Image Enhancement in

the Spatial Domain

Intensity Transformation

Some figures are provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Enhancement
• Objectives
– It makes an image more suitable for specific, problem-
oriented applications than the original.
– It is often for human perception; the viewer is the ultimate
judge of how well a particular works.

• Two broad categories of enhancement

methods
– Spatial domain (this topic)
• Works with pixels
– Frequency domain (will be covered in the other topic)
• Works with frequencies
 Spatial Domain

g(x,y) = T[ f(x,y) ]

• f(x,y) = input image

• g(x,y) = output, processed, transformed image
• T is an operator on f
– T = intensity transformation function
– T is defined over some neighborhood of (x, y), e.g.,
square or rectangular sub-image area or a point of 1x1
size
– T can be operated on a single image or a set of images.
 g(x,y) = T[ f(x,y) ]

Image g ( x,y )

Input Image Output Image

T = intensity transformation function

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 f1(x,y)
g(x,y) = T[ f1(x,y), f2(x,y), …, fn(x,y)]

f2(x,y)

fn(x,y)
Image g ( x,y )

T = intensity transformation function

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Gray-level transformation
• Simplest form of T
– neighborhood is 1 x 1 (one pixel)

• Notation: s = T(r)
– r, s denote the gray levels of f(x,y) and g(x,y) for
any point (x,y), respectively
 Examples of T(r)
s s
T(r)
T(r)

m m
0 r 0 r
Darker Lighter Darker Lighter

It produces an image of lower T is a thresholding function

contrast than the original by
(1) darkening the intensity It produces a binary image with
levels above m, and an intensity threshold at m.
(2) brightening the leaves
below m in the original Low intensity levels below m
image. map to zero and high intensity
levels above m map to s.
 Basic Gray Level Transforms: Inverse
or Image Negatives

L-1

T(r)
s

0
r
L-1

s = T(r) = (L-1)-r

Negatives of the images above

 Brightening

Darkening

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Log Transformations
L-1 s = c log(1 + r)
T(r)
s (c is some scale-factor)
(intensity rescaling maybe needed for s
to fit s into the range [0, L-1])
The image becomes brighter
0 r L-1 and details are more visible.

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Power-Law (Gamma) Transformations
produce brighter image

s  cr
where c and  are γ decreases
positive constants.

Intensity rescaling
may be needed for
s to fit s into the
range [0, L-1].
γ increases
Also, known as
gamma (  )
produce darker image
correction.
Examples: http://www.cambridgeincolour.com/tutorials/photoshop-curves.htm
Power-Law (Gamma) Transformations

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
Original
image γ = 0.6

With the help of power-law

transformation, fractures
near the vertical centre of
the spine are now made
more visible in the magnetic
resonance image.
γ = 0.4 γ = 0.3
The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Original γ=3
image

γ=4 γ=5

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
Intensity normalization

r = input intensity s = output

gamma

photo s = r
s = T[r] s

r = [0 – 255] 1000 Linear mapping s = [0 - 1000]

8 bits 1000-s Max value of s = 1000
255
s 1000-0 Min value of s = 0
output a ?Final

0 0

known unknown (Final)

Ratio 8 bits

(Full range) (Full range)

 Contrast Stretching
A low-contrast image

Contrast stretching by
a piecewise linear
transformation
function

Output image

If r1  rmin s1  0
r2  rmax s2  L  1
then intensity levels are stretched to the
full intensity range [0, L – 1].

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Contrast Stretching
A low-contrast image

Output binary image

If r1  r2 s1  0 s2  L  1
then the image is thresholded and a
binary image is produced.

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
Windowing (or Gray Level Slicing)
L-1
Operating with a “window”
T(r) of gray levels ri to r(i+x)
s
This function highlights the
intensity levels between the
ri r(i+x) range
0 r L-1

L-1
L-1 L-1
T(r)
s
s T(r) s T(r)
ri r(i+x)
ri r(i+x)
ri r(i+x)
0 r L-1
0 r L-1 0 r L-1
 Properties of T(r)
•No-inverse due to
• From r to s, ambiguity
T(r) •Does not preserve
one-to-one mapping gray level ordering
• From s to r, •For non-inverse
one-to-many mapping transformation, it is
not a one-to-one
mapping and is
0 one-to-many mapping.

• One-to-one mapping T(r) •Inverse

• Increasing •Preserves gray level
s ordering (increasing)

0 r
 Bit-Plane Slicing
• Pixel values are integers composed of bits
• 8 bits give the values between 0 and 255
• Each bit contributes to the entire image appearance
• Each bit can be viewed as a one-bit image

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Bit-Plane Slicing
• Plane 8 1-bit image is the highest-order bit, most-significant bit
• Decomposing an image into bit planes helps analyzing the image properties

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Bit-Plane Slicing
• An image can be reconstructed by combining different
bit planes
• The nth bit plane times is added to the
reconstructed image to form the grey scale image
• If there are n bits, the constructed image has image
intensity levels.

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com