EVC - Computer Vision Recap 2 Content: Global Operations - - PowerPoint PPT Presentation

EVC - Computer Vision Recap 2

• Content:
• Global Operations
• Morphological Operations
• Image Segmentation
• Multiscale Representations
• Image Features – Interest Points
• Stereo & Motion
• Computational Photography

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 1

Global Operations - Fourier Transform

• Fourier Transform:

converts a function from the spatial (or time) domain to the frequency domain

2 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

dt e t g G

x i

∫

∞ ∞ − −

⋅ =

ω

π ω ) ( 2 1 ) (

Time Domain and Frequency Domain

• Time Domain:
• Tells us how properties (air pressure in a sound function, for

example) change over time:

• Amplitude = 100
• Frequency = number of cycles in one second = 200 Hz

3 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Time Domain and Frequency Domain

• Frequency domain:
• Tells us how properties (amplitudes) change over frequencies:

4 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• A Fourier Transform is an integral transform that re-expresses a

function in terms of different sine waves of varying amplitudes, wavelengths, and phases.

• So what does this mean exactly?

Fourier Transform

Can be represented by: When you let these three waves interfere with each other you get your

• riginal wave function!

Let’s start with an example…in 1-D

5 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• Since this object can be made up of 3 fundamental frequencies an

ideal Fourier Transform would look something like this:

• Notice that it is symmetric around the central point and that the

amount of points radiating outward correspond to the distinct frequencies used in creating the image.

Fourier Transform

6 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

This image exclusively has 32 cycles in vertical direction. This image exclusively has 8 cycles in horizontal direction.

Let’s Try it with Two-Dimensions!

You will notice that the second example is a little more smeared

• ut. This is because the lines are more blurred so more sine waves

are required to build it. The transform is weighted so brighter spots indicate sine waves more frequently used.

7 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Magnitude vs. Phase

8 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Power Spectrum

A B C 1 2 3

9 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Hough-Transformation

• Detection of straight lines in grayscale images

10 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• P.V.C. Hough 1962: US Patent
• Line in Parameterform (Hess):

θ θ sin cos y x r + =

Image Space

r

• r

Parameter Space

θ θ

• r

y x

θ θ sin cos y x r + = θ

Hough Transformation

11 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Image space Parameter space

Hough Transform for Collinear Points

• All lines that pass through a point P(x,y) are defined by:
• Collinear points are detected in

parameter space:

θ θ sin cos y x r + =

Example

12 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

A simple example

φ p

13 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Morphology

• Morph means “Shape”
• We do Morphology for

Shape Analysis & Shape Study.

• Shape analysis easy in case of binary images, pixel locations

describe the shape.

• Digital Morphology is a way to describe or analyze the shape of
• bjects in digital images

14 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Structuring Element

• Essential part of morphological operations

is the Structuring Element used to probe the input image.

• Two-dimensional structuring elements

consist of a matrix, much smaller than the image being processed.

• Structuring Elements can have varying

sizes

• Element values are 0,1 and none (!)
• Structural Elements have origin (anchor

pixel)

15 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Erosion

• Brief Description
• To delete or to reduce something
• Pixels not matching a given pattern are deleted from the image.
• Basic effect
• Erode away the boundaries of regions of foreground pixels (i.e.

white pixels, typically).

Common names: Erode, Shrink, Reduce

16 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Erosion

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 17

A 3×3 square structuring element

Set of coordinate points = { (-1, -1), (0, -1), (1, -1), (-1, 0), (0, 0), (1, 0), (-1, 1), (0, 1), (1, 1) }

Effect of erosion using a 3×3 square structuring element

Dilation

• Dilation is the set of all points in the image, where the

(reflected) structuring element “touches” the foreground.

