start edge curces and k_low threshold to continue them)
Lösung ausblenden
TESTE DEIN WISSEN

Pro/Con of Hough Transform

Lösung anzeigen
TESTE DEIN WISSEN

Pros

• all points processed independently, so can cope with occlusion
• some robustness to noise
• can detect multiple instances of a model in a single pass

Cons

• complexity of search time increases exponentially with number of model parameters
• non-target shapes and produce spurious peaks in parameter space
• hard to pick a good grid size (for the bins)
Lösung ausblenden
TESTE DEIN WISSEN

MoG, EM pros/cons

Lösung anzeigen
TESTE DEIN WISSEN

Pros

- probabilistic interpretation

- soft assignments between data points and clusters

generative model, can predict novel data points

- relatively compact storage

Cons

- local minima

- inizialization (often a good idea to start with some k-means interations)

- need to know number of components

- numerical problems are often a nuisance

Lösung ausblenden
TESTE DEIN WISSEN

Mean-shift

Lösung anzeigen
TESTE DEIN WISSEN

Mode = local maximum of the density of a given distribution (search iteratively)

1. Initialize random seed and window W
2. Calculate center of gravity ("mean") of w
3. shift w around the mean
4. Repeat until convergence

for segmentation:

1. find features
2. initialize windows at individual pixel locations
3. perform mean-shift
4. merge windows that end up near the same mode
Lösung ausblenden
TESTE DEIN WISSEN

Mean-shift pros/cons

Lösung anzeigen
TESTE DEIN WISSEN

Pros

- general, application-independent

- model-free, does not assume shapes

- single parameter with physical meaning (window size)

- finds variable number of modes

- robust to outliers

Cons

- output depends on window size

- finding window size is not trivial

- computation (relatively) expensive

- does not scale well with dimension of feature space

Lösung ausblenden
TESTE DEIN WISSEN

Graph cuts steps of the algorithm

Lösung anzeigen
TESTE DEIN WISSEN

graph g;

for all pixels p:

1. add a node p to the graph

2. set cost of terminal edges (to source and sink --> switched! FG to sink, BG to source)

3. set weights between all adjacent pixels

4. compute max-flow on g

5. label with source/sink-labels for BG/FG

Lösung ausblenden
TESTE DEIN WISSEN

Graph Cuts Pros/cons

Lösung anzeigen
TESTE DEIN WISSEN

Pros

- powerful technique, based on probabilistic models (MRF)

- applicable for a wide range of problems

- very efficient algorithms available for vision problems

- de-facto-standard for many segmentation tasks

Cons

- can only solve a limited class of models

- submodular energy functions E(t,t) + E(s,s) <= E(s,t) + E(t,s)

- can capture only part of the expressiveness of MRFs

- only approximate algorithms available for multi-label case

Lösung ausblenden
TESTE DEIN WISSEN

Support Vector Machines (steps)

Lösung anzeigen
TESTE DEIN WISSEN

1. define representation for each example

2. Select kernel function

3. Compute pairwise kernel values between labeled examples

4. Pass this "kernel matrix" to SVM optimizationsoftware to identify support vectors and weights

5. To classify a new example: Compute kernel values between new input and support vectors, apply woights, check sign (+/-) of output

Lösung ausblenden
TESTE DEIN WISSEN

HOG detector

Lösung anzeigen
TESTE DEIN WISSEN

- Image Window

- Gamma compression (x -> sqrt(x) optional)

- Weighted vote in spatial and orientation cells (8x8 cells with 8/9 orientations in each histogram)

- Contrast normalize over overlapping cells

- Collect HoGs over detection window

- Linear SVM

- Object/Non-Object

Lösung ausblenden
TESTE DEIN WISSEN

Sliding Window Pros/Cons

Lösung anzeigen
TESTE DEIN WISSEN

Pros

- Simple implementation

- past success for certail classes

- good detectors available

Cons

- high computational complexity

- very many windows (false positive rate has to be low)

- not all objects are box-shaped

- assumes 2D and fixed viewpoint (deformable objects not caputed well and less-regular textured ones)

- no context if windows are considered in isolation

Lösung ausblenden
TESTE DEIN WISSEN

Matching (general steps)

Lösung anzeigen
TESTE DEIN WISSEN
1. Find a set of distinctive keypoints
2. Define a region around each keypoint
3. Extract and normalize the region content
4. Compute local descriptor from normalized region
5. Match local descriptors
Lösung ausblenden
TESTE DEIN WISSEN

Boundary Issues (filtering)

Lösung anzeigen
TESTE DEIN WISSEN

- filter window falls off the edge

- extrapolate using ...

