brighten darken - tau

31

61

Dr. Yoram Tal � � � � � � � � � � � � � �� � � � � � � � � � � � � !

"minmax

min255

−

−⋅=II

IIIout

brighten darken

62

Dr. Yoram Tal# $ % & ' ( ) * + , - . * / $ 0 * & $ ' 1 2 3 4Adam

17

33

Dr. Yoram Tal

Histograms

A histogram of a discrete (integer) data set is a frequency table which counts the number of appearances of each member

First introduced by Abu Yusuf Yaqub ibnIshaq al-Sabbah Al-Kindi as a tool for deciphering the substitution code – a breakthrough at that time, 850 AD. (see: simon singh “the code book” 1999)

Who invented it and why?

34

Dr. Yoram Tal

Which Histogram ?

0 5 0 1 0 0 1 5 0 2 0 0 2 5 0

0 50 100 150 200 250

0

0

0

0

0

0

0 5 0 1 0 0 1 5 0 2 0 0 2 5 0

Red Histogram: imhist(im(:,:,1))

Green Histogram: imhist(im(:,:,2))

Blue Histogram: imhist(im(:,:,3))

0 50 100 150 200 250

Gray Histogram: (rgb2gray)

24

47

Dr. Yoram Tal � � � � � � � �Given an array of numbers: �� �� ��� ��������� �������� !� �""��# $#�"�����

A = { }Find the frequency of each number:

%&'()*+,-. /01223/202Number:

Count:

48

Dr. Yoram Tal4 5 6 7 8 9 : ; < = > 8 : < ; ? @x image matr ix (intensity)

imhist (x) display histogram[count, bin] = imhist (x) ;

Gray level

Count# pixels

25

49

Dr. Yoram Tal A B C D E F G H I C J K L M N O PThe histogram of an image contains valuable informationconcerning the distr ibution of gray levels

I t does not contain any spatial information

All the following images have exactly the same histograms!

50

Dr. Yoram Tal

Q R S T U V W X Y S X S Z U T R [ VA histogram is a result of voting it counts the number of supporters (i.,e., pixels) of each candidate (graylevel)

In the simple case of a binary image there are only two candidates: Mr zero and Ms one.

Voting has many useful applications in image processing (as well as in democracy).

26

51

Dr. Yoram Tal

\ ] ^ _ ` _ a b c ^ a d ` e b f ] ^ g h b i j k l m

Too dark

Too bright

52

Dr. Yoram Tal

n o p q r q s t u p s v r w t x o p y z t { | } ~ �

Low contrast

High contrast

27

53

Dr. Yoram Tal� � � � � � � � � � � � � � � � � � �I

I = imread('pout.tif');a = min(I(:));b = max(I(:));J = 255*(I-a)/(b-a);J = uint8(J);

a = 74 b = 224

Histogram of I

255minmax

min ⋅−

−=II

IIJ

54

Dr. Yoram Tal

� � � � � � � � � � � � � � � � � � � � � � � � � �J Histogram of J

Histogram of I

28

55

Dr. Yoram Tal� � � � � � � � � � � � � � � � � �   � ¡¢ £ ¤ ¥ £ ¤ ¦ ¦ § ¨ § © ª« ¬ ­ ® ¯ ° ± ² ­ ³ ± °´ µ ¶ ® ¯ ±· ° ¸ ¬ ¹ ± ¬ º ¶ ° ­ ® ¯

The minimum is 0 and the maximum is 255

56

Dr. Yoram Tal» ¼ ½ ¾ ¿ À Á Â Ã Â Ä Å Æ ½ ¾ Ã Ç È ¾

a = 74

Histogram of I

255 if I i >= b

255*(I-a)/(b-a) if a <= I i < b

0 if I i < a

b = 170

K =

K = imadjust(I,[0.3 0.67],[]);

This is a non-linearoperation

29

57

Dr. Yoram TalÉ Ê Ë Ì Í Î Ï Ð Ñ Ð Ò Ó Ô Ë Ì Ñ Õ Ö Ì × Ø Ù Ú Û Ü ÝI J K

58

Dr. Yoram TalÞ ß à á â ã ä å æ ç è é å ê ß ë å á ß â ìy = histeq(x,256);

30

59

Dr. Yoram Tal í î ï ð ñ ò ó î ô õ ð ö ÷ ø ù ðOriginal Histogram

Equalized Histogram

n = prod(size(x))tf = cumsum(255*h/n )

h

tf

60

Dr. Yoram Talú û û ü ý þ ÿ � � � � � ú ý ÿ �

X (in [0 255])

