1. What is computer vision? What are the applications of computer vision?

Computer vision is a field of computer science that gives computers the ability to see,
understand, and interpret the world around them. It is a rapidly growing field with
applications in a wide range of industries, including:

Self-driving cars: Computer vision is essential for self-driving cars to navigate safely
and efficiently. Cameras mounted on the car's exterior are used to detect and track
objects in the environment, such as other cars, pedestrians, and traffic signs. This
information is then used by the car's computer to make decisions about how to drive.
Pedestrian detection: Computer vision is used to detect pedestrians in real time, which
is important for safety applications such as self-driving cars and traffic monitoring
systems. Pedestrian detection algorithms typically use a combination of techniques,
such as image segmentation and object tracking, to identify pedestrians in images or
videos.
Parking occupancy detection: Computer vision can be used to automatically detect
whether a parking space is occupied or not. This information can be used to help
drivers find parking spaces more easily, and it can also be used to optimize parking lot
usage.
Traffic flow analysis: Computer vision can be used to analyze traffic flow and identify
congestion hotspots. This information can be used to improve traffic management and
make roads safer for drivers and pedestrians.
Road condition monitoring: Computer vision can be used to monitor road conditions
for signs of wear and tear, such as potholes and cracks. This information can be used
to schedule maintenance and repairs before they become a safety hazard.
These are just a few of the many applications of computer vision. As the technology
continues to develop, we can expect to see even more innovative and creative
applications in the years to come.

Here are some other applications of computer vision:

Healthcare: Computer vision is used to diagnose diseases, detect cancer, and track the
progression of diseases. It is also used to develop new medical treatments and
improve the quality of patient care.
Manufacturing: Computer vision is used to inspect products for defects, automate
assembly lines, and optimize production processes. It is also used to develop new
products and improve the quality of manufacturing.
Agriculture: Computer vision is used to monitor crops for pests and diseases, optimize
irrigation, and automate harvesting. It is also used to develop new crop varieties and
improve the efficiency of agriculture.
Retail: Computer vision is used to track inventory, analyze customer behavior, and
personalize shopping experiences. It is also used to develop new retail products and
improve the efficiency of retail operations.
Security: Computer vision is used to monitor public spaces for suspicious activity,
detect crime, and identify criminals. It is also used to develop new security systems
and improve the safety of public places.
Computer vision is a powerful technology with the potential to revolutionize many
industries. As the technology continues to develop, we can expect to see even more
innovative and creative applications in the years to come.

2. Explain transformation and their types [Orthogonal, Euclidean, Affine, Projective]

with applications.

A transformation is a mathematical operation that changes the position, size, or shape

of an object. In computer graphics, transformations are used to manipulate objects in
2D or 3D space.
 There are four main types of transformations: orthogonal, Euclidean, affine, and
projective.

Orthogonal transformations preserve the length of any vector that is transformed. This
means that they do not change the shape of an object, only its position. Orthogonal
transformations are often used to translate, rotate, or scale objects.
Euclidean transformations are a special case of orthogonal transformations that also
preserve the angle between any two vectors. This means that they do not change the
shape or orientation of an object, only its position and scale. Euclidean
transformations are often used to perform simple geometric operations, such as
translation, rotation, and scaling.
Affine transformations do not preserve the length or angle between vectors. However,
they do preserve the parallel relationship between vectors. This means that they can
be used to shear, stretch, or skew objects. Affine transformations are often used to
perform more complex geometric operations, such as perspective projection.
Projective transformations do not preserve any of the properties of vectors. However,
they do preserve the cross product of vectors. This means that they can be used to
map objects from one coordinate system to another. Projective transformations are
often used to render 3D objects onto a 2D screen.
Here are some examples of applications of transformations in computer graphics:

Translation: Translation is a simple transformation that moves an object by a certain

amount in a particular direction. It is often used to move objects around the screen or
to reposition them within a scene.
Rotation: Rotation is a transformation that rotates an object around a particular point.
It is often used to change the orientation of an object or to create the illusion of
movement.
Scaling: Scaling is a transformation that changes the size of an object. It is often used
to make objects larger or smaller, or to create the illusion of perspective.
Shear: Shear is a transformation that skews an object in a particular direction. It is
often used to create the illusion of perspective or to distort an object.
Perspective projection: Perspective projection is a transformation that maps 3D
objects onto a 2D screen. It is often used to create the illusion of depth in a scene.
Transformations are a powerful tool for manipulating objects in computer graphics.
They can be used to create a wide variety of effects, from simple translations and
rotations to complex perspective projections.

3. Explain convolution and filtering with example.

Convolution and filtering are two closely related concepts in signal processing and
image processing. They both involve the application of a kernel to an input signal or image to
produce a new output signal or image.

In convolution, the kernel is a small matrix of numbers that is used to weight the input signal
or image. The kernel is slid over the input signal or image, and the output signal or image is
calculated by multiplying the kernel elements with the input signal or image elements at each
location and then summing the products.

For example, let's say we have an input image of a cat and we want to apply a kernel to detect
edges in the image. The kernel might look like this:

[[-1, 0, 1],
[-2, 0, 2],
[-1, 0, 1]]
This kernel is designed to detect edges by looking for changes in brightness. The negative
values in the kernel will help to highlight edges, while the positive values will help to smooth
out the image.

To apply this kernel to the image, we would slide it over the image, starting at the top left
corner. At each location, we would multiply the kernel elements with the image elements at
that location and then sum the products. The output of this operation would be a new image
with enhanced edges.
Filtering is a more general term than convolution. It refers to any process that modifies an
input signal or image by applying a kernel. Convolution is a specific type of filtering, but
there are many other types of filters that can be used.

For example, we could use a filter to blur an image, sharpen an image, or change the contrast
of an image. The type of filter that we use will depend on the desired effect that we want to
achieve.

Convolution and filtering are powerful tools that can be used to manipulate images and
signals in a variety of ways. They are used in a wide variety of applications, including
computer vision, image processing, and signal processing.

Here are some other examples of convolution and filtering in image processing:

Edge detection: Edge detection is a technique that uses convolution to find the edges in an
image. This can be used to identify objects in an image or to extract features from an image.
Smoothing: Smoothing is a technique that uses convolution to reduce the noise in an image.
This can be used to make an image easier to see or to improve the performance of image
processing algorithms.
Sharpening: Sharpening is a technique that uses convolution to increase the contrast in an
image. This can be used to make an image look more realistic or to improve the performance
of image processing algorithms.
Blurring: Blurring is a technique that uses convolution to reduce the detail in an image. This
can be used to hide sensitive information or to make an image more aesthetically pleasing.
Convolution and filtering are powerful tools that can be used to manipulate images in a
variety of ways. They are used in a wide variety of applications, including computer vision,
image processing, and signal processing.
4. Explain the processing of image enhancement.
Image enhancement is the process of improving the quality and information content of an
image. It can be used to make an image look more visually appealing, to bring out specific
features, or to remove noise from an image.

There are two main categories of image enhancement techniques:

Spatial domain techniques: Spatial domain techniques manipulate the pixels of an image
directly. This process happens in the image's coordinate system, also known as the spatial
domain.
Frequency domain techniques: Frequency domain techniques transform an image from the
spatial domain to the frequency domain. In this process, Mathematical transformations (such
as the Fourier transform) are used. The image can be modified by manipulating its frequency
components.
Here are some common image enhancement techniques:

Brightness adjustment: This technique adjusts the brightness of an image by adding or

subtracting a constant value to each pixel.
Contrast adjustment: This technique adjusts the contrast of an image by stretching or
shrinking the range of gray levels.
Sharpening: This technique enhances the edges in an image by increasing the contrast
between adjacent pixels.
Noise removal: This technique removes noise from an image by averaging the values of
neighboring pixels.
Color correction: This technique adjusts the colors in an image to make them more accurate
or pleasing to the eye.
The processing of image enhancement can be divided into the following steps:

Preprocessing: The preprocessing step involves preparing the image for enhancement. This
may involve removing noise, correcting for uneven illumination, or converting the image to a
different color space.
Enhancement: The enhancement step is where the actual image enhancement is performed.
This may involve adjusting the brightness, contrast, sharpness, or colors of the image.
Postprocessing: The postprocessing step involves finalizing the enhanced image. This may
involve cropping the image, resizing the image, or adding a watermark.
The processing of image enhancement can be a complex process, and the specific steps
involved will vary depending on the specific image and the desired effect. However, the
general steps outlined above are common to most image enhancement techniques.

