UNIT IV

IMAGE ANALYSIS AND PATTERN RECOGNIZATION

 Spatial feature extraction

 Edge detection
 Boundary extraction and representation
 Region and moment Representation
 Structure – Texture
 Scene matching and detection
 Image segmentation
 Classification techniques
 Pattern recognition
 Perceptron for two pattern classes
 Training algorithm
 Non separable classes.
SPATIAL FEATURE EXTRACTION

Amplitude features: e.g. the brightness levels can identify regions of interest in the
image:
Amplitude features may be discriminative enough if intensity is enough to distinguish
wanted info from the rest of the scene
 => defining the best parameters of the transformation for feature extraction – most
difficult
=> amplitude feature space representation is not necessarily binary; just that unwanted
parts of the scenes should be represented uniquely (i.e. black) in the feature space
=> sometimes adaptive thresholding/adaptive grey scale slicing is needed.

Histogram based features

Local histogram = a local statistical description of the image;

If u = an image pixel; x=a grey level => pu(x)=the probability of appearance of the grey

level x in the image region = a value in the normalized histogram

=> One can compute: the standard deviation; the entropy; the median; percentiles, of
pu(x).

Region Measurements
of interest
(ROI) ROI histogram
 Tissue of interest is well
discriminated from the
microscopic slide by the
standard deviation of the
local histogram

TRANSFORM FEATURES
EDGE DETECTION

 Edge detection significantly reduces the amount of data and filters out
useless information, while preserving the important structural
properties in an image.
 Edges are boundaries between different textures.
 Edge also can be defined as discontinuities in image intensity from one
pixel to another.
 The edges for an image are always the important characteristics that offer an
indication for a higher frequency.
 Detection of edges for an image may help for image segmentation, data
compression, and also help for well matching, such as image
reconstruction and so on.
 Edge detection is difficult in noisy images, since both the noise and the
edges contain high-frequency content.
Edge Detectors

 Robert
 Sobel
 Prewitt
 Laplacian and
 Canny
Sobel Edge Detector

 The salient features of the sobel edge detectors are listed as follows
 It has two 3x3 convolution kernels or masks, Gx and Gy, as shown in fig 1.
both Gx and Gy can be joined together to find the absolute magnitude and
the orientation of the gradient.
  Used to detect edges along the horizontal and vertical axis
 Based on convolving the image with a small, integer valued filter (3×3 kernel)
in both horizontal and vertical direction. So this detector requires less
computation
 The sobel edge detection masks search for edges in horizontal and vertical
directions and then combine this information into a single metric
 In this, image intensity is calculate at every pixel (pixel) and presenting the
direction of the maximum possible change from white (light) to black (dark)
and the rate of change in that direction.
 Simplicity
 Detection of edges and their orientations

 Sensitivity to noise
 Inaccurate

PREWITT EDGE DETECTOR

The salient features of the Prewitt Edge Detector are listed as follows
 This edge detector is very similar of sobel operator
 Simplicity
 Detection of horizontal and vertical edges and their orientations
 Sensitivity to noise
 Inaccurate
 The kernel used in the Prewitt detector is shown in fig 2.
LAPLACIAN OF GAUSSIAN EDGE DETECTOR

The salient features of the Laplacian of Gaussian edge detector is listed as follows
 Proposed by Marr and Hildreth in 1980
 This is a combination of the Gaussian filtering and Laplacian gradient
operator.
 Laplacian gradient operator determines the regions where the rapid
intensity changes. So it is best suit for edge detection.
 After the laplacian process is over, the image is given to Gaussian filter to
remove the noise pixels.
 The laplacian gradient of an image is given by

Where I(x, y) = pixel intensity values in image

• In this, an image is divided where the intensity varies to detect the edges
effectively.
• It is very difficult to find the orientation of edges due to laplacian filter.
• Used to determine exact location of edges
• Does not produce good result where the gray level function varies (corners,
curves)
• Not useful for finding the orientation of edges
• The two 3X3 kernels used for laplacian edge detector is shown in fig 3.
ROBERTS EDGE DETECTOR

The salient features of the Roberts Edge Detector is listed as follows

 This detector has two 2x2 convolution kernels in which one of the kernels is
0
rotated by 90 .
 Fast computation
 Performs a simple and fast two dimensional spatial gradient measurement on
an image.
 Each point (pixels) in the output image represents the expected absolute
magnitude of the spatial gradient of the input Image at that point.
 Convolution mask is shown in fig 4.
CANNY EDGE DETECTOR

 The canny edge detector can be used to identify a wide range of real edges
in images.
 The detector eliminates the unwanted noise pixels by the process of
smoothening edges in an image because noise pixels generate false edges. \
 In this edge detection, the signal to noise ratio is improved as compared to
other methods.
 This is the reason why this detector is extensively used for edge detection in
image processing.
The procedure to find edges in images is explained as follows.

