MACHINE LEARNING PROJECT

TITLE: IMAGE SEGMENTATION

PROJECT IDEA:
The goal of segmentation is to simplify and/or change the representation of an image into
something that is more meaningful and easier to analyze. Image segmentation is typically used to
locate objects and boundaries (lines, curves, etc.) in images. More precisely, image segmentation is
the process of assigning a label to every pixel in an image such that pixels with the same label share
certain characteristics. The result of image segmentation is a set of segments that collectively cover
the entire image, or a set of contours extracted from the image.
METHOD-1: Image Segmentation by Using Threshold Techniques
RESEARCH PAPER READ:
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 5, MAY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
ABSTRACT:
This paper attempts to undertake the study of segmentation image techniques by using five
threshold methods as Mean method, P-tile method, Histogram Dependent Technique (HDT), Edge
Maximization Technique (EMT) and visual Technique and they are compared with one another so
as to choose the best technique for threshold segmentation techniques image. These techniques
applied on three satellite images to choose base guesses for threshold segmentation image.
KEY POINTS:

Local techniques are based on the local properties of the pixels and their neighbourhoods.
Global techniques segment an image on the basis of information obtain globally (e.g. by
using image histogram; global texture properties ).
Split, merge and growing techniques use both the notions of homogeneity and geometrical
proximity in order to obtain good segmentation results.
Finally image segmentation, a field of image analysis, is used to group pixels into
regions to determine an images composition.
Threshold techniques can be categorized into two classes: global threshold and local
(adaptive) threshold.
1. In the global threshold, a single threshold value is used in the whole image.
2. In the local threshold, a threshold value is assigned to each pixel to determine whether it
belongs to the foreground or the background pixel using local information around the
pixel.

LEARNT:

Threshold technique is one of the important techniques in image segmentation.

This technique can be expressed as:
T=T[x, y, p(x, y), f(x, y]
(1)

Where: T is the threshold value.

x, y are the coordinates of the threshold value point.
p(x,y) ,f(x,y) are points the gray level image pixels.
Threshold image g(x,y) can be define:

METHOD-2: Image Segmentation by Using Digital Techniques

RESEARCH PAPER READ:
International Journal of Computer Science & Information Technology (IJCSIT) Vol 3, No 5, Oct
2011
ABSTRACT:
The problems of digital image segmentation represent great challenges for computer vision. The
wide range of the problems of computer vision may make good use of image segmentation. This
paper study and evaluate the different methods for segmentation techniques. It discuss the main
tendency of each algorithm with their applications, advantages and disadvantages. This study is
useful for determining the appropriate use of the image segmentation methods and for improving
their accuracy and performance and also for the main objective, which designing new algorithms.
KEY POINTS:

In digital image processing, we use computer algorithms to perform image processing.

Digital image processing has several advantages over the analog image processing; first it
gives a high number of algorithms to be used with the input data, second it can avoid some
processing problems such as creating noise and signal distortion during signal processing.
The goal of image segmentation process is partitioning the image into regions.
The purpose of image segmentation is to cluster pixels of an image into image regions.
We can use segmentation for image compression, object recognition, and image editing
processing.

LEARNT:
The segmentation of an image I, which represent a set of pixels is partitioning into n disjoint
sets R1,R2, . . . , Rn, called segments or regions such that their union of all regions equals I,
I =R 1 U R 2 U..... U R n .
1. Inversion Technique
The principle of the inversion is to continuously update the muscle activity to produce a face
movement following a given face trajectory.
When the inversion had been carried out for all frames, the inverted activity was used to
generate an animation. The cost function E was the sum of the squares of the Euclidean
distances between the markers and the corresponding face model nodes:

The inversion could produce different activity patterns, depending on the initial conditions.
Constraints may be added to the inversion to limit the number of solutions.
The new positive constraint cost function E ' was redefined in the second case by:

2. Pattern Recognition Techniques

Pattern Recognition Techniques is a non-linear modelling tools and we can be used to model the
inputs and outputs relationships. The extracted information from the training set provides
important cues of the structures such as intensity, position and shape, which can be valuable
complementary information for the segmentation of test images. Active appearance models
(AAM) are statistical models of the shape of structures. Training samples are used to extract the
mean shape, mean appearance and define ranges of shape parameters.
METHOD-3: Contour Detection And Hierarchical Image Segmentation
RESEARCH PAPER READ:
Pablo Arbel aez, Member, IEEE, Michael Maire, Member, IEEE, Charless Fowlkes, Member, IEEE,
and Jitendra Malik, Fellow, IEEE.
ABSTRACT:
This paper investigates two fundamental problems in computer vision: contour detection and image
segmentation. We present state-of-the-art algorithms for both of these tasks. Our contour detector
combines multiple local cues into a globalization framework based on spectral clustering. Our
segmentation algorithm consists of generic machinery for transforming the output of any contour
detector into a hierarchical region tree. In this manner, we reduce the problem of imag segmentation
to that of contour detection. Extensive experimental evaluation demonstrates that both our contour
detection and segmentation methods significantly outperform competing algorithms. The
automatically generated hierarchical segmentations can be interactively refined by user- specified
annotations. Computation at multiple image resolutions provides a means of coupling our system to
recognition applications.
KEY POINTS:

Early approaches to contour detection aim at quantifying the presence of a boundary at a

given image location through local measurements. More recent local approaches take into
account color and texture information and make use of learning techniques for cue
combination.
An orthogonal line of work in contour detection focuses primarily on another level of
processing, globalization, that utilizes local detector output. The simplest such algorithms
link together high-gradient edge fragments in order to identify extended, smooth contours.
The Mean Shift algorithm offers an alternative clustering framework. Here, pixels are
represented in the joint spatial-range domain by concatenating their spatial coordinates.

METHOD-4: Intensity Based Segmentation

RESEARCH PAPER READ:
International Journal of Soft Computing and Engineering (IJSCE)
ISSN: 2231-2307, Volume-3, Issue-4, September 2013
ABSTRACT:
Image segmentation is the fundamental step to analyze images and extract data from them. It is the
field widely researched and still offers various challenges for the researchers. This paper tries to put
light on the basic principles on the methods used to segment an image. This paper concentrates on
the idea behind the basic methods used. Image segmentation can be broadly be categorized as semiinteractive approach and fully automatic approach and the algorithms developed lies in either of
this approaches. Image segmentation is a crucial step as it directly influences the overall success
tounderstand the image.
KEY POINTS:

An image is basically a two dimensional function of spatial coordinates, f(x, y), and
amplitude of this function at a given coordinate gives the intensity value of the image.
The image can be expressed as the product of functions of illumination and reflection.
f(x,y) = i(x,y) . r(x,y)
The threshold based segmentation techniques are inexpensive, computationally fast and can
be used in real time applications with aid of specialized hardware.

where g(x,y) is the output image; i(x,y) is the input image and t is the threshold value.
Image segmentation is a mid-level processing technique used to analyze the image and can
be defined as a processing technique used to classify or cluster an image into several disjoint
parts by grouping the pixels to form a region of homogeneity based on the pixel
characteristics like gray level, color, texture, intensity and other features.

LEARNT:
Local thresholding is also called as adaptive thresholding. In this technique the threshold value
varies over the image depending on the local characteristic of the subdivided regions in the image.
The algorithm followed for adaptive thresholding as:
1. Divide the image into subimage.
2. Choose a local threshold for subimage considered.
3. Compare the pixels in that subimage and segment the region.
4. Consider all subimages individually and choose corresponding threshold values.
5. Stop segmentation when all the subimages are processed.

In case of global thresholding the threshold value chosen remains the same for the entire
image and acts as a cutoff value.
In case of local thresholding the image is to be subdivided in to subimages and the threshold

is to be chosen depending on the properties of local pixels in that subimage.

METHOD-5: Clustering Based Methods
RESEARCH PAPER READ:
International Journal of Soft Computing and Engineering (IJSCE)
ISSN: 2231-2307, Volume-3, Issue-4, September 2013
ABSTRACT:
The objective of clustering techniques is to identify bunch in data. A cluster usually contains a
group of similar pixels that belongs to a specific region and different from other regions.
Images can be grouped based on its content. In content based clustering, grouping is done
depending on the inherited characteristics of the pixels like shape, texture etc. There are various
clustering techniques employed, the most widely used are K-means algorithm and fuzzy C-means
algorithm.
LEARNT:

K-means algorithm: This algorithm clusters the point nearest to the centroid. The centroid
is basically the average of all the points in that cluster and has coordinate as the arithmetic
mean over all points in the cluster, separately for each dimension.
This algorithm minimizes the total distance of data points to the cluster center, of the cluster
they are assigned to. Also it does not require the actual computation of distances. A
drawback of k-means algorithm is that the number of desired clusters needs to be set before.

Fuzzy C-Means [FCM] Algorithm: In this algorithm the test pixel is allowed to be
member of two or more clusters with different membership coefficient. FCM algorithm is
iterative in nature and generates fuzzy partition matrix and also requires cluster centre along
with objective function. The values for cluster centre and objective function are updated for
every single iteration and are stopped when the difference between two successive object
function values is less than some predefined threshold value.

EXAMPLE IMAGE SEGMENTATIONS:

ORIGINAL

SEGMENTED