DigitalImageProcessing14-Color Image Processing 2

8/4/2019 DigitalImageProcessing14-Color Image Processing 2

1/61

[email protected]

Digital Image Processing

Color Image Processing


2/61

2

of

62Contents

Full-color processing

Color transformations

Smoothing and sharpening

Color segmentation

Noise in Color Image


3/61

3

of

62

A full color image has a vector at each pixel. For colour

images, these vectors each have 3 or 4 components.

There are two Categories to process vectorial images:

First Category Process each component image individually

Form a composite processed color image

Second Category

Work with color pixel directly.

Color pixels are vectors.

Full Color Image Processing


4/61

4

of

62

Scalar process.

Red

Scalar process.Green

Scalar process.Blue

Red

Green

Blue

Marginal Processing

Each channel is processed separately:



5/61

5

of

62

Red

Green

Blue

Vectorialprocess.

Red

Green

Blue

The colour triplets are processed as single units:



6/61

6

of

62Full Color Image Processing

For example: in RGB system, each color point can beinterpreted as a vector extending from the origin to that point in

the RGB coordinate system.

For an image with size M*N, there are MN such vectors.


7/61

7

of

62

In order for per-color-component and vector basedprocessing to be equivalent:

1 The process has to be applicable to both vectors andscalars

2 the operation on each component of a vector must beindependent of other components.

In order for per-color-component and vector basedprocessing to be equivalent:

1 The process has to be applicable to both vectors andscalars

2 the operation on each component of a vector must beindependent of other components.For example Neighborhood averaging

per-color-component processing = vector based processing

For example Neighborhood averaging

per-color-component processing = vector based processing



8/61

8

of

62Color Transformation

Different with gray level: pixel values are triplets or quartets.


9/61

9

of

62

For example:

RGB color space

n=3, r1,r2, r3 denote the red, green, blue

CMYK or HSI space are chosen,

n=4 or n=3

Color Transformation


10/61

10

of



11/61

11

of

62

Remarks

Any transformation can be performed in any color

model.

Actually, some operations are better suited to

specific models.



12/61

12

of



13/61

13

of



14/61

14

of

62

Si=kri Si=k ri +1-k S3=kr3=k*I


Wrong in text book.


15/61

15

of

62

HSI transformation involves the fewest number ofoperations.

But the computations required to convert an RGB or

CMYK image to HSI space more than offsets theadvantages of the simpler transformation.

Regardless of the color space selected, the output is

the same.



16/61

16

of

62

Reducing the red color plane by 50

levels moves the color balance

towards cyan.

Reducing the green color plane by

50 levels moves the color balance

towards magenta.

Reducing the blue color plane by 50levels moves the color balance

towards yellow



17/61

17

of

62Color Complements

Same as gray level

negative.

Color complements are

useful for enhancing detail

that is embedded in dark

regions of a color image


18/61

18

of

62Color Complements

Note: have no corresponding

functions in HSI space in color

complements.


19/61

19

of

62

Highlighting a specific range of colors

Separating objects from their surrounds.

Basic idea

Display the colors of interest

Use the regions defined by the colors as a maskfor further processing.

More complicated than gray-level counterpartsRequire each pixels transformed color components

to be a function of all n original pixels colorcomponents.

Color Slicing


20/61

20

of

62Color Slicing


21/61

21

of

62Color Slicing


22/61

22

of

62

Most Common UsesPhoto enhancement

Color reproduction.

Consistent: since these transformations are developed,

refined, and evaluated on monitors, it is necessary to maintainhigh color consistency between monitor and output devices.

Device-Independent Color Model.

Tone and Color Corrections


23/61

23

of

62Tone and Color Corrections

The principle of calibrated

imaging systems is that they

allow tonal and color

imbalances to be corrected

interactively and

independently. That is, in two

sequential operations.


24/61

24

of

62Tone and Color Corrections

Notes: every action

affects the overall colorbalance of the image.

That is the perception of

one color is affected by

its surrounding colors.


25/61

25

of

62

Application Examples: Face and tongue diagnosis

Image

capture

Color

CorrectionSegmentation

Feature

Extraction1 1 1 2 1

2 1 2 2 2

1 2

.. .

.. .

... ... ... ...

.. .

n

n

m m m n

a a a

a a a

a a a

Classification

Data Base

Results!



26/61

26

of

62

Correction

Model



27/61

27

of

62

CIE L*a*b* CIELAB Model: Device independent L*: Lightness

a* b*: Color, a* represent red minus green, and b* forgreen minus blue



28/61

28

of

62

After

correction

Tone and Color CorrectionsTone and Color Corrections

Before

correction


29/61

29

of

62

Automated way, and very efficient in Gray image

processing.

Unwise to histogram equalize the components of a color

image independently.

Lead to erroneous color.

A more logical approach

Spread the colorintensities uniformly

Leaving the colors themselves (e.g Hue) unchanged.

Histogram Processing


30/61

30

of

62Histogram Processing


31/61

31

of

62

Modify value based on the characteristics of the

surrounding pixels.

Smoothing

Sharpening

Color image smoothing

Gray image: Each pixel is replaced by the average of thepixels in the neighborhood defined by the mask.

