Image Segmentation – Complete & Easy Notes
Learning Outcomes
After studying this chapter, students will be able to understand image
segmentation, explain its importance, describe different segmentation
approaches, and explain common segmentation techniques in simple and
exam-oriented words.
What is Image Segmentation?
Image segmentation is the process of dividing an image into meaningful
regions or parts. Each region contains pixels with similar properties such
as intensity, color, or texture. The main purpose of segmentation is to
make image analysis easier by separating objects from the background.
Why Image Segmentation is Needed?
Image segmentation is needed to identify and locate objects in an image.
It helps in simplifying image representation so that further processing
becomes easy. Segmentation is widely used in medical imaging, object
detection, face recognition, and computer vision applications.
Basic Idea of Segmentation
Segmentation works on the idea that neighboring pixels in the same
object have similar properties. Pixels that are very different usually belong
to different objects. Based on this similarity or difference, the image is
divided into regions.
Types of Image Segmentation Approaches
Image segmentation techniques are generally divided into two main
approaches: region-based methods and boundary-based methods. Each
approach uses a different strategy to separate objects in an image.
Region Based Segmentation
Region based segmentation divides an image into regions based on
similarity between neighboring pixels. Pixels having similar intensity,
color, or texture are grouped together. This approach produces connected
and meaningful regions and is less sensitive to noise compared to edge-
based methods.
Common Region Based Methods
Thresholding Based Segmentation
Thresholding is the simplest segmentation technique. It separates objects
from the background based on pixel intensity values. Pixels above a
threshold belong to one region, and pixels below belong to another region.
It works well for images with clear contrast.
Global Thresholding
In global thresholding, a single threshold value is used for the entire
image. All pixels are compared with this value. This method is simple but
not effective for images with uneven lighting.
Multiple Global Thresholding in Image Processing
Multiple global thresholding is a segmentation technique that uses more
than one threshold value to divide an image into multiple regions. Instead
of separating only foreground and background, it segments the image into
several intensity levels. All thresholds are applied globally to the entire
image. It is useful for images containing multiple objects with different
intensity ranges and works well when gray levels are clearly separated.
Region Growing
Region growing is a region based segmentation technique that starts with
one or more seed pixels. Neighboring pixels are added to the region if
they are similar to the seed pixel based on a similarity condition. The
process continues until no more pixels satisfy the condition.
Region Merging
Region merging combines adjacent small regions into larger regions based
on similarity. If neighboring regions satisfy a similarity condition, they are
merged into one region. This method reduces over-segmentation and
produces meaningful regions.
Region Splitting and Merging
Region splitting and merging is a segmentation technique that divides an
image into smaller regions and then combines similar regions. Splitting is
done when a region is not homogeneous, and merging is done when
neighboring regions are similar. This method ensures meaningful and
connected regions.
Heuristic in Splitting and Merging (Important)
A heuristic is a rule or condition used to decide whether to split or merge
regions. It is usually based on similarity of intensity, color, or texture. If a
region satisfies the homogeneity condition, it is not split; otherwise, it is
divided further. Similarly, regions are merged if they satisfy the heuristic
condition.
Watershed Segmentation
Watershed segmentation treats the image as a topographic surface.
Bright areas represent peaks and dark areas represent valleys. It is useful
for separating touching or overlapping objects.
Applications of Image Segmentation
Image segmentation is used in medical diagnosis, satellite image analysis,
traffic monitoring, object detection, and face recognition. It plays a key
role in computer vision systems.
Advantages of Image Segmentation
Image segmentation simplifies image analysis. It helps in accurate object
detection and feature extraction. It improves the performance of image
processing systems.
Limitations of Image Segmentation
Segmentation results depend on image quality and method used. Noise
and poor contrast can affect accuracy. No single segmentation method
works best for all images.
Classical Clustering Algorithms (Image Processing)
Classical clustering algorithms are traditional unsupervised methods
used to group image pixels based on similarity. These algorithms do not
require labeled data and are widely used in image segmentation. Pixels
are clustered using features such as intensity, color, or texture
Common Classical Clustering Algorithms
K-Means Clustering: Divides pixels into K clusters based on
distance from cluster means.
Hierarchical Clustering: Builds clusters step by step by merging
or splitting based on similarity.
Fuzzy C-Means: Allows a pixel to belong to more than one cluster
with different membership values.
Mean-Shift Clustering: Groups pixels by finding dense regions in
feature space.
Least Square Error (LSE) Measure in Clustering
The least square error (LSE) measure is used in clustering to evaluate
how well pixels fit into their assigned clusters. It calculates the sum of
squared distances between each data point (pixel) and the mean
(centroid) of its cluster. The goal of clustering algorithms like K-means is
to minimize this error so that pixels in the same cluster are as similar as
possible.
Formula (Exam-important):
K
LSE=∑ ∑ ∣∣ x−¿ μ k ∣ ∣2 ¿
k=1 x ∈C k
Advantages of Clustering Based Segmentation
Clustering does not require prior knowledge or labeled data. It works well
for complex images and color images. It is simple to understand and
widely used in practical applications.
Limitations of Clustering Based Segmentation
The result depends on the value of K. It may converge to a local minimum.
Noise and poor initialization can affect segmentation quality.