IMAGE SEGMENTATION TECHNIQUES

Chhavi Jajoo
Asmita Jain
Shubham Kumar

Guided By :- Dr. Biplab Banerjee

ABSTRACT perceptual completion procedures will be required. Some

researchers argue that segmentation algorithms should be
Image segmentation can be defined as in which we divide evaluated in the context of a particular task, such as object
the image into multiple parts in the form of pixels. In recognition, that is, different algorithms should be compared
segmentation, we simply represent the image into more in terms of the potential benefit they provide for a particular
understandable form. Segmentation basically used to detect higher-level task. Other researchers propose that
the objects, boundaries and other relevant data in the digital segmentation algorithms should be evaluated as stand-alone
images. There are different approaches to implement modules, by comparing their output to ground truth which
segmentation like threshold, clustering and transform is usually a segmentation produced by human observers. .[3]
methods etc. After performing these approaches, the Recent research by Martin et al indicates that human
resultant segmented image is a collective pixel set of the segmentations do not vary randomly, instead they show
entire image. Pixels in the image corresponds to some regularities that can be exploited to design and evaluate
characteristics of image like colour, texture etc.[4] segmentation algorithms[1]. It also suggests ways in which
the use of higher level knowledge by human observers can
be accounted for, thus allowing for the direct comparison of
1. INTRODUCTION segmentations produced by human observers and
segmentation algorithms.
For a computer images are not objects, or persons but
pixels. Pixels in simple words are the smallest controllable 3. IMAGE SEGMENTATION
element of a picture. Pixels can be made of multiple
components (RGB) or just contain a single color The goal of image segmentation is to cluster pixels into
(Grayscale.) Now how a computer understands the image as salient image regions, i.e., regions corresponding to
a whole is another story. individual surfaces, objects, or natural parts of objects.
In imaging science, image processing is processing of Segmentations of simple gray-level images can provide
images using mathematical operations by using any form useful information about the surfaces in the scene.
of signal processing for which the input is an image, a series
of images, or a video, such as a photograph or video frame; 3.1 Region Growing
the output of image processing may be either an image or a
set of characteristics or parameters related to the Region growing is a simple region-based image
image.]Most image-processing techniques involve treating segmentation method. It is also classified as a pixel-based
the image as a two-dimensional signal and applying standard image segmentation method since it involves the selection of
signal-processing techniques to it. Images are also processed initial seed points.
as three-dimensional signals with the third-dimension being 3.1.1 Single Seeded
time or the z-axis.
This approach to segmentation examines neighboring pixels
2. RELATED WORKS of initial seed points and determines whether the pixel
neighbors should be added to the region. The process is
The process of segmentation is directly tied to recognition, iterated on, in the same manner as general data
and it is likely that to achieve a complete separation of clustering algorithms.
objects from background, information that can only be The first step in region growing is to select a set of seed
obtained from higher level recognition, inference, and points. Seed point selection is based on some user criterion
(for example, pixels in a certain grayscale range, pixels
evenly spaced on a grid, etc.). The initial region begins as A statistical method of examining texture that considers the
the exact location of these seeds. The regions are then grown spatial relationship of pixels is the gray-level co-occurrence
from these seed points to adjacent points depending on a matrix (GLCM), also known as the gray-level spatial
region membership criterion. The criterion could be, for dependence matrix. For the program blocks of 3x3 pixels
example, pixel intensity, grayscale texture, or color. were taken to form a patch. All the glcm properties are
calculated on this patch and are used for the texture analysis.
After that the same algorithm for single seeded region
growing is applied by taking a particular threshold of the
image and comparing the patch values to these thresholds.

3.1.2 Multi Seeded

As, can be seen in the IMG 1 the result we get from a single
seeded region growing is not satisfactory enough. It
basically segments the image in two parts with black and 3.1.4 Multiple patch witch GLCM
white regions. So to enhance the quality of the output, the
same region growing technique is applied with multiple The output in the case of single seeded region growing with
seeds. In this concept instead of taking one random seed, a glcm is not satisfactory enough and not much of the
number of them are taken. The more the number of seeds information is derived, so to improve the result multiple
initially, the better is the output image but the output after a patches are taken and multi seeded region growing is applied
particular number of seeds basically remains the same and is incorporating the 10 values three RGB values of the pixel,
the optimum output. three RGB values of the patch and the four glcm properties
To take multiple seeds we first need to draw a histogram, as (contrast, correlation, homogeneity and energy).
the seeds that are taken from different regions, i.e. from or

3.1.5 Single Seeded with HSV

Again to enhance the output of the image, same single

seeded algorithm is applied but including the HSV features
of the image. Much change in the output is not seen, but the
still the one with HSV features included is better than the
previous one.

near the modes of the histogram would give a better output

rather than just some random seeds from any part of the
image. And so the different seeds are taken and region
growing is applied.

3.1.3 Single patch with GLCM

 In this paper, we have discussed about the image
segmentation, the various techniques of it and image
engineering. These techniques are applicable in different
fields like medical imaging, object recognition, pattern
recognition etc. by studying this topic in depth, I got to
know that, image segmentation is having vital use and
challenging future in image processing.
3.1.6 Multiple seeded with HSV

HSV is applied with taking multiple seeds as input. Only 5. REFERENCES

difference is that the HSV features of the image are
[1] W. X. Kang, Q. Q. Yang, R. R. Liang, The
included.
Comparative Research on Image Segmentation
Algorithms, IEEE Conference on ETCS, pp. 703-707,
2009.
[2] Zhang, Y. J, An Overview of Image and Video
Segmentation in the last 40 years, Proceedings of the
6th International Symposium on Signal Processing and
Its Applications, pp. 144-151, 2001
[3] Previous Work in Image Segmentation (Chapter 2).
www.cs.utoronto.ca/~strider/publications/Chapter2.pdf
[4] International Journal of Science and Research (IJSR),
3.1.7 Single patch with GLCM and HSV Amanpreet Kaur, M. Tech CSE, Global Institute of
Management and Technology, Amritsar, India
A patch of 3x3 pixels is taken and the mean RGB values are
calculated, and the four GLCM properties as well are
calculated. Entire procedure is same but this time the HSV
features are also added.

3.1.8 Multiple patch with GLCM and HSV

The same multiple patch growing technique is applied but

now the HSV features are also included.

4. CONCLUSION