Module 4

An image degradation model is a mathematical representation of the process by which an

image is degraded. The model typically includes a degradation function and a noise model.
The degradation function describes how the image is transformed, such as by blurring,
downsampling, or compression. The noise model describes the type of noise that is added to
the image, such as Gaussian noise, salt-and-pepper noise, or speckle noise.
There are many different image degradation models that have been proposed. Some of the
most common models include:
The Wiener model: This model assumes that the degradation function is linear and that the
noise is additive and white.
The Lucy-Richardson model: This model is a nonlinear iterative method for image
restoration.
The Total Variation model: This model is a non-blind image restoration model that minimizes
the total variation of the restored image.
The Wavelet model: This model uses wavelets to represent the image and the noise.
The choice of image degradation model depends on the specific application. For example,
the Wiener model is often used for image deblurring, while the Lucy-Richardson model is
often used for image denoising.

TRACE KTU
Noise models are used to describe the random fluctuations that occur in images. There are
many different types of noise, but some of the most common types include:
Gaussian noise: This is the most common type of noise. It is characterized by a bell-shaped
probability distribution.
Salt-and-pepper noise: This type of noise is characterized by the presence of isolated white
and black pixels.
Speckle noise: This type of noise is characterized by the presence of randomly distributed
bright and dark spots.

The choice of noise model depends on the specific application. For example, Gaussian noise
is often used for image denoising, while salt-and-pepper noise is often used for image
deblurring.
Image degradation models and noise models are essential tools for image restoration. By
understanding how images are degraded and how noise affects images, we can develop
more effective methods for restoring degraded images.
Mean filters, order statistic filters, and adaptive filters are all types of filters that can be used
to process images.
Mean filters are the simplest type of filter. They work by averaging the values of the pixels in
a local neighborhood. This can be used to smooth out noise and reduce the amount of detail
in an image.
Order statistic filters are more sophisticated than mean filters. They work by ordering the
values of the pixels in a local neighborhood and then selecting the median, minimum, or
maximum value. This can be used to remove noise and preserve edges in an image.
Adaptive filters are the most complex type of filter. They work by adjusting their coefficients
to the local characteristics of an image. This can be used to improve the performance of a
filter in specific applications, such as image denoising or edge detection.
The choice of filter depends on the specific application. For example, mean filters are often
used for smoothing images, while order statistic filters are often used for denoising images.
Adaptive filters can be used for a variety of applications, but they are often more
computationally expensive than other types of filters.

TRACE KTU
Edge detection is a technique used in image processing to identify the boundaries between
regions of an image with different brightness or color. Edges are important features in
images, as they can be used to represent objects, shapes, and textures.
There are many different edge detection techniques, but they all work by identifying pixels
that have a sharp change in brightness or color. One common approach is to use gradient
operators. Gradient operators are filters that calculate the first derivative of an image. The
first derivative of an image measures the rate of change of brightness or color at each pixel.
The Laplace operator is a second-order differential operator that can be used to detect
edges in images. The Laplace operator measures the overall smoothness of an image.
Regions of an image with sharp edges will have a high Laplace value, while regions of an
image with smooth gradients will have a low Laplace value.
Zero crossings are points in an image where the value of the Laplace operator is zero. Zero
crossings often correspond to edges in images, as they are the points where the gradient of
the image changes sign.
The choice of edge detection technique depends on the specific application. For example,
gradient operators are often used for edge detection in low-noise images, while the Laplace
operator is often used for edge detection in high-noise images.
Here are some of the most common edge detection techniques:
Sobel operator: The Sobel operator is a gradient operator that uses a 3x3 mask to calculate
the first derivative of an image.
Prewitt operator: The Prewitt operator is another gradient operator that uses a 3x3 mask to
calculate the first derivative of an image.
Roberts operator: The Roberts operator is a gradient operator that uses a 2x2 mask to
calculate the first derivative of an image.
Laplace of Gaussian (LoG): The LoG operator is a second-order differential operator that uses
a Gaussian kernel to calculate the Laplace of an image.
Canny edge detector: The Canny edge detector is a popular edge detection algorithm that
uses a multi-stage process to detect edges in images.
Thresholding is a technique used in image processing to convert a grayscale image into a
binary image. In a binary image, each pixel is either black or white. Thresholding is often
used to segment objects in images, to remove noise, or to improve the contrast of an image.
There are two main types of thresholding: global thresholding and local thresholding.

