Robotics and Computer Vision

Intensity Transformations

Dr. Eva Hriekes

Email: evahriekes@gmail.com
 Lecture Outline
 Why transform images?
 Spatial Processing
 Some Basic Intensity Transformation Functions
 Image Negatives

 Log Transformations

 Power-Law (Gamma) Transformations

 Piecewise-Linear Transformation Functions

 Contrast Stretching
 Intensity-level Slicing
 Bit-plane Slicing
 Histogram Processing
 Histogram Equalization

2
 Why Transform images?
 Image has noise in it
 Image is low contrast
 Want to find things in an image
 Want to emphasize things in an image
 Want to remove things in an image
 Want to de-emphasize things in an image

3
 Spatial Processing
 The term spatial domain refers to the image plane itself, and
image processing methods in this category are based on direct
manipulation to the image pixels.
 There are two principle categories of spatial processing:
 Intensity Transformation (Point Processing)

 Spatial Filtering (Local Processing)

 Intensity transformations operate on single pixels of an image,

principally for the purpose of contrast manipulation and image
thresholding.
 Spatial filtering deals with performing operations such as image
smoothing or sharpening, by working on a neighbourhood of
every pixel in an image.

4
 General Spatial Processing
 Spatial Processing Algorithms can be reduced to the form:

𝒈 𝒙, 𝒚 = 𝑻 [𝒇 𝒙, 𝒚 ] Origin x

where f(x, y) is the

input image, g(x, y) is
the processed image
and T is some operator
(x, y)
defined over some
neighbourhood of (x, y)

y Image f (x, y)

5
 Intensity Transformation
 The simplest spatial domain operations occur when the
neighbourhood is simply the pixel itself.
 Here, 𝐠 depends only on the value of 𝐟 at (𝐱, 𝐲). In this case 𝐓 is
referred to as a point processing operation (or intensity
transformation function).
 Point processing operations take the form:
𝑠 = 𝑇(𝑟)
 Where 𝐬 and 𝐫 are variables denoting, respectively, the
intensity of 𝐠 and 𝐟 at any point (𝐱, 𝐲).
 𝐓 is a transformation that maps pixel value 𝐫 into pixel value 𝐬.

6
 Intensity Transformation
 There are three basic types of transformations used frequently for
image enhancement:
 Linear (identity and
negative transformations)
 Logarithmic (log and
inverse-log transformations)
 Power-law (𝑛𝑡ℎ power and
𝑛𝑡ℎ root transformations)

7
 Image Negatives
 The negative of an image with intensity levels in the range [0,L-1]
can be described by:
𝑠 = 𝐿 −1− 𝑟

 This type of processing is practically suited for enhancing white or

gray detail embedded in a dark regions of an image, especially
when the black areas are dominate.

8
 Log Transformations
 General form: 𝑠 = 𝑐 log(1 + 𝑟)
Where c is a constant and r ≥ 0.

 The Log transformation maps a narrow range of low intensity

values in input to a wider range of output levels. The opposite is
true for high intensity input values.

 We use transformation of this type to expand the values of dark

pixels in an image while compressing the higher level values. The
opposite is true of the inverse log transformation.

 The log function has the important characteristic that it

compresses the dynamic range of images with large variations in
pixel values.

9
 Log Transformations
 A plot (image) of the Fourier Spectrum usually has a fairly large
dynamic range.

 The image can be enhanced by applying the log transformation.

10
 Power-Law (Gamma) Transformations
 Basic form: 𝑠 = 𝑐𝑟 𝛾
Where c and γ are positive constants.

 There exists a family of possible

transformation curves by varying 𝛄.

 Power-law curves with fractional

values of γ map a narrow range
of dark input values to a wider
range of output values. The
opposite is true for higher values
of input levels.

11
 Gamma Correction
 Many devices used for image capture, display and printing
respond according to a power law.

 The exponent in the power-law equation is referred to as gamma.

 The process of correcting for the power-law response is referred

to as gamma correction.

 Example:
 CRT devices have an intensity-to-voltage response that is a power
function (exponents typically range from 1.8-2.5)
 Gamma correction in this case could be achieved by applying the
transformation: s = r1/2.5 = r 0.4 .

12
Gamma Correction Example

13
 Power-Law transformation (MRI Example)
 The Figure shows magnetic resonance image (MRI) of an upper
thoracic human spine with a fracture dislocation and spinal cord
impingement.

 Because the image is predominantly

dark, an expansion of intensity levels
is desirable.

14
Power-Law transformation (MRI Example)

s = r 0.6

1
Transformed Intensities

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Old Intensities

15
Power-Law transformation (MRI Example)

s = r 0.4

1
0.9
Transformed Intensities

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Original Intensities

16
Power-Law transformation (MRI Example)

s = r 0.3

1
0.9
Transformed Intensities

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Original Intensities

17
 Power-Law transformation (MRI Example)
 as gamma decreased from 0.6 to 0.4,
more detail became visible.
 Further decrease of gamma to 0.3
enhanced a little more detail in
the background, but began to reduce
original 𝜸=0.6
contrast to the point where the image
started to have a very slight
“washed-out” appearance, especially
in the background.
 The best enhancement in terms
of contrast and discernible detail
𝜸=0.4 𝜸=0.3
was obtained with 𝛾=0.4.
18
 Power-Law transformation (Aerial Example)
 The Aerial image has washed-out appearance, so a compression of
intensity levels is desirable.

19
 Power-Law transformation (Aerial Example)
 Suitable results were obtained 𝜸=3.0

with gamma values of 3.0 and

4.0, the latter having a slightly
more appealing appearance
because it has higher contrast.

