introduction to image processing and linear filtering

70
Athens Week 1 Introduction to image processing and linear filtering F. Tupin

Upload: opal-ray

Post on 08-Jan-2018

257 views

Category:

Documents


0 download

DESCRIPTION

Introduction to image processing What is an image ? Spatial resolution and Fourier transform Quantification and histogram Linear filtering Athens Week

TRANSCRIPT

Page 1: Introduction to image processing and linear filtering

Athens Week 1

Introduction to image processing and linear filtering

F. Tupin

Page 2: Introduction to image processing and linear filtering

Athens Week 2

Introduction to image processing

What is an image ?Spatial resolution and Fourier transformQuantification and histogramLinear filtering

Page 3: Introduction to image processing and linear filtering

Athens Week 3

Introduction to image processing

What is an image ?Spatial resolution and Fourier transformQuantification and histogramLinear filtering

Page 4: Introduction to image processing and linear filtering

Athens Week 4

What is an image ?Message support Signal (continuous 2D signal) of a physical measure

Passive imagery (Colour intensity) Active imagery (X-ray transmission,

electromagnetic radiation, …)Other dimensions:

Volumetric measures (3D medical images) 2D ½ (stereo images + intensity) Video (2D+t) 3D+t (images sequences)

Page 5: Introduction to image processing and linear filtering

Athens Week 5

What is a numerical image ?A digital image (2D) is a matrix defined by its resolution (amount of pixels) its depth (amount of potential values for each

pixel) its palette (colour look up table (CLUT))

Page 6: Introduction to image processing and linear filtering

Athens Week 6

Digital Image

Page 7: Introduction to image processing and linear filtering

Athens Week 7

Different kind of imagesImage Application Size Channels

Video VisiophonyTVHD TV

256 x 256720 x 6251920 x 1150

133

Bio-medical Tomography (IRM)Radiography

512x512256x256x2561024 x 1024

1-3

Remote Sensing

1970-19801985-19901990-

2000 x 30006000 x 600015000 x 15000

3-73-203-256

Robot vision Quality controlAutomatic driving

256 x 256512 x 512

12-3

Defense Surveillance, tracking

256 x 256512 x 512

12-3

Page 8: Introduction to image processing and linear filtering

Athens Week 8

From the continuous world to the numerical one Image digitization Sampling (number of pixels) Quantification (number of bits per

pixel)

Page 9: Introduction to image processing and linear filtering

Athens Week 9

From the continuous world to the numerical one Image digitization Sampling (number of pixels) Quantification (number of bits per

pixel)

Page 10: Introduction to image processing and linear filtering

Athens Week 10

Introduction to image processing

What is an image ?Spatial resolution and Fourier transformQuantification and histogramLinear filtering

Page 11: Introduction to image processing and linear filtering

Athens Week 11

Resolution : Vision (physical Image) vs. Digital Image

Resolution Human vision is able to look (in stereo) at from

very small objects to huge ones The digital image has a fixed resolution

determined by the amount of pixels of the image.Sampling Preserving the image frequencies (Shannon

theory, 2D Fourier transform) Preserving the image content: depends of each

application (psycho-visual tests, size of the smallest element to detect, …)

Page 12: Introduction to image processing and linear filtering

Athens Week 12

Resolution of an imageNl : amounts of lines per picture height

Maximum frequency : Nl/2 cycles per picture heightIf D is given (in TV, D=6H), there is a direct relationship between cycles per degree and cycles per picture height

ViewerViewerHH

DD

Page 13: Introduction to image processing and linear filtering

Athens Week 13

Sampling and informationintuitive approach with example

Aliasing 1D : false frequency

Page 14: Introduction to image processing and linear filtering

Athens Week 14

Sampling and informationintuitive approach with example

Aliasing 2D : false frequency and false direction of periodical structures !