Here are some examples of image enhancement in practice:

Medical imaging: Image enhancement is often used in medical imaging to improve the
visibility of tumors, blood vessels, and other structures.
Astronomy: Image enhancement is used in astronomy to improve the visibility of faint
objects and to remove noise from images.
Remote sensing: Image enhancement is used in remote sensing to improve the visibility of
features on the Earth's surface, such as vegetation, buildings, and water bodies.
Security: Image enhancement is used in security applications to improve the visibility of
faces, license plates, and other objects of interest.
Photography: Image enhancement is used in photography to improve the quality of images
and to create special effects.
Image enhancement is a powerful tool that can be used to improve the quality and
information content of images. It is used in a wide variety of applications, from medical
imaging to security to photography.
5. Types of gray-level transformation used for image enhancement.
There are three main types of gray-level transformation used for image
enhancement:

Linear transformation: A linear transformation is a function that maps each pixel value in the
input image to a new pixel value in the output image. The transformation is linear in the
sense that it can be represented by a matrix.
Logarithmic transformation: A logarithmic transformation is a function that maps each pixel
value in the input image to a logarithmic value in the output image. This can be used to
improve the contrast of an image by stretching the range of gray levels.
Power-law transformation: A power-law transformation is a function that maps each pixel
value in the input image to a power of that value in the output image. This can be used to
adjust the contrast of an image and to enhance specific features.
Here is a more detailed explanation of each type of gray-level transformation:

Linear transformation: The most basic type of gray-level transformation is a linear

transformation. A linear transformation can be represented by a matrix, where each element
in the matrix represents the weight that is given to the corresponding pixel value in the input
image. The output pixel value is then calculated by multiplying the input pixel values by the
matrix elements and then adding them together.
For example, let's say we have an input image with 8 bits per pixel, which means that each
pixel can have a value from 0 to 255. We can use a linear transformation to map each pixel
value to a new pixel value in the range of 0 to 100. To do this, we would create a matrix with
8 rows and 8 columns. The first row of the matrix would contain all zeros, the second row
would contain all ones, and so on. The last row of the matrix would contain all 100s.

Now, we can multiply the input pixel values by the matrix elements and then add them
together to get the output pixel values. For example, if the input pixel value is 128, then the
output pixel value would be 64. This is because the first element in the first row of the matrix
is 0, the second element is 1, the third element is 2, and so on. When we multiply 128 by 0, 1,
2, and so on, we get 64.

Linear transformations are a simple but powerful tool for image enhancement. They can be
used to adjust the brightness, contrast, and sharpness of an image. They can also be used to
enhance specific features in an image.

Logarithmic transformation: A logarithmic transformation is a function that maps each pixel

value in the input image to a logarithmic value in the output image. This can be used to
improve the contrast of an image by stretching the range of gray levels.
The logarithmic transformation is given by the following formula:

s = log(r + 1)
where s is the output pixel value, r is the input pixel value, and log is the natural logarithm
function.
The logarithmic transformation has the effect of stretching the range of gray levels in the
output image. This can be useful for images that have a wide range of brightness values, such
as images of outdoor scenes.

Power-law transformation: A power-law transformation is a function that maps each pixel

value in the input image to a power of that value in the output image. This can be used to
adjust the contrast of an image and to enhance specific features.
The power-law transformation is given by the following formula:

s = cr^g
where s is the output pixel value, r is the input pixel value, c is a constant, and g is a power
factor.

The power-law transformation has the effect of stretching or shrinking the range of gray
levels in the output image, depending on the value of the power factor g. If g is greater than
1, then the range of gray levels is stretched. If g is less than 1, then the range of gray levels is
shrunk.

The power-law transformation can be used to enhance specific features in an image by setting
the power factor g to a value that is appropriate for the feature. For example, to enhance
edges in an image, we would set g to a value greater than 1.

The three types of gray-level transformation described above are just a few of the many that
can be used for image enhancement. The choice of transformation will depend on the specific
image and the desired effect.
6. Suitable technique for enhancing low contrast images
There are several suitable techniques for enhancing low contrast images. Some of the most
common techniques include:

Histogram equalization: Histogram equalization is a simple but effective technique for

enhancing low contrast images. It works by redistributing the gray levels in an image so that
they are more evenly distributed. This has the effect of stretching the range of gray levels in
the image, which can improve the contrast.
Contrast limited adaptive histogram equalization (CLAHE): CLAHE is a more advanced
version of histogram equalization that is designed to avoid over-enhancement of the image. It
works by dividing the image into small blocks and then applying histogram equalization to
each block individually. This helps to ensure that the contrast is enhanced without
introducing artifacts.
Local adaptive thresholding: Local adaptive thresholding is a technique for thresholding
images that is based on the local image statistics. It works by calculating a threshold for each
pixel in the image based on the gray levels of the neighboring pixels. This helps to ensure
that the threshold is appropriate for the local image content, which can improve the contrast
of the image.
Image sharpening: Image sharpening is a technique for enhancing the edges in an image. It
works by increasing the contrast between adjacent pixels. This can help to make the image
look more visually appealing and to improve the visibility of edges.
The choice of technique for enhancing a low contrast image will depend on the specific
image and the desired effect. However, histogram equalization, CLAHE, and local adaptive
thresholding are all good techniques to consider.

Here is a more detailed explanation of each technique:

Histogram equalization: Histogram equalization is a technique that redistributes the gray

levels in an image so that they are more evenly distributed. This has the effect of stretching
the range of gray levels in the image, which can improve the contrast.
The histogram of an image is a plot of the number of pixels in each gray level. A low contrast
image will have a histogram that is skewed towards the center, with most of the pixels
concentrated in the middle gray levels. Histogram equalization works by remapping the gray
levels so that the histogram is more evenly distributed. This has the effect of stretching the
range of gray levels, which can improve the contrast of the image.

Contrast limited adaptive histogram equalization (CLAHE): CLAHE is a more advanced

version of histogram equalization that is designed to avoid over-enhancement of the image. It
works by dividing the image into small blocks and then applying histogram equalization to
each block individually. This helps to ensure that the contrast is enhanced without
introducing artifacts.
CLAHE works by calculating a threshold for each block in the image based on the gray
levels of the neighboring pixels. This threshold is then used to equalize the gray levels in the
block. The threshold is calculated so that the contrast in the block is enhanced without
introducing artifacts.

Local adaptive thresholding: Local adaptive thresholding is a technique for thresholding

images that is based on the local image statistics. It works by calculating a threshold for each
pixel in the image based on the gray levels of the neighboring pixels. This helps to ensure
that the threshold is appropriate for the local image content, which can improve the contrast
of the image.
Local adaptive thresholding works by calculating a threshold for each pixel in the image
based on the gray levels of the neighboring pixels. The threshold is calculated so that the
pixels with gray levels above the threshold are set to 1 and the pixels with gray levels below
the threshold are set to 0. This helps to ensure that the threshold is appropriate for the local
image content, which can improve the contrast of the image.

Image sharpening: Image sharpening is a technique for enhancing the edges in an image. It
works by increasing the contrast between adjacent pixels. This can help to make the image
look more visually appealing and to improve the visibility of edges.
Image sharpening works by convolving the image with a filter that enhances the edges. The
filter is designed to amplify the differences between adjacent pixels, which has the effect of
sharpening the edges in the image.

The choice of technique for enhancing a low contrast image will depend on the specific
image and the desired effect. However, histogram equalization, CLAHE, and local adaptive
thresholding are all good techniques to consider.
7. Stereo Vision method of Depth Estimation.
Stereo vision is a technique for estimating depth by using two or more images of the same
scene taken from different viewpoints. The principle behind stereo vision is that the same
object will appear to be at different distances in the two images, depending on its position in
the scene. This difference in distance is known as the disparity.

