RAJIV GANDHI PROUDYOGIKI VISHWAVIDYALAYA,

BHOPAL New Scheme Based On AICTE

Flexible Curricula Computer Science and Engineering,

Scan for YouTube

UNIT V

Explain Knowledge Based Vision: Knowledge representation, Control

Control-strategies,
strategies, Information
Integration
Knowledge-Based Vision is an approach to computer vision that focuses on the
integration of knowledge and information processing techni
techniques
ques with computer vision.
This approach is based on the idea that the knowledge of the world and the ability to
reason about it is crucial to the understanding of visual information. In this article, we
will discuss the key concepts of knowledge
knowledge-based vision,
ion, including knowledge
representation, control strategies, and information integration.

Knowledge Representation Knowledge representation is the process of encoding

knowledge in a form that can be used by a computer. In knowledge
knowledge-based
based vision, the
representation
entation of knowledge is crucial because it enables the computer to reason about
the visual information it receives. There are various methods of knowledge
representation, including rule--based systems, frame-based
based systems, and semantic
networks.

Rule-based systems use a set of rules that describe the relationships between objects
in the world. These rules are usually in the form of "if
"if-then"
then" statements that define the
conditions under which a particular action should be taken. For example, a rule-based
rule
system
em for identifying objects in an image might use rules that specify the shape and
color of the object.

Frame-based systems represent knowledge using a hierarchical structure of frames that

define the properties and relationships of objects in the world. EaEachch frame contains slots
that represent the attributes of the object, such as its size, color, and shape. Frames can
also have inheritance relationships, where one frame inherits the properties of another.

Semantic networks represent knowledge using a graph structure that defines the
relationships between objects in the world. Nodes in the graph represent concepts, and
edges represent the relationships between them. For example, a semantic network for

YouTube Channel: RGPV

PV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial
identifying objects might have nodes for "chair," "table," and "desk," with edges that
indicate that a chair can be found in a room.

Control Strategies Control strategies refer to the methods used to control the flow of
information in a knowledge-based vision system. In particular, control strategies
determine how knowledge is used to guide the interpretation of visual information.
There are various control strategies used in knowledge-based vision, including goal-
driven control, data-driven control, and hybrid control.

Goal-driven control involves using high-level goals to guide the interpretation of visual
information. For example, a goal-driven system for object recognition might have a goal
to identify all the objects in an image. The system would then use knowledge of object
properties to guide the interpretation of the image.

Data-driven control involves using the visual information itself to guide the
interpretation process. In a data-driven system, the computer first extracts features from
the image, such as color, texture, and shape. The system then uses this information to
guide the interpretation of the image.

Hybrid control combines both goal-driven and data-driven approaches. In a hybrid

system, the computer first extracts features from the image, and then uses this
information to guide the interpretation process based on high-level goals.

Information Integration Information integration refers to the process of combining

multiple sources of information to improve the interpretation of visual information. In a
knowledge-based vision system, information integration can involve combining
knowledge from different sources, such as visual knowledge and contextual knowledge.

Visual knowledge refers to the knowledge of the visual properties of objects, such as
their shape, color, and texture. Contextual knowledge refers to the knowledge of the
context in which the visual information is presented. For example, contextual knowledge
might include information about the lighting conditions, the location of the objects, and
the relationships between the objects.

To integrate information from different sources, a knowledge-based vision system might

use techniques such as Bayesian inference, fuzzy logic, or neural networks. These
techniques allow the system to combine information in a way that accounts for
uncertainty and ambiguity.

In conclusion, knowledge-based vision is a powerful approach to computer vision that

combines knowledge representation, control strategies, and information integration. By

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial

 integrating knowledge and reasoning with visual information, knowledge-based vision
systems can achieve a deeper

Explain Object recognition-Hough transforms and other simple object recognition methods
Object recognition is a process in computer vision that involves identifying objects in
an image or a video stream. It is a challenging problem due to variations in lighting
conditions, object pose, occlusions, and background clutter. In this article, we will
discuss two simple object recognition methods: Hough transforms and template
matching.

Hough Transforms Hough transforms is a popular method for detecting simple shapes,
such as lines and circles, in an image. The method was developed by Paul Hough in
1962 and has since been extended to detect other shapes such as ellipses and
rectangles.

The basic idea behind Hough transforms is to transform the image space into a
parameter space, where each point in the parameter space represents a line or a circle in
the image. The transformation is achieved by mapping each pixel in the image to a
curve in the parameter space, where the curve represents all the possible lines or circles
that pass through the pixel.

To detect a line in an image using Hough transforms, the following steps are typically
performed:

1. Edge detection: Detect edges in the image using edge detection algorithms such as
Canny or Sobel.
2. Hough transform: For each edge pixel in the image, compute the set of curves that
pass through the pixel in the parameter space. Accumulate the curves in a Hough
accumulator array.
3. Peak detection: Identify the peaks in the accumulator array, which correspond to the
lines in the image.
4. Line extraction: Extract the lines corresponding to the peaks in the accumulator array
and draw them on the image.

Template Matching Template matching is a simple object recognition method that

involves comparing a template image with the target image to find a match. The
template image is a small image that represents the object to be detected, and the
target image is the larger image in which the object is to be found.