Page 15: Introduction to image processing and linear filtering

Athens Week 15

1-D Fourier transform

Sampling at Fe spectrum periodization at Fe Avoid aliasing: low pass filtering in [-Fe/2;Fe/2]All information in [-Fe/2;Fe/2]: normalized frequency

dfefGtg

dtetgfG

tfj

tfj

c

tc

2

2

)()(

)()(

Page 16: Introduction to image processing and linear filtering

Athens Week 16

1D discrete Fourier transform

Signal sampling Fourier periodizationFrequency sampling signal periodizationNumerical signals = implicit periodization in both temporal and frequency domains !

0.5;0.5ou1;0)()( 2

ffengfG nfj

n

Page 17: Introduction to image processing and linear filtering

Athens Week 17

2D Fourier transform

Fourier transform = signal projection on 2D complex sinusoidFrequency domain = new visualization of image informationFourier transform = complex signal (phase + modulus)

Page 18: Introduction to image processing and linear filtering

Athens Week 18

2D Discrete Fourier Transform

2D DFT

Number of frequencies = number of pixels f(i,j) is the image in the spatial domain. Exponential term is the basis function

corresponding to each point, F(k,l), in the Fourier space.

1

0

1

0

)(2),(1),(

N

i

N

j

Nlj

Nkii

ejifN

lkF

Page 19: Introduction to image processing and linear filtering

Athens Week 19

2D DFT - Properties

1

0

1

0

)(2),(1),(

N

k

N

l

Nlj

Nkii

elkFN

jif

F(0,0) corresponds to the average brightnessF(N-1, N-1) represents the highest frequency

Inverse Fourier transform:

Page 20: Introduction to image processing and linear filtering

Athens Week 20

2D DFT computation

Separation of DFT Formula:

Expressing two-dimensional Fourier transform in terms of series of two N one-dimensional transforms decreases the number of required computations.

1

0

2),(1),(

N

i

Nkii

ejifN

jkP

21

0),(1),(

NljiN

jejkP

NlkF

1

0

1

0

)(2),(1),(

N

i

N

j

Nlj

Nkii

ejifN

lkF

Page 21: Introduction to image processing and linear filtering

Athens Week 21

FFTFast Fourier Transform(FFT)

N2 complexity is reduced to Nlog2N. Limited to image size N = 2n.

FFT is complex Real / imaginary or magnitude /phase Magnitude linked to geometric information Phase linked to spatial position To re-transform we need to use both magnitude and phase. If real image, symmetry vs 0

Page 22: Introduction to image processing and linear filtering

Athens Week 22

2-D discrete Fourier transformSimple examples

2 pixel wide vertical stripes Fourier transform

Page 23: Introduction to image processing and linear filtering

Athens Week 23

DFT : simple examplesVertical lines : No variation along y axis Variation along x axis linked to the spatial

frequency of the lines

Fourier peak P(k,l) interpretation: |P(k,l)| related to signal frequency (O,P(K,l)) perpendicular to the pattern

direction

Page 24: Introduction to image processing and linear filtering

Athens Week 24

DFT simple examples

Page 25: Introduction to image processing and linear filtering

Athens Week 25

DFT Real examples

Original image Fourier transform Logarithmic operator applied

Page 26: Introduction to image processing and linear filtering

Athens Week 26

DFT Real examples

Page 27: Introduction to image processing and linear filtering

Athens Week 27

DFT Real examples

Page 28: Introduction to image processing and linear filtering

Athens Week 28

DFT Phase information

The value of each point determines the phase of the corresponding frequency.

The phase information is crucial to reconstruct the correct image in the spatial domain.

Phase image Re-transform using only magnitude

Page 29: Introduction to image processing and linear filtering

Athens Week 29

DFT properties

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

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

),()),((

),()),((

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

),(),(),(

2

2

002

**

00

HFmnhmnf

HFmnhmnf

Femnf

Fmnf

FmnfFFmnfmnf

qpFqpF

D

D

mnj

TFD

TFD

TFDTFDTFDTFD

ZxZ

Page 30: Introduction to image processing and linear filtering

Athens Week 30

Back to the sampling problem…

Must be achieved at a frequency superior to 2 times the highest frequency in the signal (Nyquist frequency)

Or the original image has to be filtered at a frequency below the half of the sampling frequency