To estimate depth using stereo vision, we need to first find the disparity between
corresponding points in the two images. This can be done using a variety of techniques, such
as:

Matching: This is the most common technique for finding disparity. It works by finding the
best match for each pixel in one image to a pixel in the other image. The best match is
typically the pixel that has the most similar gray level.
Triangulation: This technique uses the disparity between two points to estimate their distance
from the camera. This is done by using the law of sines.
Structure from motion: This technique uses the disparity between multiple points to estimate
the 3D structure of the scene. This is done by using a technique called bundle adjustment.
Once we have estimated the disparity, we can use it to estimate the depth of the object. The
depth is simply the distance between the camera and the object, multiplied by the disparity.

Stereo vision is a powerful technique for estimating depth. It is used in a variety of

applications, such as:

Self-driving cars: Stereo vision is used in self-driving cars to estimate the distance to other
cars, pedestrians, and obstacles. This information is used to navigate the car safely.
Robotics: Stereo vision is used in robotics to estimate the distance to objects in the
environment. This information is used to plan the robot's movements.
Virtual reality: Stereo vision is used in virtual reality to create the illusion of depth. This is
done by displaying different images to each eye, which creates the illusion that the objects in
the scene are three-dimensional.
Stereo vision is a mature technology that is used in a variety of applications. It is a powerful
tool for estimating depth and creating the illusion of three-dimensionality.
8. Camera Geometry.
Camera geometry is the study of how cameras work and how they relate to the real
world. It is a fundamental concept in computer vision and image processing.

The basic idea of camera geometry is that a camera projects a 3D scene onto a 2D
image plane. This projection is not perfect, and there are a number of factors that
can distort the image, such as the lens distortion, the sensor size, and the camera's
position and orientation.

Despite these distortions, camera geometry can be used to recover information

about the 3D scene from the 2D image. This can be done using a variety of
techniques, such as:

• Triangulation:
Triangulation is a technique for determining the 3D coordinates of
a point from its projections in two or more images.
• Structure
from motion: Structure from motion is a technique for determining the
3D structure of a scene from a sequence of images.
• Camera calibration: Camera calibration is a technique for determining the
intrinsic and extrinsic parameters of a camera.

Camera geometry is a complex and challenging field, but it is also a very powerful
tool. It is used in a variety of applications, such as:

• Self-driving
cars: Self-driving cars use camera geometry to estimate the
distance to other cars, pedestrians, and obstacles. This information is used to
navigate the car safely.
• Robotics:Robotics use camera geometry to estimate the distance to objects in
the environment. This information is used to plan the robot's movements.
• Virtualreality: Virtual reality uses camera geometry to create the illusion of
depth. This is done by displaying different images to each eye, which creates
the illusion that the objects in the scene are three-dimensional.

Camera geometry is a mature technology that is used in a variety of applications. It

is a powerful tool for understanding how cameras work and for recovering
information about the 3D world from 2D images.

Here are some of the key concepts in camera geometry:

• Pinholecamera model: The pinhole camera model is a simplified model of a

camera that is used to understand the basic principles of camera geometry. In
the pinhole camera model, the camera is a point that projects rays onto a flat
image plane.
• Principal
point: The principal point is the point on the image plane that
corresponds to the center of the camera.
• Focal length: The focal length is the distance between the principal point and
the image plane.
• Image coordinate system: The image coordinate system is a coordinate system
that is used to represent points on the image plane. The origin of the image
coordinate system is the principal point, and the x-axis is horizontal and the y-
axis is vertical.
• Distortion:
Distortion is a phenomenon that occurs when a camera does not
project rays onto the image plane perfectly. There are a number of different
types of distortion, such as radial distortion and tangential distortion.
• Camera calibration: Camera calibration is the process of determining the
intrinsic and extrinsic parameters of a camera. The intrinsic parameters are
the parameters of the camera that do not change with the camera's position
and orientation, such as the focal length and the principal point. The extrinsic
parameters are the parameters of the camera that do change with the
camera's position and orientation, such as the camera's position and
orientation in the world.

Camera geometry is a complex and challenging field, but it is also a very powerful
tool. It is used in a variety of applications, such as self-driving cars, robotics, and
virtual reality.

9. Write the theory of RANSAC with the example.

RANSAC stands for RANdom SAmple Consensus. It is an iterative algorithm for finding the
best fit model to a set of data points. The algorithm works by repeatedly sampling a random
subset of the data points and fitting a model to the subset. The model that best fits the random
subset is then used to fit the entire data set.

The RANSAC algorithm is robust to outliers, which are data points that do not fit the model
well. This is because the algorithm only uses the data points that fit the model well to fit the
entire data set.

Here is an example of how the RANSAC algorithm can be used to find the best fit line to a
set of data points:

Initialize: The algorithm starts by randomly sampling two data points from the set.
Fit a line: The algorithm fits a line to the two data points.
Count inliers: The algorithm counts the number of data points that are within a certain
distance of the line.
Repeat: If the number of inliers is greater than a certain threshold, then the algorithm stops.
Otherwise, the algorithm repeats steps 2-4.
Return the best model: The algorithm returns the model that had the most inliers.
In this example, the line that had the most inliers is the best fit line to the data set. The
RANSAC algorithm is a powerful tool for finding the best fit model to a set of data points. It
is robust to outliers and can be used to fit a variety of models, such as lines, planes, and
curves.

Here are some of the advantages of RANSAC:

Robust to outliers: RANSAC is robust to outliers, which are data points that do not fit the
model well. This is because the algorithm only uses the data points that fit the model well to
fit the entire data set.
Can fit a variety of models: RANSAC can be used to fit a variety of models, such as lines,
planes, and curves.
Simple to implement: RANSAC is relatively simple to implement.
Here are some of the disadvantages of RANSAC:

Slow: RANSAC can be slow, especially for large data sets.

Not always accurate: RANSAC may not always find the best fit model, especially if the data
set contains a lot of outliers.
Sensitive to the parameters: The performance of RANSAC can be sensitive to the parameters
that are used, such as the number of data points to be sampled and the distance threshold.
Overall, RANSAC is a powerful tool for finding the best fit model to a set of data points. It is
robust to outliers and can be used to fit a variety of models. However, RANSAC can be slow
for large data sets and may not always find the best fit model.
10. Explain the functioning of a digital camera with the help of pipeline diagram.
A digital camera is a device that captures images and stores them in a digital format. It works
by taking in light through a lens and projecting it onto a sensor. The sensor converts the light
into electrical signals, which are then processed by the camera's processor and stored in
memory.

The following is a pipeline diagram of a digital camera:

1. Light enters the camera through the lens.

2. The lens focuses the light onto the sensor.
3. The sensor converts the light into electrical signals.
4. The processor amplifies and digitizes the signals.
5. The processor applies image processing algorithms to the signals.
6. The processor stores the processed signals in memory.
7. The processor outputs the image to a display or other device.
Let's take a closer look at each step in the pipeline:

Light enters the camera through the lens. The lens is responsible for focusing the light onto
the sensor. The type of lens used in a camera will affect the quality of the image. For
example, a wide-angle lens will capture a wider field of view, while a telephoto lens will
magnify objects that are far away.
The lens focuses the light onto the sensor. The sensor is a light-sensitive device that converts
the light into electrical signals. The sensor is made up of millions of tiny light-sensitive cells
called photodiodes. When light hits a photodiode, it generates an electrical signal. The more
light that hits a photodiode, the stronger the electrical signal.
The sensor converts the light into electrical signals. The electrical signals from the sensor are
then amplified and digitized by the camera's processor. Amplification increases the strength
of the signals, while digitization converts them into a digital format.
The processor applies image processing algorithms to the signals. The processor applies
image processing algorithms to the signals to improve the quality of the image. These
algorithms can be used to adjust the brightness, contrast, saturation, and white balance of the
image. They can also be used to remove noise from the image and to sharpen the edges.
The processor stores the processed signals in memory. The processed signals are then stored
in memory. This memory can be either volatile or non-volatile. Volatile memory is lost when
the camera is turned off, while non-volatile memory retains its data even when the camera is
turned off.
The processor outputs the image to a display or other device. The processed signals are then
output to a display or other device. This can be a computer monitor, a TV, or a printer.
The pipeline diagram above is a simplified representation of the functioning of a digital
camera. The actual process is much more complex, but this diagram gives you a general
overview of how a digital camera works.

11. Explain different types of stereopsis.

Stereopsis is the ability to perceive depth from two images of the same scene taken from
slightly different viewpoints. It is one of the most important cues for depth perception in
humans and other animals.

