0% found this document useful (0 votes)
90 views9 pages

DIP Unit-5

The document discusses various techniques for detecting discontinuities in image segmentation, including point, line, and edge detection, as well as combined detection methods. It also covers edge linking and boundary deduction approaches, such as local processing, Hough Transform, and graph-theoretic techniques, along with morphological watershed segmentation and thresholding methods. These techniques are essential for applications in image segmentation, object detection, and feature extraction across various fields, including medical image analysis.

Uploaded by

215 Logavani BK
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
90 views9 pages

DIP Unit-5

The document discusses various techniques for detecting discontinuities in image segmentation, including point, line, and edge detection, as well as combined detection methods. It also covers edge linking and boundary deduction approaches, such as local processing, Hough Transform, and graph-theoretic techniques, along with morphological watershed segmentation and thresholding methods. These techniques are essential for applications in image segmentation, object detection, and feature extraction across various fields, including medical image analysis.

Uploaded by

215 Logavani BK
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Detection and Discontinuities in Image Segmentation:

Discontinuities in an image, such as points, lines, and edges, often correspond to


significant features or object boundaries. Detecting and analyzing these discontinuities
is a fundamental step in image segmentation.

1. Point Detection

Point detection aims to identify isolated points of interest within an image. These points
can represent corners, intersections, or other significant features.

How it works:

●​ Calculates the gradient matrix at each pixel.


●​ Computes the Harris corner response function, which measures the change in
intensity in different directions.
●​ Points with high response values are considered corners.

Methods:

●​ Template Matching: A small template or kernel is convolved with the image.


Points with high correlation to the template are identified as potential points of
interest.
●​ Corner Detection: Algorithms like the Harris corner detector and the Shi-Tomasi
corner detector identify points with high variation in intensity in multiple
directions.
2. Line Detection

Line detection aims to identify linear structures within an image.

How it works:

●​ Parameterizes lines using the slope-intercept form or polar coordinates.


●​ For each edge pixel, votes are cast into accumulator cells corresponding to
potential lines.
●​ Peaks in the accumulator space indicate the presence of lines.

Methods:
●​ Hough Transform: This technique maps image points onto a parameter space,
where lines are represented as points. By accumulating votes in the parameter
space, lines can be detected.
●​ Line Fitting: After detecting edge points, line fitting techniques like least-squares
fitting can be used to approximate lines.
3. Edge Detection

Edge detection aims to identify abrupt changes in image intensity, which often
correspond to object boundaries.

Methods:

●​ First-Order Derivative Operators:


○​ Sobel Operator: Computes the gradient magnitude and direction at each
pixel.
○​ Prewitt Operator: Similar to the Sobel operator, but with different filter
weights.
○​ Roberts Cross Operator: Detects edges along diagonals.
●​ Second-Order Derivative Operators:
○​ Laplacian Operator: Detects zero-crossings in the second derivative of
the image.
○​ Laplacian of Gaussian (LoG): Combines Gaussian smoothing with the
Laplacian operator to reduce noise sensitivity.
●​ Canny Edge Detector: A multi-stage algorithm that includes noise reduction,
gradient calculation, non-maximum suppression, double thresholding, and edge
tracking.
4. Combined Detection

Often, a combination of these techniques can be used to achieve more robust and
accurate segmentation. For example, edges can be detected to outline objects, and
points or lines can be used to refine the segmentation or identify specific features within
the objects.

Applications

●​ Image Segmentation: Detecting edges and lines can help segment images into
meaningful regions.
●​ Object Detection: Identifying objects in images can be facilitated by detecting
their contours and features.
●​ Feature Extraction: Discontinuities can be used to extract relevant features for
image analysis and computer vision tasks.
●​ Medical Image Analysis: Detecting edges and points in medical images can aid
in diagnosis and treatment planning.
By effectively detecting and utilizing discontinuities, we can enhance the accuracy and
robustness of various image processing and computer vision applications.

Edge Linking and Boundary Deduction:

After detecting edges, the next step is to connect them into meaningful boundaries.
Here are the primary approaches:

Local Processing

Local processing involves linking edges based on the similarity of neighboring pixels.
This approach is simple but can be sensitive to noise and gaps in edges.

