Scribd
Upload
15 views15 pages

Pattern and Pattern Classes 17

The document discusses two main types of pattern arrangements in image pattern classification: quantitative patterns, represented as pattern vectors, and structural patterns, which utilize symbols arranged in various forms. It explains how pattern vectors can represent features of images and highlights the importance of tightly grouped and well-separated classes for better recognition. Additionally, it covers the use of high-level descriptors and structural methods for representing objects with symbolic features, emphasizing their role in shape detail and hierarchical structure analysis.

Uploaded by

Inchara Vaishnavar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
15 views15 pages

Pattern and Pattern Classes 17

The document discusses two main types of pattern arrangements in image pattern classification: quantitative patterns, represented as pattern vectors, and structural patterns, which utilize symbols arranged in various forms. It explains how pattern vectors can represent features of images and highlights the importance of tightly grouped and well-separated classes for better recognition. Additionally, it covers the use of high-level descriptors and structural methods for representing objects with symbolic features, emphasizing their role in shape detail and hierarchical structure analysis.

Uploaded by

Inchara Vaishnavar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 15

PATTERN AND PATTERN

CLASSES

K R INCHARA
2GP22CS017
INTRODUCTION

 In image pattern classification, there are two main types of pattern arrangements:
 Quantitative patterns
 Structural patterns
 Quantitative patterns:
 Represented as pattern vectors
 Structural patterns:
 Made up of symbols
 Arranged as:
• Strings
• Trees
• Sometimes graphs
PATTERN VECTORS

 A pattern vector (e.g., x, y, z) is a column or row of numbers representing features.

 Form:

 Each component xi is a feature descriptor.

 n is the total number of features.

 A pattern vector can be written in two ways:

 Column form (vertical)

 Row form with transpose:

 x=(x1​,x2​,...,xn​)T (The T means it's a column vector.)


 Visually, a pattern vector can be thought of as a point in n-dimensional Euclidean space.
 A group of such points (same class) forms a "hyper cloud" in that space.
 For better recognition, pattern classes should be:
 Tightly grouped (close together within the same class)
 Well-separated from other classes (far apart in space)
 Pattern vectors can be made directly by flattening image pixel intensities (e.g., using linear indexing).
 Instead of raw pixels, pattern vectors are usually built from meaningful features.

fig: Using linear indexing to vectorize a grayscale image


Example – Fisher's Iris Data (1936):

 Used discriminant analysis to classify 3 iris types:


Iris setosa, virginica, versicolor
 Described each flower using 4 features:
 Petal length
 Petal width
 Sepal length
 Sepal width
 Each flower became a 4-dimensional vector.
 Collected 50 samples per class, forming 3 classic
pattern classes.
 Today, extra features (like color and shape) might
be added for more detailed vectors.
SHAPE-BASED PATTERN VECTORS:

 High-level descriptors are used for better pattern representation.


 Example: Using boundary shape descriptors in industrial inspection.
 A shape can be represented by its signature (a 1D signal).
 Sample the signature at intervals to form a vector:

xi=r(Ꝋi) where u is the sampling step.


 Instead of raw values, apply a function g to enhance features:

xi=g(r(Ꝋi))
 Vectors from Boundary and Region Features:

• Objects can be represented by vectors


• combining boundary and region features.
• Example: 3-D vectors capture shape info of binary objects

 Vectors for Image Regions:


• Pattern vectors can describe image regions.
• Example: 6-D vectors use texture features
• In Fig. 12.5 are texture measures based on
the feature descriptors in Table 11.3.
 Invariant Feature Vectors:

• Some pattern vectors use features invariant to transformations like rotation and scaling
 Vectors from Registered Image Sequences:
•Pattern vectors can be created from corresponding pixels across image sequences (Fig. 12.7).
•Useful for tasks like multispectral image recognition .
 High-Dimensional Vectors for Whole Images:
• Full images may require high-dimensional vectors (e.g., using SIFT features, Section 11.7).
 Deep Learning Approach:
• A more advanced method for analyzing entire images is using deep convolutional neural
networks (Sections 12.5 and 12.6).
STRUCTURAL PATTERNS:

 Pattern vectors are not ideal for representing objects described by symbolic or structural features,
such as shapes made of lines or symbolic sequences.
 In structural pattern recognition, shape and spatial relationships play a key role rather than just
numeric values.
 Structural methods are used when shape detail and hierarchical structure are important for
interpretation or recognition.
Figure 12.8 – Symbolic String Representation of Shapes

 Objects (e.g., bottles) are outlined and approximated using polygonal boundaries.
 The boundary is split into line segments, and the interior angles at their junctions are
computed.
 As the boundary is traversed (e.g., counterclockwise), a sequence of symbols is generated.
 This sequence is a structural pattern, which can be compared with other symbol strings for
recognition.
 This method is ideal for shape-based object comparison.
Figure 12.9 – Tree-Based Structural Representation

 A tree is a symbolic way to represent an image by breaking it down into component regions.
 The root node ($) represents the entire image.
 Using the relationship "composed of", the image is split into downtown and residential areas.
 These areas are further divided:
 Residential areas → housing, highways, shopping malls
 Housing and highways can be further subdivided for more detail.
 This hierarchical breakdown continues until regions become too small to resolve.
 Such tree structures help in analyzing complex scenes and are useful for tasks like image
understanding or semantic segmentation.
Thank you…

You might also like