There are two main types of stereopsis:

Binocular stereopsis: Binocular stereopsis is the most common type of stereopsis. It relies on
the fact that our eyes are spaced apart, which gives us two slightly different views of the same
scene. The brain uses these two views to calculate the distance to objects in the scene.
Monocular stereopsis: Monocular stereopsis is a type of stereopsis that does not require two
eyes. It relies on cues such as perspective, occlusion, and shading to calculate the distance to
objects in the scene.
Here is a more detailed explanation of each type of stereopsis:

Binocular stereopsis: Binocular stereopsis is the most common type of stereopsis. It relies on
the fact that our eyes are spaced apart, which gives us two slightly different views of the same
scene. The brain uses these two views to calculate the distance to objects in the scene.
The brain calculates the distance to an object by comparing the images from the two eyes.
The brain does this by measuring the disparity between the two images. Disparity is the
difference in the position of an object in the two images. For example, if an object is closer to
the left eye than the right eye, then there will be more disparity between the two images.

The brain uses the disparity to calculate the distance to the object. The greater the disparity,
the closer the object is to the observer.

Monocular stereopsis: Monocular stereopsis is a type of stereopsis that does not require two
eyes. It relies on cues such as perspective, occlusion, and shading to calculate the distance to
objects in the scene.
Perspective is the way that objects appear smaller as they get further away. Occlusion is the
way that objects obscure other objects when they are in front of them. Shading is the way that
objects appear lighter or darker depending on their position relative to the light source.

The brain uses all of these cues together to calculate the distance to objects in the scene. For
example, if an object is blocking another object, then the brain knows that the blocking object
is closer to the observer than the blocked object.

Stereopsis is a powerful cue for depth perception. It allows us to perceive depth in a wide
range of situations, even in low-light conditions or when one eye is closed. It is also
important for tasks such as driving, playing sports, and navigating through unfamiliar
environments.
12. What is meant by an “Epipolar Constraint”? How is it represented
algebraically?
In computer vision, an epipolar constraint is a geometric constraint that relates the projections
of a point in 3D space onto two different images. The constraint states that the two
projections lie on a line called the epipolar line.

The epipolar constraint can be represented algebraically as follows:

x′F′T+t′=0
where:

x′ is the projection of the point in 3D space onto the second image.

F′ is the fundamental matrix, which is a matrix that relates the projections of points in 3D
space onto the two images.
T is the translation vector, which is the difference between the positions of the two cameras.
t′ is a scalar that represents the disparity between the two projections.
The epipolar constraint can be used to solve for the 3D coordinates of a point in space given
the projections of the point onto two images. The constraint can also be used to find the
fundamental matrix, which is a valuable tool for stereo vision.

Here is a more detailed explanation of the epipolar constraint:

The epipolar constraint states that the two projections of a point in 3D space onto two
different images lie on a line. This is because the two cameras are taking pictures from
different viewpoints, so the projections of the point will be in different places.

The line that the two projections lie on is called the epipolar line. The epipolar line can be
calculated using the fundamental matrix, which is a matrix that relates the projections of
points in 3D space onto the two images.

The fundamental matrix can be calculated using the following steps:

Find the correspondences between the two images. This means finding the pairs of pixels in
the two images that correspond to the same point in 3D space.
Calculate the cross-correlation between the two images. The cross-correlation is a measure of
the similarity between two images.
Use the cross-correlation to find the fundamental matrix. The fundamental matrix can be
found by solving a system of equations.
Once the fundamental matrix is calculated, it can be used to solve for the 3D coordinates of a
point in space given the projections of the point onto two images. The constraint can also be
used to find the disparity between the two projections.

The epipolar constraint is a powerful tool for stereo vision. It can be used to solve for the 3D
coordinates of points in space, find the fundamental matrix, and calculate the disparity
between two images.
13. Camera parameters.
Camera parameters are the physical characteristics of a camera that affect the way it
captures images. These parameters include:

• Focallength: The focal length is the distance between the lens and the image
sensor. It determines the field of view of the camera.
• Principalpoint: The principal point is the center of the image sensor. It is the
point on the image sensor that corresponds to the center of the camera.
• Image center: The image center is the point on the image where the optical axis
 of the camera intersects the image sensor.
• Distortion:Distortion is a phenomenon that occurs when a camera does not
project rays onto the image sensor perfectly. There are a number of different
types of distortion, such as radial distortion and tangential distortion.
• Resolution: The resolution of a camera is the number of pixels that it can
capture. The higher the resolution, the more detailed the image will be.
• White balance: White balance is the process of adjusting the colors in an image
so that they appear natural. This is important because the color of light can
vary depending on the time of day, the weather, and the location.
• Exposure: Exposure is the amount of light that is allowed to hit the image
sensor. This is important because it affects the brightness of the image.

Camera parameters are important for understanding how cameras work and for
ensuring that images are captured correctly. They are also important for computer
vision algorithms that need to process images.

Here are some additional camera parameters that are not as commonly used as the
ones mentioned above:

• Pixel size: The pixel size is the size of each pixel on the image sensor. It affects
the resolution of the image and the amount of noise in the image.
• Sensor size: The sensor size is the size of the image sensor. It affects the field
of view of the camera and the amount of light that can hit the sensor.
• ISO: The ISO is a setting on the camera that controls the sensitivity of the
sensor to light. It affects the brightness of the image and the amount of noise
in the image.
• Shutter speed: The shutter speed is the amount of time that the shutter is open.
It affects the brightness of the image and the amount of motion blur in the
image.

These are just some of the many camera parameters that can affect the way images
are captured. It is important to understand these parameters in order to use cameras
effectively and to ensure that images are captured correctly.

14. Perspective Projection and its types

Perspective projection is a type of projection in which parallel lines in the real world
are projected onto the image plane as converging lines. This creates the illusion of
depth in images.

There are three main types of perspective projection:

• One-point perspective: In one-point perspective, all parallel lines converge to a

single point on the horizon. This type of projection is often used for cityscapes
and other scenes where there is a strong sense of depth.

Opens in a new window

www.studentartguide.com
One-point perspective
• Two-point perspective: In two-point perspective, there are two vanishing points
 on the horizon. This type of projection is often used for scenes where there is
a strong sense of depth and perspective.

Opens in a new window

makeamarkstudios.com
Two-point perspective
• Three-point perspective: In three-point perspective, there are three vanishing
points on the horizon. This type of projection is rarely used, but it can be used
for scenes where there is a very strong sense of depth and perspective.

Opens in a new window www.craftsy.com

Three-point perspective

Perspective projection is a powerful tool for creating realistic images. It is used in a

wide variety of applications, such as film, video games, and computer graphics.

Here are some of the key features of perspective projection:

• Vanishing points: Parallel lines in the real world are projected onto the image
plane as converging lines. These lines converge at points on the horizon.
• Foreshortening: Objects that are closer to the camera appear larger than
objects that are further away. This is because the parallel lines that make up
the object converge at a closer vanishing point.
• Depth cues: Perspective projection provides a number of depth cues that help
us to perceive the depth of an image. These cues include occlusion, relative
size, and relative height.

Perspective projection is a complex topic, but it is an essential tool for understanding

how images are created and perceived.

15. Reflectance map in detail.

A reflectance map is a two-dimensional image that stores the reflectance properties
of a surface. It is used in computer graphics to simulate the way that light reflects off
of surfaces.

The reflectance map is typically represented as a grayscale image, where each pixel
stores the reflectance of the surface at that point. The reflectance is a value between
0 and 1, where 0 represents a perfectly black surface and 1 represents a perfectly
white surface.

The reflectance map can be used to simulate a variety of surface properties, such
as:
 • Diffuse reflection: Diffuse reflection is the most common type of reflection. It
occurs when light is scattered in all directions by the surface. The reflectance
map for a diffuse surface is uniform, meaning that the reflectance is the same
at all points on the surface.

Opens in a new window en.wikipedia.org

Diffuse reflection
• Specular reflection: Specular reflection occurs when light is reflected in a single
direction. The reflectance map for a specular surface is not uniform, but it has
a bright spot in the direction of the light source.

Opens in a new window en.wikipedia.org

Specular reflection
• Glossy reflection: Glossy reflection is a combination of diffuse and specular
reflection. The reflectance map for a glossy surface has a bright spot in the
direction of the light source, but it also has a diffuse component.