 Initially the image is smoothened using a suitable filter such as mean filter,
Gaussian filter etc., to reduce the effect of noise.

 Then local gradient and edge direction is calculated for every point. This
point has a maximum strength in the direction of the gradient.

 These edge points give rise to ridges in the gradient magnitude image.
  The edge detector tracks along the top of these ridges and make all the
pixels to zero that are not actually on the top of the ridge. So a thin line is
generated in the output.

 These ridge pixels are threshold using two threshold values: upper threshold
(T2) and lower threshold (T1).
 Ridge pixels are classified as strong edge pixels if ridge pixel values are
greater than upper threshold (T2) and ridge pixels are classified as weak
edge pixels if ridge pixel values are between the lower threshold (T1) and the
upper threshold (T2).
 Finally, the edges in the image are linked by integrating the weak pixels
which are connected to the strong pixels.

EDGE DETECTORS EXAMPLES

COMPARISON OF EDGE DETECTORS
BOUNDARY EXTRACTION AND REPRESENTATION
Boundary Extraction and Representation using Hough Transform
Boundary Representation
TEXTURE-BASED TECHNIQUES

What is Texture?
No one exactly knows.
In the visual arts, texture is the perceived surface quality of an artwork.
 "Texture" is an ambiguous word and in the context of texture synthesis may
have one of the following meanings:
 1. In common speech, "texture" used as a synonym for "surface
structure".
 Texture has been described by five different properties in the
psychology of perception: coarseness, contrast, directionality, line-
likeness and roughness [1].
 2. In 3D computer graphics, a texture is a digital image applied to the surface
of a three-dimensional model by texture mapping to give the model a more
realistic appearance. Often, the image is a photograph of a "real" texture,
such as wood grain.
 3. In image processing, every digital image composed of repeated
elements is called a "texture."
 Stochastic textures. Texture images of stochastic textures look like noise: colour
dots that are randomly scattered over the image, barely specified by the attributes
minimum and maximum brightness and average colour. Many textures look like
stochastic textures when viewed from a distance. An example of a stochastic texture
is roughcast.

 · Structured textures. These textures look like somewhat regular patterns. An

example of a structured texture is a stonewall or a floor tiled with paving stones.

Artificial Natural
textures textures
STRUCTURE

MORPHOLOGICAL OPERATIONS

 Morphology is a broad set of image processing operations that process

images based on shapes.
 Morphological operations apply a structuring element to an input image,
creating an output image of the same size.
 The most basic morphological operations are dilation and erosion.
 In a morphological operation, the value of each pixel in the output image is
based on a comparison of the corresponding pixel in the input image with its
neighbors.
 Dilation and erosion are two fundamental morphological operations.
 Dilation adds pixels to the boundaries of objects in an image, while erosion
removes pixels on object boundaries.
 The number of pixels added or removed from the objects in an image depends
on the size and shape of the structuring element used to process the image.
 In the morphological dilation and erosion operations, the state of any given pixel
in the output image is determined by applying a rule to the corresponding pixel
and its neighbors in the input image.
 


Skeletonization
 To reduce all objects in an image to lines, without changing the essential
structure of the image, use the bwmorph function. This process is known as
skeletonization.
SCENE MATCHING AND DETECTION

Image Subtraction
 Subtract one image from another or subtract constant from image
 Syntax Z= imsubtract(X,Y)
 Description
 Z = imsubtract (X,Y) subtracts each element in array Y from the
corresponding element in array X and returns the difference in the
corresponding element of the output array Z.
SCENE MATCHING AND DETECTION

IMAGE SEGMENTATION

There are many definitions:

• Segmentation subdivides an image into its constituent regions or objects (

Gonzales, pp567)
• Segmentation is a process of grouping together pixels that have similar
attributes ( Efford , pp250)
 • Image Segmentation is the process of partitioning an image into non-
intersecting regions such that each region is homogeneous and the union
of no two adjacent regions is homogeneous ( Pal, pp1277)
• Segmentation is typically associated with pattern recognition problems. It is
considered the first phase of a pattern recognition process and is sometimes also
referred to as object isolation.

