Q1.

Define Computer Vision and Explain Its Applications in Real-World

Scenarios
Computer vision is a multidisciplinary domain of artificial intelligence that enables
machines to analyze, interpret, and understand the visual world in the same way as
human vision. It involves transforming image data into symbolic or numerical information
that can be processed and analyzed by algorithms. The imaging process in computer
vision can be modeled as g(x, y) = f(x, y) * h(x, y) + n(x, y), where g(x, y) represents the
observed image, f(x, y) is the true scene, h(x, y) is the point spread function of the imaging
system, and n(x, y) represents the noise introduced during image acquisition. The aim of
computer vision is to infer meaningful and structured representations of objects and
scenes from unstructured visual inputs. Computer vision systems work through several
stages, beginning with image acquisition using sensors, followed by preprocessing
operations like filtering, histogram equalization, and noise reduction. This is followed by
feature extraction using methods such as gradient-based or texture-based operators, and
finally, decision-making using classification or segmentation algorithms. Modern computer
vision systems rely heavily on deep learning, especially Convolutional Neural Networks
(CNNs) and Vision Transformers (ViTs), to achieve high accuracy. In real-world
applications, computer vision plays a vital role in diverse fields. In autonomous vehicles, it
enables lane detection, pedestrian recognition, and obstacle avoidance through real-time
image interpretation. In medical imaging, computer vision assists radiologists in detecting
diseases like tumors, fractures, and cardiovascular abnormalities. In surveillance systems,
it supports activity recognition, motion tracking, and facial recognition for security
purposes. Industrial automation utilizes computer vision for defect inspection, product
counting, and robotic assembly. In agriculture, it is applied for plant disease identification
and crop yield estimation. Additionally, it is integral to augmented reality (AR), virtual
reality (VR), and robotics for environmental mapping and interaction. Thus, computer
vision forms a crucial bridge between raw visual data and intelligent decision-making,
revolutionizing several industries through automation and precision.
Q2. Describe the Photometric Image Formation Process in Computer
Vision
The photometric image formation process focuses on how light interacts with objects and
surfaces before being captured by an imaging device. Unlike geometric models that
emphasize spatial relationships, the photometric model describes the brightness and
intensity distribution within an image. The intensity I(x, y) of a pixel is determined by two
factors: illumination L(x, y), which refers to the amount of light incident on the surface, and
reflectance R(x, y), which defines the proportion of light reflected by the surface material.
The relationship is expressed as I(x, y) = L(x, y) * R(x, y). In computer vision,
understanding this process is crucial for accurate interpretation of brightness variations
and shading. The illumination component depends on the position and intensity of light
sources, while reflectance is determined by the material properties and surface orientation.
Various lighting models are used to approximate real-world conditions. The Lambertian
model assumes diffuse reflection, where brightness depends only on the cosine of the
angle between the light source and the surface normal. In contrast, specular reflection
models, such as Phong and Torrance–Sparrow, account for mirror-like reflections where
intensity depends on both viewing direction and light source angle. The photometric
process also considers phenomena like interreflections, shading, and shadows, which
affect image intensity. For example, the shading equation I = k_d * (N • L) represents
diffuse reflection, where k_d is the diffuse reflectance coefficient, N is the surface normal,
and L is the light direction vector. Advanced techniques like shape-from-shading and
photometric stereo exploit variations in illumination to reconstruct 3D surface geometry
from 2D images. Thus, photometric modeling plays a crucial role in applications such as
surface inspection, material recognition, and 3D reconstruction.
Q3. Explain the Working of a Digital Camera in Capturing Images
A digital camera captures images by converting light energy from a scene into electrical
signals that are processed and stored digitally. The process begins when light passes
through the camera lens, which focuses the incoming rays onto an image sensor. The lens
system controls key parameters such as focal length and aperture. The aperture regulates
the amount of light entering the system, while the shutter controls exposure time. The
focused light is then projected onto a sensor, typically a CCD (Charge-Coupled Device) or
CMOS (Complementary Metal-Oxide Semiconductor) array. Each pixel on the sensor
contains a photodiode that converts incoming photons into electrons through the
photoelectric effect. The number of electrons generated is proportional to the light intensity
falling on that pixel. The analog signal produced by this conversion is then amplified,
digitized using an analog-to-digital converter (ADC), and stored as a digital image. A color
filter array, commonly the Bayer filter, is used to separate light into red, green, and blue
components, allowing the creation of full-color images through interpolation techniques
known as demosaicing. Additional processing steps such as white balance adjustment,
gamma correction, and noise reduction enhance image quality. Lens distortion correction
and sharpening filters are applied to refine spatial details. The final image is encoded and
stored in digital formats such as JPEG or RAW. The digital camera thus acts as a
complete optical-electronic imaging system, where the transformation of optical signals
into digital pixel values enables advanced image analysis, storage, and transmission. The
high sensitivity, compactness, and accuracy of modern digital sensors have made digital
cameras the backbone of computer vision, robotics, and surveillance systems.