Deal with component vectors.

Smoothing and Sharpening


32/61

32

of

62

=

=9

99

9

i

izR

1/91/9

1/9

1/91/9

1/9

1/91/9

1/9

Color Image Smoothing


33/61

33

of

62Color Image Smoothing

Note: The results is

the same whensmoothing each

plane of starting

RBG image using

conventional gray

scale neighborhood

processing.

34


34/61

34

of


Filtered by

5*5

averaging

mask

35


35/61

35

of


Smoothing intensity

component only

36


36/61

36

of


37


37/61

37

of

62

RGB smoothingRGB smoothingI smoothingI smoothing

HIS smoothing

Wrong!

HIS smoothing

Wrong!

Color Image Smoothing

38


38/61

38

of

62Color Image Sharpening

),1(),1([),( yxfyxfyxf ++=

)1,()1,( +++ yxfyxf )],(4 yxf

fyxfyxg 2),(),( =

),1(),1(),(5 yxfyxfyxf +=

)1,()1,( + yxfyxf 39


39/61

39

of

62Color Image Sharpening

0 -1 0-1 5 -1

0 -1 0Sharpening in RGB Sharpening in HIS-Intensity

40


40/61

40

of

62

Segmentation: Partition an image into regions

Segmentation in HSI Color Space

Segmentation in RGB Vector Space

Color Edge Detection

Color Segmentation

41


41/61

41

of

62

Segment an image based on color in HSI space:

Color is conveniently represent in the hue

image.

Saturation is used as a masking image inorder to isolate further regions of interest in

the hue image.

Intensity infrequently used for segmentation,

Since it carries no color information.

Segmentation in HSI Space

42


42/61

42

of

62

Thresholding the

Saturating image with

a threshold equal to10% of the maximum

value in the

saturation image.

Segmentation in HSI Space

Region of interest

43


43/61

43

of

62

Segmentation in RGB Space

HSI Space: more intuitive.

RGB: better results.

Objective: Segment objects of a specified color range in an

RGB image.

Given a set of sample color points representative of thecolors of interest.

An estimate of the averagecolor that we wish to segment.


44


44/61

44

of

62Segmentation in RGB Space

9),( DazD

45


45/61

45

of

62

Generalization:


Where C is the covariance matrix of the samples

representative of the color we wish to segment. WhenC=I, it is as the described in the preceding page.

9),( DazD

46


46/61

46

of


However, implementing of last two methods iscomputationally expensive for images of practical size,

even if the square roots are not computed.A compromise is to use a bounding box.

Given an arbitrary color point, we segment it bydetermining whether or not it is on the surface or inside the

box.

47


47/61

47

of

62

Select sample by rectangularregion

compute its mean vector

a

Compute standard deviation

r g9 bGet box:

ar- 1.25r ~ ar+ 1.25r

ag- 1.25g ~ ag+ 1.25g

ab- 1.25b ~ ab+ 1.25b

Codin 1in the box


48


48/61

48

of


Segmentation results in RGB space Segmentation results in HSI space

49


49/61

49

of

62

An important tool for image segmentation.

Computing edges on an individual-image basis

Vs.

Computing edges in color vector space.

Edge detection

Gradient operators.

Color Edge Detection

50


50/61

50

of

62Color Edge Detection

Consider two M*M images (M is odd), considering the gradient

value at point [(M+1)/2, (M+1)/2]

51


51/61

51

of


52


52/61

52

of


53


53/61

of


Represented by Di Zenzo in 1986

54

C


54/61

of


Obtain by

vector

method

Obtain by

add three

RGB

gradient

components

Thedifference

55

C l Ed D i


55/61

of


56

N i i C l I


56/61

of

62

The noise content of a color image

The same characteristics in each color channel

For color channels to be affected differently by noise.

The relative strength of illumination: Red filter in a

CCD camera

CCD sensors are noisier at lower levels ofillumination.

Noise in Color Images

57

N i i C l I


57/61

of

62Noise in Color Images

Take a brief look at noise in color images and how noise carries

over when converting from one color model to another.

Noise is fine grain

noise such as thistends to be less

visually noticeable in a

color image than it is in

a monochrome image.

58

N i i C l I


58/61

of


HSI

With noise

HSI

No noise

59

N i i C l I


59/61

of


Note:

Hue and saturation components of

noisy image are significantly

degraded.This is due to the nonlinearity of

the cos and min operations in theRGB to HIS transformation

On the other hand, the intensity

component is slightly smoother than

any to the three noisy RGBcomponent images.Regarding the fact that image

averaging reduces random noise.

( ) ( )[ ]

( ) ( ) ( )[ ]

( )[ ]

( )BGRI

BGRBGR

S

BGBRGR

BRGR

with

GBif

GBifH

++=

++=

+

+=

>

=

3

1

),,min(1

1

cos

360

212

21

1

60

N i i C l I


60/61

of


61

S


61/61

of

62Summary

In this lecture we have learned:Color Fundaments

Color Model

Pseudo-color processingFull-color transformation

Smoothing and Sharpening

Color SlicingNoise in color images

Next we will look at Representation &

DigitalImageProcessing14-Color Image Processing 2

Documents