Image Enhancement in

the Spatial Domain

Histogram Processing

Some figures are provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Image Histogram
• Image Histogram
– digital image with gray levels [0, L-1]

– p(rk) = nk/N, probability of occurrence of gray level rk

– rk is the kth gray level

– nk = number of pixels with kth gray level
– N = total number of pixels
– k = 0, 1, 2, 3, 4, 5 . . ., L-1
 Image Histogram
• p(rk) is the probability of the occurrence of gray-
level rk
• is the probability density function (PDF) of
the variable ,

p(rk)

rk
 Image Histogram
• Histogram can tell you a lot about an image
• Gray level distribution
• It can be modeled by a statistical distribution (red line)

p(rk)= nk/N

rk
 Image Histogram
• Gray level distribution helps define intensity
threshold
T(r)

p(rk)= nk/N

rk
Threshold
 Image Histogram
• Histograms are not unique. Two images below
give the same image histogram.
• No spatial information is captured
 Image Histogram
• Image types

Bright Image Dark Image

P(rk) P(rk)

rk rk
 Image Histogram
• Image types

Low Contrast High Contrast

P(rk) P(rk)

rk rk
Some
Examples

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Contrast Stretching
Low Contrast
L-1 T(r)
P(rk)
s

r
0 r L-1

P(sk)

s
 Contrast Compressing
High Contrast
L-1
P(rk) T(r)
r s

0 r L-1

P(sk)

s
 Histogram Equalization
• We want an image with equally many pixels at every
gray level, or the output intensity approximately
follows a uniform distribution.

• That is, a flat histogram, where each gray level, rk,

appears (N/rm) times

– where “rm” is the maximum gray level

– N is number of pixels in the image

• Example
– https://demonstrations.wolfram.com/HistogramEqualization/
 Histogram Equalization
• The weighted cumulative distribution function (CDF) of r is
represented by (in continuous form), (T is the transformation
function)

• This transformation function seeks to generate an output

image with a uniform (flat) probability density function (PDF),
independently of the form of

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Use T(r) to equalize the histogram
• Assume the input variable r has been
normalized between [0,1]

• s = T(r), there are two properties

– (a) T(r) is single-valued in the interval 0 <= r <= 1

– (b) 0 <= T(r) <= 1 for 0 <= r <= 1

 Conditions (a) and (b)
• (a) T(r) is single-valued and non-decreasing in the interval
0 <= r <= 1
– Can preserve the order from black to white in the gray
scale.
– Can preserve the basic appearance of an image.

• (b) for 0 <= r <= 1, 0 <= T(r) <= 1

– Can guarantee a mapping that is consistent with the
allowed range of pixel values. No intensity rescaling is
needed.
 Discrete approximation of T
• Let’s look at the discrete approximation

sk is the output continuous intensity value, 0 <= sk <= L-1

rk is the input continuous intensity value, 0 <= rk <= L-1
nj is number of pixels with the jth gray level
k = 0, 1, 2, 3, . . L-1 (gray levels)
 Example

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 sk = T(rk) sk
T
Input image New image
rk

Total number of pixels = 64*64 = 4096 pixels

23-1=7
k 28-1=255
rk

Pr(rk)

Sk*(1/7)

Gray
levels

Stretching Compressing Compressing

 Example
http://www.mathworks.com/help/toolbox/images/ref/histeq.html

nk

Image rk
Histogram
Notice, this is not normalized, y axis is nk.
To normalize, let y axis = p(rk)= nk /N
 Example
sk
nk

rk rk
stretching compressing
Histogram
Notice, this is not normalized, y axis is nk.
To normalize, let y axis = p(rk)= nk /N
 Example
sk
nk nk

rk T(r) sk
The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Histogram Equalization
• Can significantly improve image appearance
– Automatic
– users doesn’t have to perform manual windowing or
manual gray level slicing

• Nice pre-processing step before image comparison

– Account for different lighting conditions
– Account for different camera/device properties
 Local Enhancement
• Histogram equalization is a global operation
– Each pixel is processed based on information of the entire
image
– Often enhances global details

• We would like to enhance details over small areas

– Each pixel is processed based on information of a small
area/sub-image
 Local Histogram Equalization

place result
(x,y) in (x,y)

Result
calculate histogram
using neighborhood of mxm
about (x,y)
 Local Histogram
• Apply histogram equalization about a
neighborhood around (x,y)

• Transform the gray level for pixel (x,y)

• Move the neighborhood over the rest of

the image
 Local Histogram
Reveals detail in local areas

Original Global Histogram Local Histogram

 Histogram Specification
• It is useful sometimes to be able to specify the shape of the
histogram that we wish the processed image to have

• Histogram specification is also known as histogram matching

• Let be the intensity of the input image

• Let be the PDF (probability density function) of the

input image
 Histogram Specification
• Let be the PDF (probability density function) of the
input image

Input image

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Histogram Specification
• The histogram equalized image of the input image

A lot of low intensity

pixels are mapped to the
high intensity values

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Histogram Specification
• Given the specified PDF , compute all values of the
transformation function

The figure is provided by Pearson Education, Digital Image Processing, Gonzalez & Woods, www.ImageProcessingPlace.com
 Histogram Specification
• For each value of (rounded to integer), find the
corresponding value of so that (rounded to integer)
is closest to
• Generate the histogram-
specified image by
mapping every equalized
pixel with value to the
corresponding pixel with
value

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