Digital Image Processing

Image Restoration
Noise models and additive noise removal
Image Restoration
 Image Restoration
 What is noise (in the context of image processing) and how can
it be modeled?
 What are the main types of noise that may affect an image?
 What are the possible solutions?

 Subjective Vs Objective (Enhancement Vs

Restoration)
Degradation Model for a Digital Image
Noise Models
 Noise and Noise Models
 Gaussian (normal)
 Impulse (salt-and-pepper)
 Uniform
 Rayleigh
 Gamma (Erlang)
 Exponential
 Effect of Noise on Images & Histograms

 Gaussian

 Exponential

 Impulse
(salt-and-pepper)
 Effect of Noise on Images & Histograms

 Rayleigh

 Gamma (Erlang)

 Uniform
Noise Models: Gaussian Noise
Noise Models: Rayleigh Noise
Noise Models: Erlang (Gamma) Noise
 Noise Models: Exponential Noise

Where
 Noise Models: Uniform Noise

 1

 , if a  z  b
p(z)   b  a
 otherwise
0


The mean and variance are

given by

ab (b  a) 2
 2 ,  2  12
Noise Models: Impulse (Salt and Pepper) Noise
Effect of Noise on Images & Histograms
Effect of Noise on Images & Histograms
Effect of Noise on Images & Histograms
 Periodic Noise (Example)
 Spatially Dependent Case
Applicability of various noise models
Estimation of noise parameters
Estimation of noise parameters (example)
Estimation of noise parameters (example)
Estimation of noise parameters
 Restoration of noise-only degradation

Filters to be considered
 Mean Filters: Arithmetic mean filter

Causes a certain amount of blurring (proportional to the window size)

to the image, thereby reducing the effects of noise.
Can be used to reduce noise of different types, but works best for Gaussian,
uniform, or Erlang noise.
 Mean Filters: Geometric mean filter

– A variation of the arithmetic mean filter

– Primarily used on images with Gaussian noise
– Retains image detail better than the arithmetic mean
 Mean Filters: Harmonic mean filter

Harmonic mean filter

– Another variation of the arithmetic mean filter

– Useful for images with Gaussian or salt noise
– Black pixels (pepper noise) are not filtered
Arithmetic and geometric mean filters (example)
Mean Filters: Harmonic mean filter
Mean Filters: Harmonic mean filter
Mean Filters: Contra-harmonic mean filter
Classification of contra-harmonic filter applications
Contra-harmonic mean filter (example)
Contra-harmonic mean filter (example)
 Rank / Order / Order Statistics Filters
– Known as Rank filters, Order filters OR Order Statistics filters

– Operate on a neighborhood around a reference pixel by

ordering (ranking) the pixel values and then performing an
operation on those ordered values to obtain the new value
for the reference pixel

– They perform very well in the presence of salt and pepper

noise but are more computationally expensive as compared to
mean filters
Rank / Order Statistics Filters: Median filter
 Rank / Order Statistics Filters: Median filter

– Most popular and useful of the rank filters.

– It works by selecting the middle pixel value from the ordered set
of values within the m × n neighborhood (W) around the
reference pixel.
• If mn is an even number, the arithmetic average of the two
values closest to the middle of the ordered set is used
instead.

– Many variants, extensions, and optimized implementations

in the literature.
Median filter (Example)
Rank / Order Statistics Filters: Max and Min filter
 Rank / Order Statistics Filters: Max and Min filter

– Max filter also known as 100th percentile filter

– Min filter also known as zeroth percentile filter
– Max filter helps in removing pepper noise
– Min filter helps in removing salt noise
Max and Min filter (Example)
 Rank / Order Statistics Filters: Midpoint filter

– Calculates the average of the highest and lowest pixel

values within a window
– What would it do with salt and pepper noise ?
Midpoint filter (Example)
 5.3.3 Adaptive Filters

• The behavior of adaptive filters changes based

on statistical characteristics (var. mean) of the
image inside the filter region Sxy, m x n local
window region

- The mean gives a measure for average intensity in the

region
- The variance characterizes contrast in that region,

 they both are reasonable parameters to base an

adaptive filter.
45
a) Adaptive local noise reduction filter:

n/ L

n = variance of noise
= local variance of the pixels in Sxy
mL = local mean of the pixels in Sxy

46
5.3.3 Adaptive Filters

47
 5.3.3 Adaptive Filters

b) Adaptive median filter: can handle impulse

noise with larger probabilities than traditional
median filter. It operates on a rectangular
region whose size is changing

• Adaptive median filter has 3 goals to remove

impulse noise, to provide smoothing of other
noise, and to reduce distortion

48
5.3.3 Adaptive Filters

49