Why segmentation is difficult?

It can be difficult for many reasons:

• Non- uniform illumination

• No control of the environment
• Inadequate model of the object of interest
• Noise

Why segmentation is useful?

Segmentation algorithms have been used for a variety of applications. Some

examples are :
• Optical character recognition(OCR)
• Automatic Target Acquisition
• Colorization of Motion Pictures
• Detection and measurement of bone, tissue, etc, in medical images.

Overview of Segmentation Techniques

CLASSIFICATION TECHNIQUES
PATTERN RECOGNITION

“The assignment of a physical object or event to one of several prespecified categeries”

-- Duda & Hart
 A pattern is an object, process or event that can be given a name.
 A pattern class (or category) is a set of patterns sharing common attributes
and usually originating from the same source.
 During recognition (or classification) given objects are assigned to
prescribed classes.
 A classifier is a machine which performs classification.
Applications

Components of PR system

• Sensors and preprocessing.

• A feature extraction aims to create discriminative features good for
classification.
• A classifier.
• A teacher provides information about hidden state -- supervised learning.
• A learning algorithm sets PR from training examples.
 Basic components of a pattern recognition system
 Data acquisition and sensing
 Pre-processing
 Removal of noise in data.
 Isolation of patterns of interest from the background.
 Feature extraction
 Finding a new representation in terms of features.

(Better for further processing)

 Model learning and estimation
 Learning a mapping between features and pattern groups.
 Classification
 Using learned models to assign a pattern to a predefined
category
 Post-processing
 Evaluation of confidence in decisions.
  Exploitation of context to improve performances.

Examples of pattern recognition applications

Feature extraction

Task: to extract features which are good for classification.

Good features:
• Objects from the same class have similar feature values.
• Objects from different classes have different values.

Problem can be expressed as optimization of parameters of featrure extractor .

Supervised methods: objective function is a criterion of separability (discriminability) of

labeled examples, e.g., linear discriminat analysis (LDA).

Unsupervised methods: lower dimesional representation which preserves important

characteristics of input data is sought for, e.g., principal component analysis (PCA).
Advantages
 economical to store features in memory
 experimental evidence consistent with features
Disadvantages
 lack of applicability to a wide range of stimuli
 analysis of stimuli does not always begin with features
 treats all features as equivalent

Face Recognition
 User-friendly pattern recognition application
 Weakness of face recognition
 Illumination problems
 Pose problems( profile or frontal view)

 Eigenspace-based approach
 A holistic approach
 Reducing the high dimensionality problem , and large computational
complexity.
Approaches

 Statistical PR: based on underlying statistical model of patterns and pattern

classes.
 Structural (or syntactic) PR: pattern classes represented by means of formal
structures as grammars, automata, strings, etc.
 Neural networks: classifier is represented as a network of cells modeling
neurons of the human brain (connectionist approach).
 Template matching is simple to implement but the template size must be small to
decrease computational delay.
 Statistical methods highly depend on the assumption of distribution.
 Neural networks can adaptively refine the classifier and the decision surface in
principle can be arbitrarily implemented.
 Syntactic methods concerned structural sense to encode but additional process
to define primitives is required.
2 Mark Questions

1. What is meant by spatial feature extraction?

2. Write notes on Edge detection.
3. List out the various edge detectors.
4. Write notes on Boundary extraction and representation
5. What is meant by Structure and Texture?
6. Write notes on Scene matching and detection.
7. What is meant by Image segmentation?
8. What is meant by supervised and unsupervised classification?
9. What is meant by Pattern recognition?
10. List out the applications of Pattern recognition.

12 Mark Questions

1. Explain about various spatial feature extraction methods.

2. Explain about edge detection techniques.
3. Explain about Boundary extraction and representation.
4. Explain (a) Scene matching and detection (b) Image segmentation
5. Explain about various classification techniques
6. Explain Pattern recognition with a neat sketch.