• Consider each pixel belonging to the foreground
• Replicate the structuring element

18 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Dilation

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 19

Effect of dilation using a 3×3 square structuring element 3×3 square structuring element

Set of coordinate points = { (-1, -1), (0, -1), (1, -1), (-1, 0), (0, 0), (1, 0), (-1, 1), (0, 1), (1, 1) }

Combining Dilation and Erosion

• Morphological Opening
• Opening is defined as an erosion, followed by a dilation.
• Morphological Closing
• Closing is defined as a dilation, followed by an erosion.

20 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Opening

• Structuring element: 3x3 square

1 1 1 1 1 1 1 1 1

21 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Closing

• Structuring element: 3x3 square

1 1 1 1 1 1 1 1 1

22 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Hit-and-miss Transform

• Used to look for particular patterns of foreground and

background pixels

• Very simple object recognition
• All other morphological operations can be derived from it!!
• Input:
• Binary Image
• Structuring Element, containing 0s, 1s and “don’t cares”

23 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Hit-and-Miss Transform

• Effect of the hit-and-miss based right angle convex corner

detector

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 24

Four structuring elements used for corner finding in binary images

Thinning

• A simple way to obtain the skeleton of the character is to thin

the image until convergence.

• The line is broken at some locations, which might cause

problems during the recognition process.

Original image Thresholded image Thinning result

25 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Example Thinning

• We use two Hit-and-miss Transforms

26 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Introduction to Image Segmentation

Grass Sky Tree Tree ? ?

Image Segmentation

27 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

How can we divide an Image into Uniform Regions ?

• Segmentation techniques can be classified as either contextual or

non-contextual .

• Non-contextual techniques ignore the relationships that exist

between features in an image; pixels are simply grouped together

• n the basis of some global attribute, such as grey level.
• Contextual techniques, additionally exploit the relationships

between image features. Thus, a contextual technique might group together pixels that have similar grey levels and are close to

• ne another .

28 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Greylevel Histogram-based Segmentation

• First, we will look at two very simple non-contextual image

segmentation techniques that are based on the greylevel histogram of an image:

• Thresholding
• Clustering

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 29

Greylevel Thresholding

• We can easily understand

segmentation based on thresholding by looking at the histogram of the low noise

• bject/background image
• There is a clear ‘valley’ between

to two peaks

• We can define the greylevel

thresholding algorithm as follows:

• If the greylevel of pixel p <= T

then pixel p is an object pixel else

• Pixel p is a background pixel

30

Background Object T

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• Histogramm
• Problem: Connected image regions do not always have the

same intensity => missing location relation of pixels

Original Histogram Segmentation

Pixel Classification by Threshold

31 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Local Thresholding

• A complex thresholding algorithm is to use a spatially varying
• threshold. This approach is very useful to compensate for the

effects of non –uniform illumination. If T depends on coordinates x and y, this referred to as Dynamic, Adaptive or Local Thresholding.

32 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Local Thresholding

• Finding the local threshold is to statistically examine the intensity

values of the local neighborhood of each pixel

• The statistic which is most appropriate depends largely on the

input image. Simple and fast functions include:

1.

The mean of the local intensity distribution

2.

The median value

3.

The mean of the minimum and maximum values,

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 33

mean T =

median T =

2 max min+ = T

Local Thresholding Example

34

Image contains a strong illumination gradient, global thresholding produces a very poor result Source Image

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Local Thresholding

• Improve: If threshold employed is not

the mean, but (mean-C), where C is a constant

• Using this statistic, all pixels which exist

in a uniform neighborhood (e.g. along the margins) are set to background

35

The result for a 7×7 neighborhood and C=7

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Background Object

c1 c2

Greylevel Clustering

• Consider an idealized object/background histogram
• Clustering tries to separate the histogram into 2 groups defined by

two cluster centres c1 and c2

36 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Greylevel Clustering

• A nearest neighbour clustering algorithm allows us perform a

greylevel segmentation using clustering

• A simple case of a more general and widely used K-means

clustering

• A simple iterative algorithm which has known convergence

properties

• Given a set of greylevels:
• We can partition this set into two groups
• and

37

{ }

g g g N ( ), ( )...... ( ) 1 2

{ }

g g g N

1 1 1 1

1 2 ( ), ( )...... ( )

