Cs-4495-2B-Model_Fitting

Front 1

What is a parametric model?

Back 1

A model that represent a class ofinstances where each is defined by a value ofthe parameters.

Examples include lines, or circles, or even aparameterized template.

Hint 1

2B-L1

Front 2

How can line fitting be difficult?

Back 2

There can be noise in edge points.

Only parts of the line can be detected.

There can be extra edge points.

Hint 2

2B-L1

Front 3

How does voting help detect features?

Back 3

Cycle through features, each casting a vote for model parameters.

Look for the most popular votes.

Hint 3

2B-L1

Front 4

Why is voting effective?

Back 4

Noise will not cast enough votes to be detected as a feature.

Hint 4

2B-L1

Front 5

What is the main idea behind a Hough transform?

Back 5

1. Each edge point votes forcompatible lines.

2. Look for lines that get manyvotes.

Hint 5

2B-L1

Front 6

What is Hough space?

Back 6

Hint 6

2B-L1

Front 7

What is the equation to transform a point in Hough space to a line?

Back 7

Hint 7

2B-L1

Front 8

How does the Hough algorithm work?

Back 8

Hint 8

2B-L1

Front 9

What is the polar representation of a line?

Back 9

x * cos(theta) + y * sin(theta) = 0

Hint 9

2B-L1

Front 10

How is a polar Hough array constructed?

Back 10

Hint 10

2B-L1

Front 11

What is the Hough transform algorithm for polar coordinates?

Back 11

Hint 11

2B-L1

Front 12

How does noise impact a Hough transform?

Back 12

Long segments will still get a high amount of points, but votes might end up being split across buckets.

Hint 12

2B-L1

Front 13

What is the advantage of using the gradient of an edge point with Hough voting?

Back 13

Instead of computing each possible line for an edge point, the gradient reduces it down to one possible line, which dramatically reduces computational complexity.

Hint 13

2B-L1

Front 14

What is Hough space for a circle with known radius?

Back 14

Each edge point creates a circle based on the circle center.

Hint 14

2B-L2

Front 15

What is the Hough space for a circle with unknown radius?

Back 15

Each point creates a cone in Hough space.

Hint 15

2B-L2

Front 16

What is the Hough space for a circle with unknown radius, and a gradient?

Back 16

Each point creates a three dimensional line.

Hint 16

2B-L2

Front 17

What is the algorithm for Hough voting for circles?

Back 17

There is a typo in this slide...

a = x + r * cos( theta )

Hint 17

2B-L2

Front 18

What are some pros of parameterized Hough space?

Back 18

1. All points are processed independently, so can copewith occlusion

2. Some robustness to noise: noise points unlikely tocontribute consistently to any single bin

3. Can detect multiple instances of a model in a singlepass

Hint 18

2B-L2

Front 19

What are some cons of parameterized Hough space?

Back 19

1. Complexity of search time increases exponentiallywith the number of model parameters

2. Non-target shapes can produce spurious peaks inparameter space

3. Quantization: hard to pick a good grid size

Hint 19

2B-L2

Front 20

How is a generalized Hough Transform Table created?

Back 20

1. At each boundary point,compute displacementvector: r = c – pi

2. Measure the gradientangle θ at the boundarypoint.

3. Store that displacement ina table indexed by θ.

Hint 20

2B-L3

Front 21

What is the pseudo code of a Hough Transform of a template with known orientation and scale?

Back 21

For each edge point:

Compute gradient direction θRetrieve displacement vectors r to vote for reference point.

Hint 21

2B-L3

Front 22

What is the pseudo code of a Hough Transform of a template with unknown scale S?

Back 22

For each edge point:

For each possible master scale S:Compute gradient direction θFor θ’ retrieve displacement vectors rVote r scaled by S for reference point.

Hint 22

2B-L3

Front 23

What is the pseudo code of a Hough Transform of a template with unknown orientation?

Back 23

For each edge point:

For each possible master θ*: Compute gradient direction θNew θ’ = θ - θ*For θ’ retrieve displacement vectors r to vote forreference point.

Hint 23

2B-L3

Front 24

What is a Code book?

Back 24

Clustering used on patches around an interest points. By storing displacements relative to an objects center, for each codebook entry, can "learn" to locate an object in an image.

Hint 24

2B-L3