Opens in a new window

www.researchgate.net
Glossy reflection

The reflectance map can be used to create realistic images of objects. It is especially
useful for objects with complex surface properties, such as metals and plastics.

Here are some of the advantages of using a reflectance map:

• Realism: Reflectance maps can be used to create realistic images of objects.

This is because they can simulate the way that light reflects off of surfaces in
the real world.
• Efficiency:
Reflectance maps are relatively efficient to store and process. This
makes them a good choice for real-time applications, such as video games.
• Flexibility:
Reflectance maps can be used to simulate a variety of surface
properties. This makes them a versatile tool for creating realistic images.

Here are some of the disadvantages of using a reflectance map:

 • Accuracy: Reflectance maps are not always accurate. This is because they are
a simplified representation of the way that light reflects off of surfaces in the
real world.
• Complexity: Reflectance maps can be complex to create and use. This is
especially true for objects with complex surface properties.
• Limitations:
Reflectance maps cannot simulate all of the surface properties that
can be found in the real world. This is because they are a two-dimensional
representation of a three-dimensional surface.

16. Explain Various Methods of Edge detection? Also explain the canny edge detection
technique with example.

Edge detection is a technique used to find the boundaries of objects in an image. It is

a fundamental operation in image processing and computer vision.

There are many different edge detection techniques, each with its own advantages
and disadvantages. Some of the most common edge detection techniques include:

• Robert'scross: The Robert's cross operator is a simple edge detection operator

that uses two masks to find edges in an image. The masks are oriented at 45
degrees to each other, and they are applied to the image to find the local
gradients in the horizontal and vertical directions.

Opens in a new window

homepages.inf.ed.ac.uk
Robert's cross edge detection
• Prewitt:
The Prewitt operator is another simple edge detection operator that
uses two masks to find edges in an image. The masks are oriented at 0 and
45 degrees, and they are applied to the image to find the local gradients in the
horizontal and vertical directions.

Opens in a new window

www.researchgate.net
Prewitt edge detection
• Sobel:The Sobel operator is a more sophisticated edge detection operator than
the Robert's cross and Prewitt operators. It uses two masks to find the local
gradients in the horizontal and vertical directions, and it also uses a third
mask to find the magnitude of the gradient.
 Opens in a new window
homepages.inf.ed.ac.uk
Sobel edge detection
• Canny: The Canny edge detection algorithm is a popular edge detection
algorithm that is known for its smoothness and accuracy. It uses a multi-stage
process to find edges in an image, including:
o Noise reduction: The image is first smoothed to remove noise.
o Gradient calculation: The image gradients are calculated in the
horizontal and vertical directions.
o Non-maximum suppression: The image gradients are suppressed if they
are not the local maximum in their neighborhood.
o Hysteresis thresholding: The image edges are finally extracted by
applying two thresholds to the image gradients.

Opens in a new window

towardsdatascience.com
Canny edge detection

The Canny edge detection algorithm is a powerful edge detection algorithm that is
suitable for a wide variety of applications. However, it can be computationally
expensive, especially for large images.

Here is an example of the Canny edge detection algorithm applied to an image:

Opens in a new window

towardsdatascience.com
image before and after Canny edge detection

The original image is on the left, and the Canny edge detection output is on the right.
The Canny edge detection algorithm has successfully detected the edges of the
objects in the image.
 17. Explain Harris Corner Detection in detail.
Harris corner detection is a corner detection algorithm that is used to find points in an
image where there is a large change in intensity in two or more directions. It is a
popular corner detection algorithm that is known for its accuracy and robustness to
noise.

The Harris corner detection algorithm works by calculating the Harris corner
response at each pixel in an image. The Harris corner response is a measure of the
local intensity changes in an image. The Harris corner response is calculated using
the following formula:

R = det(M) - k(trace(M)^2)

where:

•M is the corner matrix

•k is a constant
• det() is the determinant function
• trace() is the trace function

The corner matrix is a 2x2 matrix that is calculated at each pixel in the image. The
corner matrix measures the local intensity changes in the horizontal and vertical
directions.

The Harris corner response is high at pixels where there is a large change in
intensity in two or more directions. These pixels are likely to be corners.

The Harris corner detection algorithm has a number of advantages, including:

• Accuracy: The Harris corner detection algorithm is known for its accuracy. It can
accurately detect corners in images with noise.
• Robustness: The Harris corner detection algorithm is robust to noise. It can still
detect corners in images with noise.
• Speed: The Harris corner detection algorithm is relatively fast. It can be used to
process large images quickly.

However, the Harris corner detection algorithm also has some disadvantages,
including:

• Sensitivity to scale: The Harris corner detection algorithm is sensitive to scale.

This means that it can detect corners at different scales in the same image.
This can be a problem for applications that require the corners to be detected
at a specific scale.
• Sensitivity to illumination: The Harris corner detection algorithm is sensitive to
illumination. This means that it can detect corners in images with different
illumination conditions. This can be a problem for applications that require the
corners to be detected in images with consistent illumination.

Overall, the Harris corner detection algorithm is a powerful corner detection

algorithm that is suitable for a wide variety of applications. It is accurate, robust to
noise, and relatively fast. However, it is important to be aware of its limitations, such
as its sensitivity to scale and illumination.

18. Distinguish Hough Transform and generalized Hough transform.

The Hough Transform and the Generalized Hough Transform are both techniques used in
computer vision and image processing to detect patterns and shapes within images. While
they share some similarities, they have distinct differences in terms of their applications and
capabilities.

Hough Transform: The Hough Transform is a technique primarily used for detecting simple
geometric shapes, such as lines and circles, within an image. It is especially useful when the
shapes cannot be represented conveniently using Cartesian coordinates. The Hough
Transform converts the problem of shape detection into a parameter space, where each point
represents a potential parameter set (e.g., slope and intercept for lines, center and radius for
circles). The process involves:

1. Edge Detection: Detecting edges in the image using techniques like the Canny edge
detector.
2. Parameter Space: Creating an accumulator space (Hough space) to hold the votes for
each potential shape parameter.
3. Voting: For each edge point, voting in the Hough space for the parameters that could
have produced that edge.
4. Finding Peaks: Identifying peaks in the Hough space, which correspond to the most
likely parameters for the detected shapes.
5. Converting Back: Converting the peak parameters back to the image space to obtain
the detected shapes.

Generalized Hough Transform: The Generalized Hough Transform extends the concept of
the original Hough Transform to detect more complex shapes or patterns beyond simple
geometric forms. It achieves this by representing the shape using a template or reference
image, which is often called the "reference object." The process involves:

1. Reference Object: Creating a reference object template, which represents the desired
shape.
2. Voting: Instead of voting for parameters, the Generalized Hough Transform votes for
the positions where the reference object might be present in the image.
3. Hough Space: The accumulator space is used to store the votes for the positions of
the reference object.
4. Finding Peaks: Identifying peaks in the Hough space, which correspond to the likely
positions of the reference object in the image.
5. Converting Back: Converting the peak positions back to the image space to locate
instances of the reference object.

Key Differences:

• Scope of Shapes: The Hough Transform is specialized for detecting simple shapes like
lines and circles, whereas the Generalized Hough Transform can detect more complex
and arbitrary shapes represented by reference objects.
• Parameter vs. Position: In the original Hough Transform, voting is done for parameters
in the Hough space. In the Generalized Hough Transform, voting is done for positions
of the reference object.
• Application: The Hough Transform is suitable for simple shape detection tasks, while
the Generalized Hough Transform is more versatile and applicable to a broader range
of pattern recognition tasks.
• Complexity: Implementing the Generalized Hough Transform is generally more
complex due to the need to handle reference objects and their transformations.

In summary, while both the Hough Transform and the Generalized Hough Transform are used
for shape detection, the Generalized Hough Transform offers greater flexibility by allowing
the detection of more complex and arbitrary shapes using reference objects.

19. Histogram Equalization.

Histogram equalization is an image processing technique that improves the contrast
of an image by stretching its histogram so that its values are more evenly distributed.
This can make images appear brighter and more uniform, and it can also make it
easier to see details in the image.