- clip filter (black/0)

- wrap around (stretch)

- copy edge

- reflect across edge

Lösung ausblenden
• 196169 Karteikarten
• 4633 Studierende
• 312 Lernmaterialien

## Beispielhafte Karteikarten für deinen Computer Vision Kurs an der RWTH Aachen - von Kommilitonen auf StudySmarter erstellt!

Q:

Canny edge detector (steps)

A:
1.  filter image with derivative of gaussian
2. find magnitude and orientation of gradient
3. non-maximum suppression (find local maximum along multi-pixel edge)
4. linking and thresholding (hysteresis) (k_high threshold to start edge curces and k_low threshold to continue them)
Q:

Pro/Con of Hough Transform

A:

Pros

• all points processed independently, so can cope with occlusion
• some robustness to noise
• can detect multiple instances of a model in a single pass

Cons

• complexity of search time increases exponentially with number of model parameters
• non-target shapes and produce spurious peaks in parameter space
• hard to pick a good grid size (for the bins)
Q:

MoG, EM pros/cons

A:

Pros

- probabilistic interpretation

- soft assignments between data points and clusters

generative model, can predict novel data points

- relatively compact storage

Cons

- local minima

- inizialization (often a good idea to start with some k-means interations)

- need to know number of components

- numerical problems are often a nuisance

Q:

Mean-shift

A:

Mode = local maximum of the density of a given distribution (search iteratively)

1. Initialize random seed and window W
2. Calculate center of gravity ("mean") of w
3. shift w around the mean
4. Repeat until convergence

for segmentation:

1. find features
2. initialize windows at individual pixel locations
3. perform mean-shift
4. merge windows that end up near the same mode
Q:

Mean-shift pros/cons

A:

Pros

- general, application-independent

- model-free, does not assume shapes

- single parameter with physical meaning (window size)

- finds variable number of modes

- robust to outliers

Cons

- output depends on window size

- finding window size is not trivial

- computation (relatively) expensive

- does not scale well with dimension of feature space

Q:

Graph cuts steps of the algorithm

A:

graph g;

for all pixels p:

1. add a node p to the graph

2. set cost of terminal edges (to source and sink --> switched! FG to sink, BG to source)

3. set weights between all adjacent pixels

4. compute max-flow on g

5. label with source/sink-labels for BG/FG

Q:

Graph Cuts Pros/cons

A:

Pros

- powerful technique, based on probabilistic models (MRF)

- applicable for a wide range of problems

- very efficient algorithms available for vision problems

- de-facto-standard for many segmentation tasks

Cons

- can only solve a limited class of models

- submodular energy functions E(t,t) + E(s,s) <= E(s,t) + E(t,s)

- can capture only part of the expressiveness of MRFs

- only approximate algorithms available for multi-label case

Q:

Support Vector Machines (steps)

A:

1. define representation for each example

2. Select kernel function

3. Compute pairwise kernel values between labeled examples

4. Pass this "kernel matrix" to SVM optimizationsoftware to identify support vectors and weights

5. To classify a new example: Compute kernel values between new input and support vectors, apply woights, check sign (+/-) of output

Q:

HOG detector

A:

- Image Window

- Gamma compression (x -> sqrt(x) optional)

- Weighted vote in spatial and orientation cells (8x8 cells with 8/9 orientations in each histogram)

- Contrast normalize over overlapping cells

- Collect HoGs over detection window

- Linear SVM

- Object/Non-Object

Q:

Sliding Window Pros/Cons

A:

Pros

- Simple implementation

- past success for certail classes

- good detectors available

Cons

- high computational complexity

- very many windows (false positive rate has to be low)

- not all objects are box-shaped

- assumes 2D and fixed viewpoint (deformable objects not caputed well and less-regular textured ones)

- no context if windows are considered in isolation

Q:

Matching (general steps)

A:
1. Find a set of distinctive keypoints
2. Define a region around each keypoint
3. Extract and normalize the region content
4. Compute local descriptor from normalized region
5. Match local descriptors
Q:

Boundary Issues (filtering)

A:

- filter window falls off the edge

- extrapolate using ...

- clip filter (black/0)

- wrap around (stretch)

- copy edge

- reflect across edge

### Erstelle und finde Lernmaterialien auf StudySmarter.

Greife kostenlos auf tausende geteilte Karteikarten, Zusammenfassungen, Altklausuren und mehr zu.

## Das sind die beliebtesten StudySmarter Kurse für deinen Studiengang Computer Vision an der RWTH Aachen

Für deinen Studiengang Computer Vision an der RWTH Aachen gibt es bereits viele Kurse, die von deinen Kommilitonen auf StudySmarter erstellt wurden. Karteikarten, Zusammenfassungen, Altklausuren, Übungsaufgaben und mehr warten auf dich!

## Das sind die beliebtesten Computer Vision Kurse im gesamten StudySmarter Universum

##### computer vision

Hochschule Karlsruhe

##### Computer

Universität Siegen