NANDHA ENGINEERING COLLEGE

COMPUTER
VISION 1 7 A I X 0 7

UNIT - 1 INT RODUC T ION TO IM AGE

FOR M AT I ON A N D P ROC E S S I N G

J a n u a r y | 2 0 2 4
 CONTENT
1 Computer Vision
2 Geometric Primitives and
Transformation
3 Photometric Image Formation
4 Digital Camera
5 Point Operators
Computer Vision
What is CV?
 One of the most powe ul and compelling types of AI is
Computer Vision which you have almost surely
experienced in any number of ways without even
knowing.
 Computer Vision is the field of Computer Science that
focuses on replicating pa s of the complexity of the
human vision system and enabling computers to
identify and process object in image and videos in the
same way that human do.
 The Ultimate Goal is for computers to emulate the
striking perceptual capability of human eyes and brains
or even to surpass assist the humen in ce ain way.
 History of Computer Vision
• The origins of computer vision can be traced • In the 1970s and 1980s, advancements were
back to the 1950s and 1960s when made in edge detection, shape recognition,
researchers began exploring pa ern and early a empts at three-dimensional (3D)
recognition and image analysis. Early effo s vision. However, limited computational power
focused on simple tasks like character hindered progress.
recognition.

• The 1990s witnessed an increased focus on • The 2000s saw a surge in computer vision applications,
robust algorithms for image understanding, driven by improved hardware capabilities and machine
feature extraction, and object recognition. learning techniques. Notable breakthroughs include the
Applications extended to medical imaging, development of Convolutional Neural Networks (CNNs)
surveillance, and industrial automation. for image classification and object detection.
 Concept of Computer
Vision
• Computer vision involves teaching machines to interpret and make decisions
based on visual data. The concept encompasses various tasks, including:

 Image acquisition  Segmentation

 Preprocessing  3D Reconstruction

 Feature extraction  Motion analysis

 Image recognition  Machine learning Integration

 Object detection  Applications

 Convolutional Neural Networks
(CNNs)
Definition
• Convolutional Neural Networks (CNNs or ConvNets) are a class of deep
neural networks designed for processing and analyzing visual data, making
them especially effective for tasks such as image recognition and
computer vision. CNNs are characterized by their unique architecture,
which includes convolutional layers, pooling layers, and fully connected
layers.
 Key Concept for CNNs

Convolutional Layers Activation Functions

Convolutional layers apply filters (also Activation functions, such as ReLU
known as kernels) to input data, (Rectified Linear Unit), introduce non-
enabling the network to learn linearity to the network, allowing it to
hierarchical features such as edges, learn complex relationships and
textures, and pa erns. pa erns.
Pooling Layers Fully Connected Layers
Pooling layers downsample the spatial Fully connected layers connect every
dimensions of the input data, reducing neuron from one layer to every neuron
computational complexity and retaining in the next layer, consolidating learned
impo ant features. Common pooling features for classification or regression
operations include max pooling and tasks.
average pooling.
 Key Concept for CNNs

Convolutional Filters Weight Sharing

Filters capture local pa erns in the CNNs leverage weight sharing, where the
input data, effectively learning features same set of parameters (weights and biases)
like edges, corners, and textures. These is used across different regions of the input,
filters are adapted during training to facilitating the detection of similar features in
recognize higher-level features. different pa s of an image.
Striding Padding
Striding controls the step size of the Padding involves adding extra pixels to
filter as it moves across the input data, the input data, preventing information
influencing the spatial dimensions of loss at the borders during convolution.
the output. Striding helps reduce the It ensures that the spatial dimensions
spatial dimensions and computational of the input and output match
appropriately.
load.
Dropout
Dropout is a regularization technique used to
prevent ove i ing by randomly se ing a fraction of
input units to zero during training, reducing reliance
on specific neurons.
Applications
CNNs are widely used in various applications
• Image classification
• Object detection
• Facial recognition
• Image segmentation
• Medical image analysis
• Autonomous vehicles
• Natural language processing (when
combined with recurrent networks)
Geometric Primitives and
Transformation
Definition
 Geometric primitives in computer vision refer to basic shapes and
structures used to represent objects in an image. These primitives serve
as foundational elements for various computer vision tasks, including
object recognition, image analysis, and scene understanding. Common
geometric primitives include points, lines, circles, rectangles, and polygons.
 Geometric transformations involve altering the positions, orientations,
and sizes of geometric primitives within an image. These transformations
play a crucial role in tasks such as image registration, object tracking, and
image manipulation.
 Key Concept

Points Circles
Lorem ipsum dolor sit amet, consectetuer Defined by a center point and a radius, circles are
adipiscing elit. Aenean commodo ligula eget used to model rounded objects or features in
dolor. Represented by coordinates (x, y), points images.
serve as the fundamental building blocks for
constructing more complex geometric shapes.