Histogram equalization works by first creating a histogram of the image's pixels. The
histogram is a graph that shows the number of pixels in the image for each possible
grayscale value. The histogram equalization algorithm then calculates a new
mapping from grayscale values to new grayscale values. This mapping is designed
to stretch the histogram so that its values are more evenly distributed.

The new mapping is applied to the image's pixels, and the result is an image with
improved contrast. Histogram equalization can be a useful technique for improving
the visibility of details in images, and it is often used in image processing
applications such as image enhancement and image segmentation.

Here are some of the advantages of using histogram equalization:

• Improved contrast: Histogram equalization can improve the contrast of an image

by stretching its histogram so that its values are more evenly distributed. This
can make images appear brighter and more uniform, and it can also make it
easier to see details in the image.
• Noise reduction: Histogram equalization can also help to reduce noise in an
image. This is because noise tends to have a uniform distribution, and
histogram equalization can help to spread out the noise so that it is less
noticeable.
• Image enhancement: Histogram equalization can be used to enhance the
appearance of an image. For example, it can be used to make an image look
more dramatic or to make it easier to see certain details.

However, histogram equalization also has some disadvantages, including:

• Loss of detail: Histogram equalization can sometimes cause loss of detail in an

image. This is because the algorithm stretches the histogram so that its
values are more evenly distributed. This can cause some of the fine details in
the image to be lost.
• Artifacts:
Histogram equalization can sometimes introduce artifacts in an image.
These artifacts can be caused by the way that the algorithm stretches the
histogram.
• Over-saturation:Histogram equalization can sometimes cause images to
become over-saturated. This can happen if the algorithm stretches the
histogram too much.

Overall, histogram equalization is a powerful image processing technique that can be

used to improve the contrast and visibility of images. However, it is important to be
aware of its limitations, such as loss of detail and over-saturation.

20. Feature detection algorithm in computer vision to detect and describe the local
features in images.

There are many feature detection algorithms in computer vision that can be used to
detect and describe the local features in images. Some of the most common feature
detection algorithms include:

• SIFT: Scale-invariant feature transform (SIFT) is a popular feature detection

algorithm that is known for its accuracy and robustness to scale, rotation, and
illumination changes. SIFT works by detecting keypoints in an image and then
calculating a descriptor for each keypoint. The descriptor is a vector that
represents the local intensity changes around the keypoint.
 Opens in a new window

docs.opencv.org
• SURF: Speeded up robust features (SURF) is a faster alternative to SIFT that is
also known for its accuracy and robustness to scale, rotation, and illumination
changes. SURF works by detecting keypoints in an image and then
calculating a descriptor for each keypoint. The descriptor is a vector that
represents the local intensity changes around the keypoint.

Opens in a new window medium.com

• FAST: Features from accelerated segment test (FAST) is a very fast feature
detection algorithm that is not as accurate as SIFT or SURF. However, FAST
is very efficient and can be used to detect features in real-time. FAST works
by detecting edges in an image and then calculating a response for each
edge. The response is a value that indicates how likely the edge is to be a
feature.

Opens in a new window

docs.opencv.org
• ORB:Oriented FAST and rotated BRIEF (ORB) is a combination of FAST and
BRIEF. ORB is a fast and efficient feature detection algorithm that is also
known for its accuracy. ORB works by detecting edges in an image using
FAST, and then calculating a descriptor for each edge using BRIEF. The
descriptor is a vector that represents the local intensity changes around the
edge.
 Opens in a new window medium.com

Feature detection algorithms are an important part of many computer vision tasks,
such as object detection, image matching, and image stitching. The choice of feature
detection algorithm depends on the specific task and the requirements of the
application.

21. Using appropriate mathematical expressions, explain convolution used in computer

vision and briefly explain its function and applications.
Convolution is a mathematical operation that is used to combine two functions. In
computer vision, convolution is used to blur images, sharpen images, detect edges,
and extract features.

The convolution operation is defined as follows:

f(x) * g(x) = \sum_{i=-k}^{k} \sum_{j=-k}^{k} f(x - i) g(y - j)

where:

• f(x) is the first function, also known as the kernel or filter

• g(x) is the second function, also known as the image
•k is the kernel size
•x and y are the coordinates of the output pixel

The convolution operation is performed by sliding the kernel over the image and
multiplying the kernel values with the image values at each location. The output of
the convolution operation is a new image that has been filtered by the kernel.

The function of convolution in computer vision depends on the kernel that is used.
Some common kernels include:

• Gaussian kernel: The Gaussian kernel is used to blur images. It is a bell-shaped

kernel that weights the pixels in the neighborhood of the output pixel
according to their distance from the output pixel.

Opens in a new window

www.researchgate.net
Gaussian kernel
• Sobel kernel: The Sobel kernel is used to detect edges in images. It is a two-
dimensional kernel that weights the pixels in the neighborhood of the output
pixel according to their gradient.
 Opens in a new window
homepages.inf.ed.ac.uk
Sobel kernel
• Laplacian kernel: The Laplacian kernel is used to sharpen images. It is a two-
dimensional kernel that weights the pixels in the neighborhood of the output
pixel according to their second derivative.

Opens in a new window

www.researchgate.net
Laplacian kernel

Convolution is a powerful tool for image processing. It can be used to achieve a

variety of effects, such as blurring, sharpening, edge detection, and feature
extraction. Convolution is used in a wide variety of computer vision applications,
such as:

• Imageprocessing: Convolution is used to blur images, sharpen images, and

detect edges.
• Objectdetection: Convolution is used to extract features from images that can
be used to identify objects.
• Imageclassification: Convolution is used to classify images into different
categories.
• Medicalimaging: Convolution is used to process medical images, such as MRIs
and X-rays.
• Machinelearning: Convolution is used in machine learning algorithms, such as
convolutional neural networks.

Convolution is a versatile and powerful tool that can be used to achieve a variety of
effects in computer vision. It is a fundamental operation in many computer vision
algorithms and applications.

22. Discuss the basic model and statistical model algorithms used for background
subtraction.
Background subtraction is a technique used in computer vision to separate foreground
objects from background in a video sequence. It is a fundamental operation in many
computer vision applications, such as object tracking, motion detection, and video
surveillance.

There are two main types of background subtraction algorithms: basic models and
statistical models.
 Basic models are simple algorithms that use a single image to represent the
background. The background image is typically updated every few frames to account
for changes in the environment. When a new frame is captured, the algorithm
compares the pixels in the new frame to the pixels in the background image. Any
pixels that are significantly different from the background are considered to be
foreground objects.

One of the simplest basic models is the thresholding model. The thresholding model
sets a threshold on the difference between the pixels in the new frame and the pixels
in the background image. Any pixels that are above the threshold are considered to be
foreground objects.

Another basic model is the Gaussian mixture model. The Gaussian mixture model
assumes that the background image can be represented by a mixture of Gaussian
distributions. The algorithm first estimates the parameters of the Gaussian
distributions from the background image. When a new frame is captured, the
algorithm calculates the probability that each pixel in the new frame belongs to each
of the Gaussian distributions. Any pixels that have a high probability of belonging to
a foreground Gaussian distribution are considered to be foreground objects.

Statistical models are more complex algorithms that use multiple images to represent
the background. The background model is typically updated every few frames to
account for changes in the environment. When a new frame is captured, the algorithm
compares the pixels in the new frame to the pixels in the background model. Any
pixels that are significantly different from the background model are considered to be
foreground objects.

One of the most popular statistical models is the AdaBoost model. The AdaBoost
model uses a set of weak classifiers to classify pixels as foreground or background.
The weak classifiers are trained on a set of training images. When a new frame is
captured, the algorithm applies the weak classifiers to the pixels in the new frame.
The pixels that are classified as foreground by a majority of the weak classifiers are
considered to be foreground objects.

Another popular statistical model is the Hidden Markov Model (HMM). The HMM is
a probabilistic model that can be used to represent the temporal changes in a sequence
of images. The HMM is trained on a set of training images. When a new frame is
captured, the algorithm uses the HMM to predict the most likely state of the
background in the new frame. Any pixels that are significantly different from the
predicted state are considered to be foreground objects.

Background subtraction is a challenging problem, and there is no single algorithm that

is best for all situations. The choice of algorithm depends on the specific application
and the requirements of the system.