 The result obtained with 𝜸=4.0 𝜸=5.0

γ=0.5 has areas that are too
dark, in which some detail is
lost.

20
 Piecewise-Linear Transformation Functions
 We use user-defined transforms, defined by a set of lines “pieced”
together….piecewise linear.

 Piecewise functions can be arbitrarily complex.

 A disadvantage is that their specification requires significant user

input.

 Example functions:
 Contrast stretching

 Intensity-level slicing

 Bit-plane slicing

21
 Contrast stretching
 Contrast stretching
expands the range of
intensity levels in an image
so it spans a given (full)
intensity range.

 Control points (r1 , s1 ) and

(r2 , s2 ) control the shape
of the transform T(r).

 when r1 = r2 , s1 = 0 and
s2 = L − 1, this yields
a thresholding function.

22
 Intensity-level slicing
 Intensity-level slicing used to highlight a specific range of
intensities in an image that might be of interest.

 Two common approaches

 Set all pixel values within a range of

interest to one value (white) and all

others to another value (black). This
transformation Produces a binary image.
 Brighten (or darken) pixel values in

a range of interest and leave all others

unchanged.

23
Intensity-level slicing Example

24
 Bit-plane slicing
 Pixels are digital values composed of bits. For example, a pixel in a
256-level gray-scale image is comprised of 8 bits.

 We can highlight the contribution made to total image appearance

by specific bits. For example, we can display an image that only
shows the contribution of a specific bit plane.

 8-bit image may be considered as being composed of eight 1-bit

planes, with plane 1 containing the lowest-order bit of all pixels in
the image and plane 8 all the highest-order bits.

 Higher-order bits usually contain most of the significant visual

information, where lower-order bits contain subtle details.

25
Bit-plane slicing

26
 Bit-plane slicing Example
[00000001] [00000010]

[00000100] [00001000] [00010000]

[00100000] [01000000] [10000000]

27
 Bit-plane slicing Example
 Bit-plane slicing is useful in
 Determining the number of bits necessary to quantize an image
 Image compression

 The figure shows Images reconstructed using (a) bit planes 8 and
7; (b) bit planes 8, 7 and 6; and (c) bit planes 8, 7, 6, and 5.

(a) (b) (c)

original

28
 Histogram Processing
 The histogram of a digital image, f, (with intensities [0, 𝐿 − 1]) is
a discrete function: ℎ(𝑟𝑘 ) = 𝑛𝑘
Where 𝐫𝐤 is the k th intensity value and 𝐧𝐤 is the number of pixels
in 𝐟 with intensity 𝐫𝐤 .

 Normalizing the histogram is common practice:

 Divide the components by the total number of pixels in the image.
 Assuming an MxN image, this yields

𝑝(𝑟𝑘 ) = 𝑛𝑘 /𝑀𝑁 for 𝑘 = 0,1,2, … . , 𝐿 − 1

 𝐩(𝐫𝐤 ) is, basically, an estimate of the probability of occurrence of intensity
level 𝐫𝐤 in an image.
 𝜮 𝒑(𝒓𝒌 ) = 𝟏

29
 Histogram Processing
 Histogram manipulation can be used for
 Image enhancements

 Image statistics

 Image compression

 Image segmentation

30
 Histogram Examples
 Four basic image types and their
corresponding histograms
 Dark
 Light
 Low contrast
 High contrast

 Histograms commonly viewed

in plots as
 h(rk ) = nk versus rk
 p(rk ) = nk /MN versus rk

31
 Histogram Equalization
 we want to be able to learn how to go from a non-uniform
distribution to a uniform distribution.

32
 Histogram Equalization
 In order to achieve that, what we needs to find a transform that
takes basically the pixel values of the original image and
transforms them to pixel values of a different image.

33
 Histogram Equalization
 Intensity levels in an image may be viewed as random variables in
the interval [0, 𝐿 − 1].

 Fundamental descriptor of a random variable is its probability

density function (PDF).

 Let pr (r) and ps (s) denote the PDFs of 𝑟 and 𝑠 respectively:

34
 Histogram Equalization
 For discrete values, we deal with probabilities (histogram values),
instead of probability density functions and integrals.

 The probability of occurrence of intensity level 𝑟𝑘 in a digital

image is approximated by: p(rk ) = nk /MN for k = 0,1,2, … . , L − 1.

 the discrete form of the transformation function is:

35
 Histogram Equalization Example
 Suppose we have Intensity distribution and histogram values for a
3-bit, 64x64 digital image.

36
 Histogram Equalization Example
 The values of histogram equalization transformation function are
obtained using the equation:

 So:
𝑠0 = 1.33 𝑠1 = 3.08
𝑠2 = 4.55 𝑠3 = 5.67
𝑠4 = 6.23 𝑠5 = 6.65
𝑠6 = 6.86 𝑠7 = 7.00

37
 Histogram Equalization Example
 At this point, the s values still have fractions because they were
generated by summing probabilities values, so we round them to
the nearest integer:

𝑠0 = 1.33 1 𝑠1 = 3.08 3
𝑠2 = 4.55 5 𝑠3 = 5.67 6
𝑠4 = 6.23 6 𝑠5 = 6.65 7
𝑠6 = 6.86 7 𝑠7 = 7.00 7

 These are the values of equalized histogram.

38
 Histogram Equalization Example

Original Histogram Transformation Function Equalized Histogram

39
Contrast enhancement using Histogram Equalization

Transformation functions

40