Q4. Discuss the Role of Point Operators in Image Processing with
Examples
Point operators are fundamental tools in image processing that modify the intensity value
of each pixel independently of its neighbors. They are also known as pixel-wise
transformations because they act directly on pixel intensity values to enhance or correct
image properties. Mathematically, a point operation can be expressed as g(x, y) = T[f(x,
y)], where T is a transformation function applied to each pixel of the input image f(x, y) to
produce the output image g(x, y). Common point operations include brightness and
contrast adjustments, thresholding, and intensity transformations such as logarithmic and
power-law (gamma) corrections. Brightness adjustment is achieved by adding a constant
value to all pixels, shifting the intensity histogram toward higher or lower values. Contrast
enhancement modifies the spread of intensity levels, often through linear stretching or
histogram equalization. The logarithmic transformation g(x, y) = c * log(1 + f(x, y))
enhances dark regions, while power-law transformation g(x, y) = c * f(x, y)^γ adjusts
contrast dynamically. Thresholding is another important point operation used for image
segmentation, where each pixel is classified as foreground or background depending on a
chosen intensity threshold T. Binary images are generated according to g(x, y) = 1 if f(x, y)
> T, else 0. Point operations are computationally efficient and form the basis for more
advanced processing methods such as edge detection and morphological transformations.
They are essential in applications like document enhancement, medical imaging, and
preprocessing for object detection, where rapid pixel-level manipulation improves image
interpretability and quality.
Q5. Differentiate Between Linear and Non-Linear Filtering Techniques
Filtering techniques are employed in image processing to enhance or restore images by
suppressing noise or emphasizing important features. Linear filtering assumes that the
output image intensity at each pixel is a weighted linear combination of its neighborhood
pixels. It satisfies the principle of superposition. A typical linear filter is represented as g(x,
y) = ΣΣ h(i, j) * f(x - i, y - j), where h(i, j) is the filter kernel. Examples include mean filtering,
Gaussian smoothing, and Laplacian filters. Linear filters are efficient and easy to
implement but are often sensitive to edge blurring. Non-linear filters, on the other hand, do
not obey the superposition principle. They use non-linear mathematical operations to
reduce noise while preserving edges and details. The median filter is a classic example,
where each pixel is replaced by the median of its neighborhood intensities. This approach
is effective against impulse noise (salt-and-pepper noise). Other non-linear filters include
the bilateral filter, which combines spatial and intensity information, and the adaptive
median filter that changes its window size based on noise level. Linear filters perform well
in Gaussian noise conditions and for applications like smoothing, whereas non-linear filters
are better suited for preserving structural information and removing non-Gaussian noise.
In practical computer vision tasks, a combination of both techniques is often used to
achieve optimal results, balancing detail preservation with noise suppression.
Q6. Write Short Notes on the History of Computer Vision
The evolution of computer vision spans over six decades, originating from early research
in artificial intelligence and image analysis. In the 1950s and 1960s, the focus was on
enabling computers to interpret simple patterns and shapes, such as recognizing
handwritten characters or geometric objects. The 1970s saw the emergence of
model-based vision, where mathematical models were developed to describe object
geometry and motion. Techniques such as edge detection and region segmentation
became popular, forming the basis for feature extraction. During the 1980s and 1990s,
computer vision expanded rapidly with the development of robust algorithms for motion
tracking, stereo vision, and object recognition. The introduction of the Marr’s vision theory
proposed a computational model for visual perception, emphasizing multiple levels of
visual representation. The 2000s marked the rise of machine learning-based vision
techniques, with algorithms like Support Vector Machines (SVMs) being used for object
classification and detection. The introduction of large labeled datasets such as ImageNet
enabled data-driven approaches. The 2010s initiated a deep learning revolution in
computer vision, led by the success of Convolutional Neural Networks (CNNs) like
AlexNet, VGG, and ResNet, which dramatically improved performance in image
recognition benchmarks. Modern advancements include Vision Transformers (ViTs),
self-supervised learning, and multimodal AI systems capable of integrating vision with
language understanding. Today, computer vision has matured into a core component of
AI, powering technologies in healthcare, robotics, autonomous systems, and security
surveillance.