23. Using appropriate mathematical expressions and diagrams, explain the following: -
(a)Perspective projection (b) Epipolar Geometry

Sure, here are the explanations for perspective projection and epipolar geometry,
respectively:

Perspective projection is a type of projection in which parallel lines in the real world
are projected onto the image plane as converging lines. This creates the illusion of
depth in images.

The mathematical equation for perspective projection is as follows:

x = f(X/Z)
y = f(Y/Z)
z = Z

where:
 •x and y are the coordinates of the projected point in the image plane
•X and Y are the coordinates of the point in the real world
•Z is the distance from the camera to the point in the real world
•f is a function that maps the real world coordinates to the image plane
coordinates

The diagram below shows an example of perspective projection. The parallel lines in
the real world are projected onto the image plane as converging lines.

Opens in a new window

www.geeksforgeeks.org
Perspective projection

Epipolar geometry is a collection of geometric relationships between corresponding

points in two images that are taken from different viewpoints. These relationships
can be used to solve for the relative pose of the cameras and to track objects across
multiple images.

One of the most important epipolar relationships is the epipolar constraint. The
epipolar constraint states that the projections of a point in 3D space onto two images
lie on a line called the epipolar line. The epipolar line is defined by the camera
centers and the projection of the point in 3D space.

The diagram below shows an example of the epipolar constraint. The two cameras
are represented by the blue and red circles. The point P is a point in 3D space. The
projections of P onto the two images lie on the epipolar lines.

Opens in a new window en.wikipedia.org

Epipolar geometry

The epipolar constraint can be used to solve for the relative pose of the cameras.
The relative pose is the transformation that maps points from one camera coordinate
system to the other. The relative pose can be solved for by finding the intersection of
the epipolar lines.

The epipolar constraint can also be used to track objects across multiple images.
The object's position in the first image can be used to find the epipolar line in the
second image. The intersection of the epipolar line and the second image is the
object's position in the second image. This process can be repeated to track the
object across multiple images.

Epipolar geometry is a powerful tool for solving problems in computer vision. It can
be used to solve for the relative pose of cameras, track objects across multiple
images, and more.

24. Discuss about combining views from multiple cameras.

Combining views from multiple cameras can be used to improve the quality of
images and to solve problems that are difficult or impossible to solve with a single
camera.

Here are some of the benefits of combining views from multiple cameras:

• Increased resolution: The resolution of an image is the number of pixels in the

image. By combining views from multiple cameras, we can increase the
resolution of the image. This is because the total number of pixels in the
combined image is equal to the sum of the number of pixels in the individual
images.
• Improved depth perception: Depth perception is the ability to perceive the
distance of objects in the environment. By combining views from multiple
cameras, we can improve depth perception. This is because the relative
positions of the cameras can be used to calculate the depth of objects in the
scene.
• Betterobject tracking: Object tracking is the ability to track the movement of
objects in an image or video sequence. By combining views from multiple
cameras, we can better track objects in the scene. This is because the
multiple views can be used to identify the object in each view and to track its
movement between views.

Here are some of the problems that can be solved by combining views from multiple
cameras:

• 3Dreconstruction: 3D reconstruction is the process of creating a 3D model of

an object or scene from multiple 2D images. By combining views from multiple
cameras, we can create more accurate 3D models of objects and scenes.
• Structurefrom motion: Structure from motion is the process of estimating the 3D
structure of an object or scene from its motion over time. By combining views
from multiple cameras, we can more accurately estimate the 3D structure of
objects and scenes.
• Objectdetection: Object detection is the ability to automatically detect objects in
images and videos. By combining views from multiple cameras, we can more
accurately detect objects in the scene. This is because the multiple views can
be used to identify the object in each view and to track its movement between
views.

There are many different techniques for combining views from multiple cameras.
Some of the most common techniques include:

• Image stitching: Image stitching is the process of combining multiple images

into a single image. Image stitching can be used to create panoramic images,
wide-angle images, and high-resolution images.
• Stereovision: Stereo vision is the process of using two cameras to create a 3D
model of an object or scene. Stereo vision can be used to measure the depth
of objects in the scene and to create 3D models of objects and scenes.
• Multi-viewgeometry: Multi-view geometry is the study of the geometric
relationships between multiple views of an object or scene. Multi-view
geometry can be used to solve for the relative pose of cameras, track objects
across multiple images, and reconstruct 3D models of objects and scenes.

The choice of technique depends on the specific application and the requirements of
the system.
 25. Explain all edge based approaches of segmentation.
Edge-based segmentation approaches are a type of segmentation algorithm that
identifies image regions based on their edges. Edges are discontinuities in the image
intensity that can be detected using a variety of techniques.

Here are some of the most common edge-based segmentation approaches:

• Thresholding: Thresholding is a simple technique that classifies pixels as either

foreground or background based on their intensity values. Pixels with
intensities above a certain threshold are classified as foreground, and pixels
with intensities below the threshold are classified as background.

Opens in a new window

www.semanticscholar.org
Thresholding edge detection
• Gradient-based: Gradient-based approaches identify edges by detecting
discontinuities in the image gradient. The image gradient is a measure of the
change in intensity between neighboring pixels. Gradient-based approaches
typically use a threshold to classify pixels as either foreground or background
based on their gradient magnitude.

Opens in a new window

www.geeksforgeeks.org
Gradient-based edge detection
• Canny edge detection: Canny edge detection is a popular gradient-based edge
detection algorithm. Canny edge detection uses a multi-stage process to
identify edges that are both strong and well-defined.

Opens in a new window

towardsdatascience.com
Canny edge detection
• Ridge detection: Ridge detection is a gradient-based edge detection algorithm
that identifies ridges in the image. Ridges are regions of the image where the
intensity changes gradually in one direction. Ridge detection algorithms can
be used to identify objects with smooth edges, such as cars and buildings.
 Opens in a new window stackoverflow.com
Ridge detection

Edge-based segmentation approaches are relatively simple to implement and can be

used to segment a wide variety of images. However, they can be sensitive to noise
and can produce inaccurate results in images with low contrast.

Here are some of the advantages of edge-based segmentation approaches:

• Simple to implement
• Fast

• Can be used to segment a wide variety of images

Here are some of the disadvantages of edge-based segmentation approaches:

• Sensitive to noise
• Can produce inaccurate results in images with low contrast
• Not suitable for segmenting objects with smooth edges

Overall, edge-based segmentation approaches are a powerful tool for segmenting

images. However, they are not always the best choice for every application.

26. Explain principal component analysis with the help of example.

Sure. Principal component analysis (PCA) is a statistical procedure that transforms a
set of observations of possibly correlated variables into a set of values of linearly
uncorrelated variables called principal components.

PCA is a widely used technique in machine learning and data science for
dimensionality reduction. It can be used to reduce the number of features in a
dataset without losing too much information. This can be helpful for improving the
performance of machine learning algorithms and for making data visualization easier.

Here is an example of how PCA can be used to reduce the dimensionality of a

dataset. Consider the following dataset of images of handwritten digits:

[Image of handwritten digit 0]

[Image of handwritten digit 1]
[Image of handwritten digit 2]
...
[Image of handwritten digit 9]

Each image in this dataset is a 28x28 pixel image. This means that each image is
represented by 784 features (28x28). PCA can be used to reduce the dimensionality
of this dataset to 10 features. The 10 principal components are the directions in the
feature space that contain the most variance in the data.

The following image shows the first 2 principal components of the handwritten digit
dataset:
 Opens in a new window www.researchgate.net
2 principal components of handwritten digit dataset

The first principal component is a direction that captures the overall brightness of the
image. The second principal component is a direction that captures the vertical
elongation of the image.

PCA can be used to reduce the dimensionality of the handwritten digit dataset to 10
features without losing too much information. This can be helpful for improving the
performance of machine learning algorithms that are used to classify handwritten
digits.

Here are some of the advantages of PCA:

• It can be used to reduce the dimensionality of a dataset without losing too much
information.
• It is relatively easy to implement.
• It is a non-parametric technique, which means that it does not make any
assumptions about the distribution of the data.

Here are some of the disadvantages of PCA:

• It can be sensitive to noise in the data.

• It can be computationally expensive for large datasets.
• It can create principal components that are not meaningful.

Overall, PCA is a powerful tool for dimensionality reduction that can be used to
improve the performance of machine learning algorithms and to make data
visualization easier.