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial

 The template matching process involves the following steps:

1. Template creation: Create a template image of the object to be detected.

2. Image preprocessing: Preprocess the target image to improve the matching process.
This may include operations such as scaling, rotation, and filtering.
3. Matching: Slide the template image over the target image and compare the pixel values
of the two images at each location. Compute a similarity measure, such as correlation or
sum of squared differences, to determine how well the template matches the target
image at each location.
4. Thresholding: Set a threshold value for the similarity measure and identify the locations
where the similarity measure exceeds the threshold. These locations correspond to the
object in the target image.
5. Object localization: Extract the object from the target image using the location
information obtained in step 4.

Limitations of Simple Object Recognition Methods While Hough transforms and

template matching are simple and efficient methods for object recognition, they have
some limitations. One of the main limitations is that they are sensitive to variations in
lighting conditions, object pose, and background clutter. These methods may also not
be able to handle complex objects with multiple parts or textures.

To overcome these limitations, more advanced object recognition methods have been
developed, such as feature-based methods, which extract distinctive features from the
image, and deep learning-based methods, which use convolutional neural networks to
learn features directly from the image data. These methods are capable of handling
more complex objects and are more robust to variations in the image data.

Explain Shape correspondence and shape matching, Principal component analysis , feature
extraction
Shape correspondence and shape matching, principal component analysis, and feature
extraction are important concepts in computer vision that are used to identify and
match objects in an image or video stream. In this article, we will discuss these concepts
in detail.

Shape Correspondence and Shape Matching Shape correspondence is the process of

finding the correspondence between the shapes of two objects. This is an important
step in shape matching, which is the process of finding the object in an image or video
stream that matches a given shape.

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial

One approach to shape correspondence is to use geometric features, such as corners or
edges, to represent the shape of the object. These features can then be compared
between two objects to find their correspondence. Another approach is to use shape
descriptors, such as Fourier descriptors or shape context, which capture the overall
shape of the object and its distribution of features.

Shape matching algorithms typically involve finding the correspondence between the
shapes of two objects and then computing a similarity measure between them. Some
popular similarity measures include Euclidean distance, Hausdoff distance, and the
Chamfer distance.

Principal Component Analysis Principal component analysis (PCA) is a statistical

technique used to analyze and reduce the dimensionality of a dataset. In computer
vision, PCA is often used for feature extraction, which involves identifying the most
important features in an image or video stream.

PCA works by finding the principal components of a dataset, which are the directions of
greatest variation in the data. These principal components are used to create a new set
of features that capture the most important information in the data.

In computer vision, PCA can be used for image compression, object recognition, and
face recognition. In image compression, PCA is used to reduce the dimensionality of the
image data, which can lead to significant reductions in storage space and processing
time. In object recognition and face recognition, PCA is used to extract features from the
image data, which are then used to identify and match objects in the image.

Feature Extraction Feature extraction is the process of identifying and extracting

important information from an image or video stream. Features can be defined as
distinctive attributes or characteristics of an object, such as edges, corners, or color
histograms.

Feature extraction algorithms typically involve identifying salient points or regions in the
image, and then computing a set of features at each point or region. These features can
then be used to identify and match objects in the image or video stream.

Some popular feature extraction algorithms include SIFT (scale-invariant feature

transform), SURF (speeded up robust features), and HOG (histogram of oriented
gradients). These algorithms are often used in object recognition, image retrieval, and
face recognition applications.

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial

Conclusion Shape correspondence and shape matching, principal component analysis,
and feature extraction are important concepts in computer vision that are used to
identify and match objects in an image or video stream. These concepts are
fundamental to many computer vision applications, including object recognition, image
retrieval, and face recognition.

Explain , Neural network and Machine learning for image shape recognition
Neural networks and machine learning are powerful techniques for image shape
recognition. In this article, we will discuss how these techniques work and their
applications in computer vision.

Neural Networks for Image Shape Recognition A neural network is a computational

model that is inspired by the structure and function of the human brain. It consists of
interconnected nodes, or neurons, that work together to perform a specific task, such as
image classification.

In image shape recognition, a neural network can be trained to recognize specific

shapes or patterns in an image. The network is fed a large number of training examples,
each with a label indicating the shape or pattern present in the image. The network
learns to recognize the shape or pattern by adjusting its weights and biases, which are
parameters that determine the output of each neuron.

Once the network is trained, it can be used to classify new images by feeding them into
the network and analyzing the output. Neural networks have been successfully used in a
variety of image recognition tasks, including object recognition, face recognition, and
handwriting recognition.

Machine Learning for Image Shape Recognition Machine learning is a broader term
that encompasses a range of techniques used to enable machines to learn from data
without being explicitly programmed. In computer vision, machine learning is often used
to recognize shapes and patterns in images.

One popular machine learning technique for image shape recognition is supervised
learning, which involves training a model using a set of labeled training examples. The
model learns to recognize the shape or pattern in the image by analyzing the features of
the image and the corresponding label.

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial

Another machine learning technique for image shape recognition is unsupervised
learning, which involves training a model without any labeled training examples. The
model learns to recognize patterns in the data by analyzing the features of the images
and identifying similarities and differences between them.

Machine learning techniques have been used in a wide range of image recognition
tasks, including object recognition, face recognition, and image segmentation.

Conclusion Neural networks and machine learning are powerful techniques for image
shape recognition. They have been successfully used in a variety of computer vision
applications, including object recognition, face recognition, and image segmentation. As
the field of computer vision continues to evolve, we can expect to see even more
advanced techniques being developed to improve image shape recognition and other
related tasks.

YouTube Channel: RGPV Exam official 2023 https://www.youtube.com/@RGPVExamOfficial