TRACE KTU
Global thresholding uses a single threshold value for the entire image. This is the simplest
type of thresholding, but it can be suboptimal if the image has a bimodal histogram,
meaning that the image has two distinct populations of pixels with different brightness
values.
Local thresholding uses different threshold values for different parts of the image. This is
more complex than global thresholding, but it can produce better results for images with
bimodal histograms.
Otsu's method is a popular algorithm for global thresholding. It works by finding the
threshold value that maximizes the between-class variance. The between-class variance is a
measure of how well the pixels in the image are separated into two classes, the foreground
and the background.
Here are the steps involved in Otsu's method:
Calculate the histogram of the image.

Find the mean of the image.

Calculate the between-class variance for all possible threshold values.
Choose the threshold value that maximizes the between-class variance.
Use the threshold value to convert the image into a binary image.
Otsu's method is a simple and effective algorithm for global thresholding. It is often used in
image processing applications, such as object segmentation and image denoising.
Multiple thresholds, variable thresholding, and multivariable thresholding are all image
thresholding techniques that use multiple thresholds to segment an image into multiple
regions.
Multiple thresholds is a general term for any thresholding technique that uses more than
one threshold. This can be done manually by setting multiple thresholds, or automatically by
using a thresholding algorithm that calculates the optimal thresholds for the image.
Variable thresholding is a type of multiple thresholds where the thresholds are allowed to
vary across the image. This can be done by using a different threshold for each pixel, or by
using a smoothly varying threshold across the image.
Multivariable thresholding is a type of multiple thresholds where the thresholds are allowed
to vary based on multiple image properties, such as the brightness, contrast, and texture of
the image. This can be done by using a different threshold for each property, or by using a
combination of properties to calculate the optimal thresholds.

The choice of which thresholding technique to use depends on the specific application.
Multiple thresholds can be used to segment images with complex backgrounds, variable
thresholding can be used to segment images with smooth gradients, and multivariable
thresholding can be used to segment images with multiple properties.

TRACE KTU
Here are some examples of how multiple thresholds, variable thresholding, and
multivariable thresholding can be used:

Multiple thresholds can be used to segment a medical image into different organs. For
example, a doctor might use multiple thresholds to segment a brain MRI image into the
brain, the skull, and the cerebrospinal fluid.
Variable thresholding can be used to segment an image of a landscape into different regions,
such as the sky, the ground, and the trees. For example, a photographer might use variable
thresholding to segment an image of a forest into the foreground, the middle ground, and
the background.
Multivariable thresholding can be used to segment an image of a product into different
parts, such as the packaging, the product itself, and the background. For example, a quality
control engineer might use multivariable thresholding to segment an image of a
manufactured product to identify any defects.
Thresholding is a powerful tool for image segmentation. By using multiple thresholds,
variable thresholds, or multivariable thresholds, you can segment images with complex
backgrounds, smooth gradients, or multiple properties.
Region-based segmentation is an image segmentation technique that divides an image into
regions based on the similarity of their pixel properties. This can be done using a variety of
methods, such as region growing, region splitting, and region merging.
Region growing starts with a seed pixel and grows a region by including neighboring pixels
that are similar to the seed pixel. This process is repeated until no more similar pixels can be
found.
Region splitting starts with a region and splits it into two or more regions based on the
dissimilarity of their pixel properties. This process is repeated until all regions are
homogeneous.
Region merging starts with a set of regions and merges them into larger regions based on
the similarity of their pixel properties. This process is repeated until there is only one region
remaining.
The choice of which region-based segmentation method to use depends on the specific
application. Region growing is a good choice for images with smooth gradients, region
splitting is a good choice for images with sharp edges, and region merging is a good choice
for images with multiple objects.

Here are some examples of how region-based segmentation can be used:

Region growing can be used to segment an image of a landscape into different regions, such
as the sky, the ground, and the trees.

TRACE KTU
Region splitting can be used to segment an image of a face into different regions, such as the
eyes, the nose, and the mouth.
Region merging can be used to segment an image of a document into different regions, such
as the header, the body, and the footer.
Region-based segmentation is a powerful tool for image segmentation. By using region
growing, region splitting, or region merging, you can segment images with smooth gradients,
sharp edges, or multiple objects.
Here are some of the advantages of region-based segmentation:
It is a flexible technique that can be used to segment images with a variety of properties.
It is relatively easy to implement.
It can be used to segment images with complex backgrounds.

Here are some of the disadvantages of region-based segmentation:

It can be time-consuming to segment large images.
It can be sensitive to noise.
It can produce over-segmentation, where the image is divided into too many regions.
Overall, region-based segmentation is a powerful and versatile image segmentation
technique. It can be used to segment images with a variety of properties, but it is important
to be aware of its limitations.

TRACE KTU