{ }

g g g N

2 2 2 2

1 2 ( ), ( )...... ( )

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Greylevel Clustering

38

c1 c2

g1 g2

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Greylevel Clustering

• Clustering
• Finds groups of pixels with

similar properties

• Does not guarantee that these

groups form continuous areas in the image

• Even if it does, edges of these

areas tend to be uneven

39 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Relaxation Labelling: Probabilities

• Relaxation labelling is a general technique in computer vision

which is able to incorporate constraints (such as spatial continuity) into image labelling problems

• Assume a simple object/background image
• p(i) is the probability that pixel i is a background pixel
• (1- p(i)) is the probability that pixel i is a object pixel
• Define the 8-neighbourhood of pixel i as {i1,i2,….i8}

40

i i1 i2 i3 i4 i5 i6 i7 i8

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Region-Based Segmentation

• We want smooth regions

in the image

• We still want the pixels in

each region to be similar, and those in adjacent regions to be different

• One way to do this is to

work with regions rather than pixels

• Region growing
• Start with a small ‘seed’

and expand by adding similar pixels

• Split and merge
• Splitting divides regions

that are inconsistent

• Merging combines

adjacent regions that are consistent

41 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Region Growing Summary

• Region growing starts with a small patch of seed pixels
• Compute statistics about the region
• Check neighbors to see if they can be added
• Re-compute the statistics
• This procedure repeats until the region stops growing
• Simple example: We compute the mean grey level of the pixels

in the region

• Neighbors are added if their grey level is near the average

42 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Region Growing Example

43 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Split and Merge

• The opposite approach to region growing is region shrinking

(Splitting)

• It is a top-down approach and it starts with the assumption that

the entire image is homogeneous

• If this is not true, the image is split into four sub images
• This splitting procedure is repeated recursively until we split the

image into homogeneous regions

44 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Split Example

45 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Multiscale Representations

• Image information
• ccurs at all spatial

scales

46 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

High resolution Low resolution

Multi-resolution Image Pyramids

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 47

Image Pyramid

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 48

Sub-Sampling with Gaussian Pre-Filtering

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 49

• Filter the image, then subsample

2 ) * (

↓ = Gaussian G G

The Gaussian Pyramid

Image = G

2 ) * (

↓ = Gaussian G G 2 ) * (

↓ = Gaussian G G 2 ) * (

↓ = Gaussian G G

blur blur blur blur

50 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

High resolution Low resolution

Gaussian Pyramid Frequency Composition

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 51

The Laplacian Pyramid

• Synthesis
• Compute the difference between upsampled Gaussian pyramid

level and Gaussian pyramid level.

• band pass filter - each level represents spatial frequencies

(largely) unrepresented at other level.

52 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Gaussian Pyramid Laplacian Pyramid

G

1

G

2

G

n

G

• =

L

• =

1

L

• =

2

L

n n

G L = ) expand(

1 +

− =

i i i

G G L ) expand(

1 +

+ =

i i i

G L G

The Laplacian Pyramid

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 53

Laplacian Pyramid Frequency Composition

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 54

Corner Detection

• Corner detector
• H. Moravec. Obstacle avoidance and navigation in the real world by a seeing

robot rover. Technical Report CMU-RI-TR-3, Carnegie-Mellon University, Robotics Institute, 1980.