Linear approx. of tf

L(1:150) = 0;L(151:240) = linspace (0,255 , 90) ;L(241:255) = 255;

y = L(x+1);

y

tf

L

32

63

Dr. Yoram Tal5 6 7 8 9 : ; < = > ? @ < A 6 B < 8 6 9 C D E FPout

64

Dr. Yoram Tal

G H I J K L M N O P Q R N S H T N J H K U V W XY Z [ Z \ ] ^Y Z [ Z \ ] ^ _ _ ^ _ ` [ a Z ` \ b c d^ _ ` [ a Z ` \ b c d e e b f Z g \ Z e ^ d ^ h _ h ib f Z g \ Z e ^ d ^ h _ h i b b d e b _ _ Z \d e b _ _ Z \

paper Histogram of paper

18

35

Dr. Yoram Tal

Histogram Manipulation

• Correct illumination problems • Correct contrast problems

• Use histogram information to create a Look-Up Table (LUT)

• Apply LUT to the image

Image

Histogram LUT

New_imge = LUT(image)

Why?

How?

36

Dr. Yoram Tal

Gamma Correction(point operation)

γ

minmax

min255 ⎟⎟⎠

⎞⎜⎜⎝

⎛−−

⋅=II

IIIout

brighten darken

Input graylevelOut

put

gray

leve

l

[H, b] = imhist(I);p = H/sum(H); (p is in [0,1])LUT = p.^gamma

Iout = LUT(I+1); (if Imin is 0)(gamma + stretching)

19

37

Dr. Yoram Tal

Option Ioperate on each channel separately

R = brighten(I(:,:,1),0.4);G = brighten(I(:,:,2),0.4);B = brighten(I(:,:,3),0.4);

RGB = cat(3,R,G,B);

Warning: distortion of colors!

38

Dr. Yoram Tal

Option IIoperate on the intensity channel

RGB Image HSV Image V = HSV(:,:,3);

Correct V => Vc

rgb2hsv

HSV(:,:,3) = Vc;RGB Imagehsv2rgb

brighten

20

39

Dr. Yoram Tal

Histogram Equalization

Assumption: A “good” image has a “flat” histogram, i.e., all graylevels have the same probability.

How ?See histeq

40

Dr. Yoram Tal

Histogram Equalization - Lemma

• Let I be an image having N pixels and and G+1 graylevels such that g is a graylevel of I iff

g ε {0, 1, 2, …, G}

• Let h be the histogram of I where h(g) is the number of pixels in I having graylevel g.

• Let H be the cumulative histogram of I:

• Then L(x) = (G/N) H(x) is a lookup table which transforms I into a histogram equalized image.

( )0

H ( )g x

gx h g

=

=

= ∑

21

41

Dr. Yoram Tal

Histogram Equalization - how

[H, b] = imhist(I);p = cumsum(H); p = p/p(end);LUT = p*255;

LUT

graylevel

1 Make LUT 2 Apply LUT

Iout = LUT(I+1); (if I is uint8)

In this case, a similar effect could be obtained with imadjust(use stretchlim to find thresholds)

The same result can be achieved with histeq

42

Dr. Yoram Tal

Histogram Equalization problem: a narrow histogram

22

43

Dr. Yoram Tal

Histogram Equalizationresult: too mach contrast

LUT

Resulting histogram

44

Dr. Yoram Tal

CLHEContrast Limited Histogram Equalization

Source histogram Cut at some threshold(the histogram is very high & therefore the LUT is steep)

Redistribute the “left over”Now integrate

Original LUT“softer” LUT

23

45

Dr. Yoram Tal

CLHE - Result

Source HE

CLHE

46

Dr. Yoram Tal

CLAHE Contrast Limited Adaptive Histogram equalization

Problem: a big illumination difference between the two sides of the image (High Dynamic Range).

24

47

Dr. Yoram Tal

Try Histogram Equalization

It certainly helps, but does not solve the main problem.

48

Dr. Yoram Tal

Adaptive Histogram Equalizationrgb => hsv;v1 = adapthisteq(v,{parameters});hsv1 => rgb1

• The image is partitioned into blocks.

• Each block is enhanced with CLHE.

• The results are stitched back smoothly.

“Parameters” are very important!

25

49

Dr. Yoram Tal

CLAHE – another exampleSource image Histogram Eq. adapthisteq

50

Dr. Yoram Tal