Page 31: Introduction to image processing and linear filtering

Athens Week 31

Original and spectrum

Page 32: Introduction to image processing and linear filtering

Athens Week 32

Sub-sampled image and spectrum

Page 33: Introduction to image processing and linear filtering

Athens Week 33

Image comparison

Original image Zoom of the sub-sampled image

Page 34: Introduction to image processing and linear filtering

Athens Week 34

Spectrum comparison

Original spectrum Zoom of the sub-sampled image spectrum

Page 35: Introduction to image processing and linear filtering

Athens Week 35

ExplanationOriginal image spectrum is defined in [-0.5,0.5] The 2-D pattern spectrum [-0.5,0.5] x [-0.5,0.5] is

periodically repeated in (0,1), (O,2),…(p,q),…When doing the sub-sampling: The 2-D pattern spectrum [-0.5,0.5] x [-0.5,0.5] is

periodically repeated in (0,0.5), (O,1),…(p/2,q/2),…

The different patterns overlap in the Fourier domain

The new spectrum corresponds to the [0.25,0.25] x [-0.25,0.25] area

Page 36: Introduction to image processing and linear filtering

Athens Week 36

Initial spectrum and pattern of the sub-sampled spectrum

Page 37: Introduction to image processing and linear filtering

Athens Week 37

Page 38: Introduction to image processing and linear filtering

Athens Week 38

(0,0) (1,0)(-1,0)

Periodization on (p,q)

Page 39: Introduction to image processing and linear filtering

Athens Week 39

(0,0) (1,0)(-1,0)

Periodization on (p/2,q/2)

Page 40: Introduction to image processing and linear filtering

Athens Week 40

(0,0) (1,0)(-1,0)

Selection of [-0.25,0.25]x [-0.25,0.25]

Page 41: Introduction to image processing and linear filtering

Athens Week 41

Page 42: Introduction to image processing and linear filtering

Athens Week 42

Spectrum comparison

Original spectrum Zoom of the sub-sampled image spectrum

Page 43: Introduction to image processing and linear filtering

Athens Week 43

Sub-samplingTo avoid aliasing when sub-sampling Low-pass filtering (suppression of all the

frequencies higher than Fe’/2) Blurring effect but no artefacts are

introduced

Page 44: Introduction to image processing and linear filtering

Athens Week 44

Page 45: Introduction to image processing and linear filtering

Athens Week 45

Introduction to image processing

What is an image ?Spatial resolution and Fourier transformQuantification and histogramLinear filtering

Page 46: Introduction to image processing and linear filtering

Athens Week 46

Depth of an imageBinary representation: b bits, 2b levels

whitewhite

blackblack

11

00 0000

0101

1010

1111

Page 47: Introduction to image processing and linear filtering

Athens Week 47

The depth : Vision vs. Digital Image

Vision sensitivity limited to 6 to 8 bits per colour component i.e.

max 24 bits Vision works in the (350nm) to red (700nm) range

Digital images Much larger depth (in medical imaging and in

remote sensing) Different wavelengths than vision: NMR imaging

(0,001nm) to microwaves images (100000nm) Image visualization: 8 bits = 256 grey-levels(0 for black and 255 for white)

Page 48: Introduction to image processing and linear filtering

Athens Week 48

Visible domainrayons

cosmiquesrayons

gammasrayons

Xultra violet visible infra

rougemicroondes TV radio 50 Hz

10-5 nm 10-3 nm 10 nm 500 nm 1500 nm 5 m 1000 m

VIOLET BLUE GREEN YELLOW RED

380 nm 500 nm 555 nm 600 nm 720 nm

Puech William Université Montpellier II - Nîmes

Page 49: Introduction to image processing and linear filtering

Athens Week 49

Image Histogram

Histogram: grey level distribution F(x) = number of pixels having x as grey-

level If normalized = probability of the grey-level

Used to Contrast enhancement Classification

Page 50: Introduction to image processing and linear filtering

Athens Week 50

Histogram: example

Histogram modes correspond to region of interests (clouds, parts of the boats, etc.)

Page 51: Introduction to image processing and linear filtering

Athens Week 51

Histogram …. be careful !