Lines Rectangles
Lorem ipsum dolor sit amet, consectetuer Composed of four points or defined by a
adipiscing elit. Aenean commodo ligula eget center, width, and height, rectangles are
dolor. Defined by two points or a point and a used to represent objects with rectangular
direction vector, lines are essential for shapes.
representing edges and contours in images.
Polygons
Composed of a sequence of connected points, polygons are
versatile geometric primitives used to represent complex
shapes with multiple sides.
 Key Concept

Translation Scaling
Shi s the position of geometric primitives Enlarges or shrinks geometric primitives based on
horizontally and ve ically. a scaling factor.

Rotation Affine Transformation

Rotates geometric primitives around a Combines translation, rotation, scaling, and
specified point or axis. shearing to pe orm a more generalized
transformation.

Shearing
Disto s geometric primitives by shi ing one axis relative to the
other.
 Photometric Image Formation

Definition
Photometric image formation in computer vision refers to the process by which a digital
image is created based on the interaction of light with a scene and the subsequent
capture of this light by an imaging device, such as a camera. This process involves various
factors related to illumination, reflection, and the characteristics of the imaging system.
 Key Concept

Illumination Reflection Su ace Prope ies Shading Models

Illumination represents the Reflection describes how The material prope ies of Shading models are
incident light on a scene, and su aces in a scene interact su aces, such as diffuse and mathematical representations
it plays a crucial role in image with incident light. Different specular reflectance, affect that simulate how light and
formation. The intensity, materials exhibit varying how light interacts with them. shadows interact with su aces.
direction, and color of light reflectance prope ies, Diffuse reflection is Common models include
impact how objects in the influencing how much light responsible for the sca ered Lambe ian reflectance for diffuse
scene are captured. they absorb or reflect. appearance, while specular su aces and Phong or Blinn-
reflection creates highlights. Phong models for specular
su aces.
 Key Concept

Ambient, Diffuse & Light Source Camera Characteristics

Specular Components The position, type, and
characteristics of light
The characteristics of the
imaging device, such as the
In computer graphics and sources in a scene impact camera, influence image
vision, the interaction of light how objects are illuminated formation. This includes
is o en decomposed into and, consequently, how they factors like exposure time,
ambient, diffuse, and appear in the captured ape ure size, and sensor
specular components. This image. sensitivity.
decomposition aids in
simulating realistic lighting
conditions.
Digital Camera

Definition
A digital camera in computer vision refers to an
electronic imaging device that captures visual
information in the form of digital images. It plays a
fundamental role in computer vision applications by
providing a means to acquire, process, and analyze
visual data for various tasks such as image recognition,
object detection, and scene understanding.
 Key Concept
The image sensor
conve s light into
electrical signals, forming
the basis for digital image
creation. Common types
Resolution The lens system focuses
include CMOS light onto the image
(Complementary Metal- The resolution of a digital sensor. The choice of
Oxide-Semiconductor) and camera is the number of lenses influences factors
CCD (Charge-Coupled pixels it can capture. like focal length, ape ure,
Device) sensors. Higher resolution and depth of field.
contributes to finer
details in images.
Image censor Lens System
 Key Concept

Shu er Speed
Digital cameras capture
color information using & Exposure White balance
RGB (Red, Green, Blue) adjustments ensure
channels. Color Shu er speed and accurate color
representation is crucial exposure se ings reproduction under
for various computer vision determine how long the different lighting
tasks. camera's shu er remains conditions.
open. They impact the
amount of light reaching
Colour the sensor and influence White Balance
Representation image quality.
 Key Concept

In-built image processing

capabilities enhance
Frame Rate Autofocus systems and
images and may include focus modes contribute
features like noise The frame rate indicates to capturing sharp and
reduction, sharpness how many images per clear images by adjusting
adjustment, and color second the camera can the focus based on the
correction. capture. It is crucial for scene.
applications like video
analysis.
Image Auto Focus
Processing
 Point Operators

Definition
• Point operators in computer vision refer to a class of image processing
operations where each pixel in an image is independently transformed
based on a predefined mathematical function. These operations are
applied individually to every pixel, o en without considering the
surrounding pixels, and are fundamental for enhancing or modifying image
characteristics.
 Key Concept

Pixel Transformation Gamma Correction

Point operators involve Gamma correction is a point
transforming the intensity or color operation used to adjust the
of each pixel independently based overall brightness of an image,
on a specific mathematical rule or pa icularly in cases where the
function display device's characteristics
need to be taken into account.
Brightness & Contrast Thresholding
Adjustment Thresholding is a point operation
Common point operators include that conve s an image into a
operations for adjusting the binary form by se ing pixels above
brightness and contrast of an or below a ce ain threshold to
image. These operations scale or specific values.
shi pixel values to achieve the
desired visual effect.
 Key Concept
Logarithmic & Exponential
Histogram Equalization Transformation
Histogram equalization is a point Logarithmic and exponential
operation that enhances the transformations are point
contrast of an image by operations used for enhancing
redistributing pixel values to cover details in ce ain intensity ranges.
the entire intensity range.
Negative Transformation Bit-Plane Slicing
A simple point operation involves
obtaining the negative of an Bit-plane slicing involves extracting
image, where pixel values are specific bits from the binary
inve ed. representation of pixel values,
allowing for detailed analysis or
modification.