55 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

The Basic Idea

• We should easily localize the point by looking through a

small window

• Shifting a window in any direction should give a large

change in intensity

56 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Corner Detector: Basic Idea

“flat” region: no change as shift window in all directions “edge”: no change as shift window along the edge direction “corner”: significant change as shift window in all directions

57 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Moravec Corner Detector

• Change of intensity for the shift [u,v]:

[ ]

2 ,

( , ) ( , ) ( , ) ( , )

x y

E u v w x y I x u y v I x y = + + −

∑

Intensity Shifted intensity Window function

Four shifts: (u,v) = (1,1), (1,0), (0,1), (-1, 1) Look for local maxima in min{E}

58 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Example

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 59 10 10 10 10 10 10 10 10 10 10 10 10 10 10

E(1,1) = (0-0)² + (0-0)² + (0-0)² + (0-0)² + (0-10)² + (0-0)² + (10-0)² + (0-10)² + (0-10)² = 0 + 0 + 0 + 0 + 100 + 0 + 100 + 100 + 100 = 400 E(1,0) = (0-0)² + (0-0)² + (0-0)² + (10-0)² + (0-10)² + (0-0)² + (10-0)² + (10-10)² + (0-10)² = 0 + 0 + 0 + 100 + 100 + 0 + 100 + 0 + 100 = 400 E(0,1) = (0-0)² + (0-0)² + (0-0)² + (0-0)² + (0-10)² + (0-0)² + (0-0)² + (10-10)² + (0-10)² = 0 + 0 + 0 + 0 + 100 + 0 + 0 + 0 + 100 = 200 E(-1,1) = (0-0)² + (0-0)² + (0-0)² + (0-0)² + (0-10)² + (0-0)² + (0-0)² + (0-10)² + (10-10)² = 0 + 0 + 0 + 0 + 0 + 100 + 0 + 100 + 0 = 200

10 10 10 10 10

Interest Value = 200

Look for local maxima in min{E}: Four Shifts: (u,v) = (1,1), (1,0), (0,1), (-1, 1)

Problems of Moravec Detector

• Noisy response due to a binary window function

 Harris corner detector (1988) solves these problems.

• C. Harris, M. Stephens. “A Combined Corner and Edge Detector”,

Alvey Vision Conference, pp.147-151, 1988

60 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Harris Corner Detector

Noisy response due to a binary window function

• Use a Gaussian function

61 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Harris/Moravec Detector: Some Properties

• But: non-invariant to image scale!

All points will be classified as edges Corner !

62 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

SIFT (Scale Invariant Feature Transform)1

• SIFT is an carefully designed procedure with empirically

determined parameters for the invariant and distinctive features

• Empirically found2 to show very good performance, invariant to

image rotation, scale, intensity change, and to moderate affine transformations

63 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

SIFT Stages:

• Scale-space extrema detection
• Keypoint localization
• Orientation assignment
• Keypoint descriptor

( )

local descriptor

detector descriptor

A 500x500 image gives about 2000 features

64 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• 1. Finding Keypoints – Scale
• Scale selection principle (T. Lindeberg ’94)

 Maxima/minima of Difference of Gaussian

Convolve with Gaussian Downsample Find extrema in 3D DoG space

65 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• 2. Finding Keypoints – Localization

X is selected if it is larger or smaller than all 26 neighbors

66 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Using the available pixel data, subpixel values are generated. This is done by the Taylor expansion of the image around the approximate key point

• 3. Finding Keypoints – Orientation
• Create histogram of local

gradient directions computed at selected scale

• Assign canonical orientation at

peak of smoothed histogram

• Each key specifies stable 2D

coordinates (x, y, scale,

• rientation)

2π

67 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Orientation Assignment

68 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Orientation Assignment

69 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

• 4. Creating Signature
• Thresholded image gradients are sampled over 16x16 array of locations

in scale space

• Create array of orientation histograms
• 8 orientations x 4x4 histogram array = 128 dimensions of feature vector

70 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

SIFT

• Extract features

Find keypoints Scale, Location Orientation Create signature

• Match features
• Nearest neighbor, Hough voting, Least-square affine parameter

fit

71 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

SIFT Example 1

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 72

Maxima in D

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 73

Remove Low Contrast

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 74

Remove Edges

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 75

SIFT Descriptor

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 76

Stereo Vision

• Correct 3d information using 2d images:
• 2 or more images taken from different geometric positions plus

geometric calibration of camera

• Tries to imitate human visual system
• Is also used in the entertainment industry