Page 52: Introduction to image processing and linear filtering

Athens Week 52

Histogram …. be careful !

Page 53: Introduction to image processing and linear filtering

Athens Week 53

Histogram Stretch Linear Stretch

Histogram Equalised Stretch

Page 54: Introduction to image processing and linear filtering

Athens Week 54

Page 55: Introduction to image processing and linear filtering

Athens Week 55

Colour spacesThe RGB space (3 types of phosphors for colour excitation on a screen)

BB

GG

RR

blackblack

whitewhite

Page 56: Introduction to image processing and linear filtering

Athens Week 56

Colours (cont.)Printing industry : CYM(K) Cyan Yellow Magenta (can be translated into RGB)Hue Saturation Intensity (HSI): artists, vision

II

SSHH

blackblack

whitewhiteredred

greengreen

blueblue

Page 57: Introduction to image processing and linear filtering

Athens Week 57

Colours(cont.)YUV : luminance, chrominances in TV black and white TV backward compatibility Decorrelation of the components Y is containing most of the information

Y=0.299R+0.587G+0.114BU=-0.147R-0.289G+0.437BV=0.615R-0.515G-0.100B

Page 58: Introduction to image processing and linear filtering

Athens Week 58

Video formatsCCIR 601: 576 lines of 720 pixels, 25 images per second, 2 interlaced fields per frame, format 4:2:2 (YUYVYUYV) = 165 Mbit/sHDTV: 16/9 format, 2*576 lignes, 720*2(*4/3) pixels (non interlaced and even higher for digital cinema)TV 16/9CIF: CCIR 601 /2 /2 /2QCIF: CIF /2 /2

Page 59: Introduction to image processing and linear filtering

Athens Week 59

Colour CompositesA colour composite is a colour image produced through optical combination of multiband images by projection through filters.True Colour Composite: A colour imaging process whereby the colour of the image is the same as the colour of the object imaged.False Colour Composite: A colour imaging process which produces an image of a colour that does not correspond to the true colour of the scene (as seen by our eyes).

Page 60: Introduction to image processing and linear filtering

Athens Week 60

TM7,4,1

TM5,7,2

TM5,4,3

TM4,3,2

Page 61: Introduction to image processing and linear filtering

Athens Week 61

Introduction to image processing

What is an image ?Spatial resolution and Fourier transformQuantification and histogramLinear filtering

Page 62: Introduction to image processing and linear filtering

Athens Week 62

Linear filteringProperty :F(I+aJ)=F(I)+aF(J)

Image domain: convolutiony=x*hFourier domain : multiplicationY=XH

Page 63: Introduction to image processing and linear filtering

Athens Week 63

Convolution: involves moving a 'window' of a few pixels in dimension (e.g. 3x3, 5x5, etc.) over each pixel in the image, applying a mathematical calculation using the pixel values under that window, and replacing the central pixel with the new value.

Page 64: Introduction to image processing and linear filtering

Athens Week 64

Linear filtering

Page 65: Introduction to image processing and linear filtering

Athens Week 65

Linear Filtering

),(),(),( lmknxlkhmnylk

),(),(),( yxyxyx ffXffHffY

Spatial domain (h convolution kernel)

Frequency domain

Page 66: Introduction to image processing and linear filtering

Athens Week 66

Linear FilteringLow pass filters:noise suppression (= high frequencies)

Page 67: Introduction to image processing and linear filtering

Athens Week 67

Linear filteringHigh or band pass filters: Selection of frequencies of interest

High frequencies (edges) Specific frequencies (texture analysis)

Page 68: Introduction to image processing and linear filtering

Athens Week 68

High pass filter = edge detector

Page 69: Introduction to image processing and linear filtering

Athens Week 69

Stripes of the zebra create high energy waves generally along the u-axis; grass pattern is fairly random causing scattered low frequency energy

y

x

v

u

Page 70: Introduction to image processing and linear filtering

Athens Week 70

Non-linear filtering

Simple non-linear filters = median filters: the output of the filter is the median in a given windowMin / Max : mathematical morphology