In this example, edges are linked based on the similarity of gradient magnitude and
direction of neighboring pixels. However, gaps in the edges can lead to incorrect linking

How it works

The process begins at an arbitrary edge point and examines neighboring points, adding those that
meet the similarity criteria to the current edge set. This continues until no more points can be linked,
prompting the selection of a new starting point to repeat the process. The algorithm concludes when
all edge points have been considered for linking. Various strategies, including relaxation labeling,
graph or tree search, and dynamic programming, have been developed to enhance edge linking
efficiency and accuracy
Global Processing via Hough Transform

The Hough Transform is a powerful technique for detecting shapes in images, including
lines, circles, and ellipses. It works by mapping image points to parameter space, where
each point represents a possible shape. By accumulating votes in parameter space, the
Hough Transform can identify dominant patterns.

Example:

line detection using Hough Transform

In this example, edge points are mapped to the Hough space, where lines are
represented by points. The peaks in the Hough space correspond to the most likely
lines in the image.

Global Processing via Graph-Theoretic Techniques

Graph-theoretic techniques represent edge pixels as nodes in a graph, and connections


between nodes represent the similarity between pixels. By applying graph algorithms,
such as minimum spanning trees or shortest path algorithms, it is possible to group
edge pixels into meaningful boundaries.

Example:
edge linking using minimum spanning tree

In this example, edge pixels are represented as nodes in a graph, and edges connect
similar pixels. A minimum spanning tree is then constructed to group the pixels into
connected components, representing the boundaries of objects.

The segmentation process by morphological watersheds:


The segmentation process by morphological watersheds is a technique primarily used
in image processing, especially for the segmentation of images.

This method is based on mathematical morphology and the concept of watersheds,


which is inspired by the natural geography of watersheds in landscape.

It is used to partition an image into distinct regions based on the topology of the image's
intensity values.
Morphological Watershed Segmentation: A Professional Overview

Core Principles:

1.​ Topographic Representation:

○​ The image is treated as a topographic surface, where pixel intensities

represent elevation.
○​ Higher intensity pixels correspond to peaks, while lower intensity pixels

represent valleys.

2.​ Marker Identification:

○​ Local minima in the image are identified as potential seed points or

markers.

○​ These markers represent the starting points for the flooding process.

3.​ Watershed Transformation:

○​ The image is flooded from the markers, simulating the flow of water

downhill.

○​ As the water level rises, it fills the basins formed by the peaks.

○​ When water from different basins meets, a watershed line is formed,

separating the basins.

4.​ Segmentation:

○​ The watershed lines represent the boundaries between the segmented

regions.

○​ Each basin corresponds to a distinct segmented object or region.

Advantages:

●​ Robustness to Noise: The watershed algorithm is relatively insensitive to noise in

the image.

●​ Handles Overlapping Objects: It can effectively segment overlapping objects,

which can be challenging for other techniques.

●​ Intuitive Interpretation: The hydrological analogy provides a clear understanding

of the segmentation process.

Challenges and Limitations:


●​ Over-Segmentation: The algorithm can be susceptible to over-segmentation,

particularly in the presence of noise or texture.

●​ Marker Selection: The choice of markers can significantly impact the

segmentation results. Incorrect marker placement can lead to inaccurate

segmentation.

●​ Computational Complexity: The watershed algorithm can be computationally

intensive, especially for large images.

Improving Watershed Segmentation:

●​ Pre-processing: Applying techniques like noise reduction and gradient

enhancement can improve segmentation results.

●​ Marker Selection: Careful selection of markers is crucial. Techniques like

morphological operations or region-based methods can be used to identify

suitable markers.

●​ Post-processing: Post-processing techniques like region merging or splitting can

be used to refine the segmentation results.

By understanding the principles and limitations of morphological watershed

segmentation, practitioners can effectively apply this technique to a wide range of image

analysis tasks, including medical image analysis, remote sensing, and object

recognition.

Thresholding in Image Segmentation

Thresholding is a simple yet effective technique for image segmentation. It involves partitioning
an image into two or more classes based on intensity values.

Basic Idea:

●​ A threshold value is chosen.


●​ Pixels with intensity values above the threshold are assigned to one class (often
considered foreground).
●​ Pixels with intensity values below the threshold are assigned to another class (often
considered background)

You might also like