2 dim + 2 dim + geometry => 3 dim

Stereo photographs Auto stereogram Random Dot Stereogram

77 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Stereo Analysis

• Using two views of an object

from two different positions enables to reconstruct 3d shape of the object out of the geometric differences of the two 2d projections:

• two 2d images
• geometric relation between

cameras

78 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Stereo Geometry

• Perspective projections of
• bject points
• Specific case: both image planes

are parallel => normal case

• Using geometric position of

images planes and object projection depth information is computed D = |x1 - x2|

F1 F2

79 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Disparity

d1 Disparity

Position 1

d2

Position 2

d3

Position 3

Camera Images

80 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Stereo Analysis

F: center of projection X: world coordinate x: image coordinate f: focal length Z: distance to center of projection B: basis, distance between centers of projection

2 1 x

• x

B f Z

• =

x1 B P1´ x1 x2 P P1 P2 X O1 B

• Z

f O2 x1-x2 X-B = Dx Z X x Z X − = Z X − f x1 f x1 Z B

• X

− f x2 = Z B

• X

− f x Z B X

2

− = f x Z X

1

− = f x2

81 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Epipolar Geometry

• Epipolar geometry is a consequence of the coplanarity of the

camera centers and scene point

• The camera centers, corresponding points and scene point lie in a

single plane, known as the epipolar plane

82 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Epipolar Geometry

• Epipolar geometry depends only on the relative pose (position and
• rientation) and internal parameters of the two cameras, i.e. the

position of the camera centers and image planes.

• It does not depend on the scene structure (3D points external to the

camera).

83 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Epipolar Geometry

• Note, epipolar lines are in general not parallel

84 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Correspondence Analysis

• Area Based
• Comparison of intensity values ​in the left and right image
• Correspondence due to the similarity between the intensity values
• Correspondence for each pixel
• Feature Based
• Comparison of features in left and right image
• Correspondence on the basis of selected characteristics of features (edge
• rientation, edge length, gradient, etc.)
• Correspondence only for selected pixels
• more accurate because of sub-pixel positioning of features

Left Stereo Image Right Stereo Image

85 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Area-based Matching

• Correspondence

due to similarity between the gray values ​in the left and right image

2 ,

)] , ( ) , ( [ ˆ ) , ( n j m i I j i I n m SSD

r R j i l

∆ − ∆ − − = ∆ ∆

∑

∈

)] , ( ) , ( [ ˆ ) , (

,

n j m i I j i I n m CC

r R j i l

∆ − ∆ − − = ∆ ∆

∑

∈

xl xr yr yl C(xr ) xr

86 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Feature-based Matching

• Viewpoint-independent edges as a prerequisite for correct

interpretation: violation = wrong values​.

• Edges are basis of evaluation
• Advantage: very accurate
• Disadvantage: very few range points

87 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Structure from Motion

• Motion of an observer relative to

the environment:

• Information about movement of

viewer

• Depth information of the

environment (cf. stereo)

• Problem: direction and amount of

camera movement

• Motion field estimation
• Motion field analysis

88 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Motion Field

• The motion field of a camera which is in motion relative to the

scene is characterized by vectors representing the motion of the corresponding scene points.

• Camera that does not rotate:
• Vectors point radially to or

from a focus

• FOE: Focus Of Expansion
• FOC: Focus Of Contraction
• Point where the motion

vector of the camera intersects image plane

89 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Lightfield

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 90

High Dynamic Range (HDR)

Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2 91

Short exposure Long exposure

Panoramas

1.

Pick one image (red)

2.

Warp the other images towards it (usually, one by one)

3.

blend

92 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2

Morphing = Object Averaging

• The aim is to find “an average” between two objects
• Not an average of two images of objects…
• …but an image of the average object!
• How can we make a smooth transition in time?
• Do a “weighted average” over time t

93 Robert Sablatnig/Sebastian Zambanini, CVL, Recap 2