27. KNN and ANN model.

It seems like you might want more information about K-nearest neighbors (KNN) and
artificial neural network (ANN) models. I'll provide you with a more detailed explanation of
each:

K-Nearest Neighbors (KNN) Model: K-nearest neighbors (KNN) is a simple yet effective
machine learning algorithm used for classification, regression, and sometimes clustering
tasks. Here's how it works:

1. Training Phase:
• During the training phase, KNN stores all available labeled data points in its
memory.
• It doesn't actually learn a model or tune any parameters during training. Instead,
it memorizes the training data.
2. Prediction Phase:
• When a new, unlabeled data point is given for prediction, KNN identifies the k
nearest data points from the training set based on a chosen distance metric
(e.g., Euclidean distance).
• The most common approach is to calculate the class distribution (for
 classification) or the average (for regression) of the labels of the k nearest
neighbors.
• The prediction for the new data point is based on the class with the highest
frequency (for classification) or the calculated average (for regression) among
the k neighbors.

Strengths of KNN:

• Simple to understand and implement.

• Works well with noisy or irregular data.
• Doesn't require assumptions about the underlying data distribution.

Weaknesses of KNN:

• Computationally expensive, especially for large datasets.

• Sensitive tothe choice of distance metric and the value of k.
• Doesn't capture complex patterns in the data as effectively as other algorithms like
ANNs.

Artificial Neural Network (ANN) Model: Artificial neural networks (ANNs) are powerful
models inspired by the human brain's structure. ANNs can perform a wide range of tasks,
including classification, regression, image recognition, and natural language processing.
Here's how they work:

1. Training Phase:
• ANNs consist of layers of interconnected nodes (neurons) that process and
transform data.
• During training, the network learns by adjusting the weights assigned to
connections between neurons. This is done using optimization techniques like
gradient descent and backpropagation.
• The training data is iteratively fed through the network, and the model's
predictions are compared to the actual labels to calculate the error.
• The error is then backpropagated through the network to update the weights,
minimizing the error over time.
2. Prediction Phase:
• After training, the ANN can be used to predict the output for new, unseen data.
• The input data is propagated through the network's layers using the learned
weights, and the final output is obtained.

Strengths of ANN:

• Can capture complex relationships in data, making them suitable for a wide range of
tasks.
• With proper architecture and training, ANNs can achieve state-of-the-art performance.
• Can automatically learn relevant features from the data.

Weaknesses of ANN:

• Can be difficult to train due to the need for proper hyperparameter tuning and sufficient
data.
• Prone to overfitting if not properly regularized.
• Computationally intensive, especially for large and deep networks.

In summary, KNN and ANN are distinct algorithms with their own strengths and weaknesses.
KNN is simple and intuitive, while ANN is versatile and can handle more complex tasks. The
choice between them depends on the specific task, the nature of the data, and the available
resources.

28. K-Means algorithm for clustering and its numerical.

K-means is a clustering algorithm that groups data points into k clusters. The
algorithm works by first randomly assigning k centroids to the data points. The
centroids are the mean of the data points in each cluster. The algorithm then iterates
through the following steps:

1. Assign each data point to the cluster with the nearest centroid.
2. Recalculate the centroids of each cluster.
3. Repeat steps 1 and 2 until the centroids no longer change.
The k-means algorithm is a simple and efficient algorithm that can be used to cluster
a wide variety of data. However, it can be sensitive to the initial choice of centroids.

Here is the pseudocode for the k-means algorithm:

function kmeans(data, k):

# Initialize k centroids randomly
centroids = {}
for i in range(k):
centroids[i] = data[random.randint(0, len(data) - 1)]

# Repeat until centroids no longer change:

while True:
# Assign each data point to the cluster with the nearest centroid
clusters = {}
for data_point in data:
closest_centroid = min(centroids, key=lambda c:
distance(data_point, c))
clusters[closest_centroid] = clusters.get(closest_centroid, [])
+ [data_point]

# Recalculate the centroids of each cluster

for cluster in clusters:
centroids[cluster] = mean(clusters[cluster])

# Check if the centroids have changed

if centroids == old_centroids:
break
old_centroids = centroids.copy()

return clusters

The k-means algorithm can be used to cluster a variety of data, including:

• Customer segmentation
• Image clustering
• Text clustering
• Gene clustering

K-means is a popular clustering algorithm because it is simple to understand and

implement, and it is relatively efficient. However, it can be sensitive to the initial
choice of centroids, and it can sometimes produce suboptimal clustering results.

29. Phong model, photometric stereo, albedo estimation.

Sure. Here are the explanations for Phong model, photometric stereo, and albedo
estimation, respectively:

Phong model is a model that describes the reflection of light from a surface. The
Phong model is a physically-based model that takes into account the surface's
shape, orientation, and material properties.
The Phong model is defined by the following equation:

I = Kd * L * n + Ks * L^n * r

where:

•I is the intensity of the reflected light

• Kd is the diffuse albedo, which is a measure of how much light is reflected
diffusely from the surface
• L is the incident light vector
•n is the surface normal vector
• Ks is the specular albedo, which is a measure of how much light is reflected
specularly from the surface
•n is the surface normal vector
•r is the specular reflection coefficient

The Phong model is a popular model for rendering images in computer graphics. It is
also used in computer vision for tasks such as surface reconstruction and material
estimation.

Photometric stereo is a technique for estimating the surface shape and albedo of an
object from images taken under different lighting conditions. The photometric stereo
technique works by assuming that the surface of the object is Lambertian, which
means that it reflects light equally in all directions.

The photometric stereo technique can be used to estimate the surface shape and
albedo of an object by taking three or more images of the object under different
lighting conditions. The images are then used to solve a system of equations that
relate the surface shape and albedo to the intensity of the light reflected from the
object.

Photometric stereo is a powerful technique for estimating the surface shape and
albedo of objects. It is used in a variety of applications, such as 3D reconstruction,
object recognition, and material estimation.

Albedo estimation is the process of estimating the albedo of a surface. The albedo of
a surface is a measure of how much light is reflected from the surface. The albedo of
a surface can be estimated from images using a variety of techniques, such as
photometric stereo and spectral albedo estimation.

Photometric stereo is a technique for estimating the albedo of a surface by taking

images of the surface under different lighting conditions. The images are then used
to solve a system of equations that relate the surface albedo to the intensity of the
light reflected from the surface.

Spectral albedo estimation is a technique for estimating the albedo of a surface by

taking images of the surface in different spectral bands. The images are then used to
fit a model to the spectral reflectance of the surface. The model can be used to
estimate the albedo of the surface in each spectral band.

Albedo estimation is a useful technique for a variety of applications, such as material

identification, remote sensing, and computer graphics

30. Why are color models necessary in computer vision? Explain any one color model in
detail.

Color models are necessary in computer vision because they provide a way to
represent and manipulate color information. Color models are used in a variety of
computer vision tasks, such as:
 • Image segmentation: Color models can be used to segment images into
different regions based on their color.
• Object detection: Color models can be used to detect objects in images by
finding regions of the image that match a particular color.
• Imageclassification: Color models can be used to classify images into different
categories based on their color.
• Image retrieval: Color models can be used to retrieve images from a database
that match a particular color.

One popular color model is the RGB color model. The RGB color model represents
colors as a combination of red, green, and blue light. The RGB color model is the
most common color model used in computer vision.

The RGB color model is defined by the following equation:

color = (red, green, blue)

where:

• red is the amount of red light

• green is the amount of green light
• blue is the amount of blue light

The RGB color model is a additive color model. This means that the colors are
added together to create new colors. For example, the color white is created by
adding red, green, and blue light together.

The RGB color model is not the only color model that can be used in computer
vision. Other popular color models include:

• HSV color model: The HSV color model represents colors as hue, saturation,
and value. The hue is the color itself, the saturation is the intensity of the
color, and the value is the brightness of the color.
• YUV color model: The YUV color model represents colors as luma,
chrominance U, and chrominance V. The luma is the brightness of the color,
the chrominance U is the amount of blue light, and the chrominance V is the
amount of red light.

The choice of color model depends on the specific task at hand. The RGB color
model is a good choice for most tasks, but the HSV color model or the YUV color
model may be a better choice for some tasks.