Q7. How is Digital Image Representation Achieved? Illustrate with
Examples
A digital image is represented as a two-dimensional array of discrete picture elements
known as pixels. Each pixel corresponds to a small region in the continuous scene and
contains intensity or color information. Mathematically, an image is expressed as a matrix
f(x, y), where x and y denote spatial coordinates, and f(x, y) represents the gray level or
color value at that location. In grayscale images, pixel values range from 0 (black) to 255
(white) for 8-bit representation. Color images are represented using multiple channels,
commonly Red, Green, and Blue (RGB). Each channel stores intensity values
independently, forming a 3D matrix representation. Images can also be represented in
alternative color spaces such as HSV (Hue, Saturation, Value), YCbCr, or Lab to suit
specific applications. For example, HSV separates chromatic information from luminance,
which is beneficial for color-based segmentation. Binary images, where each pixel is either
0 or 1, are used in document processing or shape analysis. Higher-bit-depth images (e.g.,
16-bit or 32-bit) enable precise scientific and medical imaging applications. The process of
digital representation involves sampling and quantization. Sampling discretizes the spatial
coordinates, while quantization discretizes intensity levels. According to the Nyquist
theorem, the sampling rate must be at least twice the highest frequency in the image to
avoid aliasing. Thus, digital image representation is fundamental to all computer vision
tasks, providing a bridge between analog real-world scenes and computational analysis.
Q8. Explain How Image Brightness and Contrast Adjustments Are
Performed
Brightness and contrast adjustments are key preprocessing operations used to improve
the visual quality and interpretability of digital images. Brightness refers to the overall
lightness or darkness of an image, whereas contrast defines the difference between the
lightest and darkest regions. Mathematically, brightness adjustment involves adding or
subtracting a constant value b to each pixel intensity: g(x, y) = f(x, y) + b. Positive b
brightens the image, while negative b darkens it. Contrast adjustment modifies the
dynamic range of intensity values and is given by g(x, y) = a * f(x, y), where a is the
contrast scaling factor. When a > 1, contrast is enhanced, making details more
pronounced, while 0 < a < 1 compresses contrast. Advanced methods such as histogram
equalization automatically redistribute pixel intensities to achieve uniform contrast
distribution. In histogram equalization, the transformation function T(r) = (L-1) * ∫■■ p_r(w)
dw maps the cumulative distribution function of pixel intensities, where L is the number of
gray levels and p_r(w) is the probability density function. Contrast-limited adaptive
histogram equalization (CLAHE) extends this concept by applying localized adjustments to
prevent over-enhancement. These techniques are essential in medical imaging, satellite
imagery, and photography, where lighting conditions vary. Brightness and contrast
manipulation ensures consistent visual appearance and enhances the performance of
downstream algorithms such as edge detection and segmentation.
Q9. Compare Photometric Image Formation with Geometric Image
Formation
Photometric and geometric image formation are two complementary processes that
describe how images are generated and perceived. The photometric process focuses on
the interaction of light with surfaces and the resulting brightness distribution, while the
geometric process deals with spatial mapping between 3D scenes and 2D images. In
photometric formation, image intensity I(x, y) = L(x, y) * R(x, y) is determined by
illumination L(x, y) and reflectance R(x, y). It explains shading, reflection, and color
properties influenced by lighting conditions. In contrast, geometric image formation models
the projection of 3D points onto a 2D image plane using the pinhole camera model. The
transformation is represented by x' = f * (X/Z) and y' = f * (Y/Z), where (X, Y, Z) are 3D
world coordinates, (x', y') are image plane coordinates, and f is the focal length. Geometric
modeling defines relationships such as perspective projection, depth perception, and
object scaling based on viewpoint. It plays a crucial role in camera calibration, 3D
reconstruction, and motion estimation. Photometric models handle the intensity domain,
while geometric models operate in the spatial domain. Together, they form the complete
image formation framework, allowing computer vision systems to infer both physical
structure and illumination characteristics of a scene.
Q10. Discuss the Advantages of Using Linear Filters in Image
Enhancement
Linear filters play a vital role in image enhancement and restoration by providing
mathematically stable and predictable operations. They are based on the principle of
linearity, where the output is a weighted sum of neighboring pixel intensities. The main
advantages of linear filters include simplicity, computational efficiency, and well-defined
frequency domain behavior. Linear filters such as mean, Gaussian, and Laplacian filters
are commonly used to perform smoothing, edge detection, and sharpening. One of the key
benefits of linear filtering is that it can be analyzed using Fourier transform theory.
Convolution in the spatial domain corresponds to multiplication in the frequency domain,
allowing for precise control of frequency components. For instance, Gaussian filters
effectively suppress high-frequency noise while preserving low-frequency image content.
Linear filters are also shift-invariant, ensuring consistent behavior across the image. Their
mathematical structure enables easy implementation using convolution kernels or
frequency domain techniques like Fast Fourier Transform (FFT). Although non-linear filters
may outperform them in preserving edges, linear filters remain widely used due to their
predictability, ease of tuning, and applicability to a variety of image types. They form the
basis of many preprocessing pipelines, particularly where real-time performance is
essential. Thus, linear filters are indispensable tools in image enhancement, providing a
balanced trade-off between performance, simplicity, and analytical clarity.