Presentation Slides

A Semi supervised
Segmentation Models
for retinal Image
ABSTRACT
ABSTRACT
Image segmentation is used in many areas including
computer vision, computer graphics, medical imaging

and so on.
Our project deals with the segmentation of retinal
image based on photometric properties.
This is achieved by a semi supervised optimization
model that determines an efficient segmentation of
input image
INTRODUCTION
WHAT IS SEMISUPERVISED MODEL?

It is a technique that makes use of both labeled and
unlabeled data for training.( i.e small amount of

labeled data with large amount of unlabeled data).
It falls between unsupervised learning and supervised
learning.
It refers to the transductive learning or inductive
learning.
WHY SEMISUPERVISED MODEL IS

REQUIRED?
When unlabeled data is used in conjunction with small
amount of labeled data, produces higher accuracy.

It is used most popularly in some cases in which vast
quantities of unlabeled data is unavailable.
Example: Text on website, protein sequences or images
etc
The cost associated with the labeling process may
render a fully labeled training set infeasible, whereas
acquisition of unlabeled data is inexpensive
WHAT IS SEGMENTATION?
Segmentation is the process of dividing or separating
image in to regions based on similar attributes.

The main goal of image segmentation is
o To divide an image into parts that have a strong
correlation with objects or regions of the real world
depicted in the image.
EXISTING SYSTEM
EXISTING SYSTEM:
In present situations the images are segmented by
considering the geometric properties(i.e. by

considering the position of pixels)
The existing models have been very successful in
segmenting single images alone.
Disadvantage of existing
system:
Time complexity
Difficult to segment
Difficult to identify the problem exactly
Provision for segmenting single image alone.
It is not feasible for many applications
PROPOSED SYSTEM
Proposed System
In the proposed model the image is segmented by
considering the photometric property(i.e. images are

segmented based on colors).
We propose a semi supervised optimization model that
determines an efficient segmentation of retinal image.
Advantage:
Time complexity is reduced
Fully automatic
Results are accurate
It allows the user to easily specify what he/she wants.
The model requires only minimal tuning of model
parameters during the initial stage.
SYSTEM ARCHITECTURE
SYSTEM ARCHITECTURE
USER
IMAGES
QUERY
IMAGE
VISUAL
COLOR
TRANSFORM
GROUND
TRUTH
BLOCK DIAGRAM
DESCRIPTION
IMAGE
The input image is obtained from the database.
The images to be processed are stored in database.
Here the retinal image is used as input image.
QUERY IMAGE:
It is the image to be manipulated.
Here retinal image is used as query image.
The retinal image consist of a background and four
retinal layers.
They are ganglion cell layer, inner nucleus layer, outer
nucleus layer and rod outer layer.
The outer nucleus layer is segmented and it is used for
further analysis.
COLOR TRANSFORM:
Before segmenting the image, a color transform is
applied to each unlabeled image so that its color

histogram matches with that of the labeled image.
This is to correct the difference in brightness and
contrast between images.
GROUND TRUTH:
The resulting output is known as ground truth.
It is the required color transformed and segmented
image.
This image is used for further analysis.
Using the color changes the diseased area of the image
can be easily identified.
After identifying the diseased area proper diagnosis is
given.
SOFTWARE
IMPLEMENTATION
ALGORITHM
Get an input image
Read the image size
Declare the initial parameters
Use for loop to achieve initial to end co-ordinate
Color transform the input image
Segment the input image
End the for loop
Get the output image
K-MEANS CLUSTERING
K-means clustering is a method of cluster analysis
which aims to partition n observations into k

clusters in which each observation belongs to the
cluster.
ADVANTAGE:
It is very fast thus it is common to run it multiple
times with different starting conditions
STEPS OF K-MEANS CLUSTERING

ASSIGNMENT STEP:
Place K points into the space represented by the
objects that are being clustered.
K clusters are created by associating every observation
with the nearest means.
UPDATE STEP:
The centroid of each k-clusters becomes the new
means.
Repeat steps1 and 2.
ADVANTAGE OF MATLAB
Several mathematical operations that work on arrays
or matrices are built-in to the MATLAB environment.

Operations can be performed using command line.
MATLAB provides HELP option which makes the
user easy to use it.
RESULT
ORIGINAL RETINAL IMAGE

Name
Size
Bytes Class
Attributes
inImg
500x500x3
750000 uint8
nBins
1x1
8 double
nClass
1x1
8 double
winSize
Input 1
Name
Attributes
inImg
nBins
nClass
winSize
Size
500x500x3
1x1
1x1
1x1
Bytes Class
750000 uint8
8 double
8 double
8 double
Input 2
Name
inImg
nBins
nClass
winSize
Size
365x600x3
1x1
1x1
1x1
Bytes Class
Attributes
657000 uint8
8 double
8 double
8 double
GROUND TRUTH
Input 3
Name
inImg
nBins
nClass
winSize
Size
365x600x3
1x1
1x1
1x1
Bytes Class
Attributes
657000 uint8
8 double
8 double
8 double
RESULT
The resulting ground truth image is used for further
analysis.
By using the color variation in the segmented output
the corresponding disease is identified and proper
diagnosis is given.
CONCLUSION
CONCLUSION
In this project , we have proposed and developed a
semiautomatic optimization model for segmentation of

retinal image
In our application the parameters can be tuned easily
by the user.
Hence the results are accurate.
FUTURE
ENHANCEMENT
FUTURE SCOPE
In future we have planned to extend our work for
several collection of images, since segmenting

collection of image is a serious issue
Further our future projects is planned to deal with
segmentation of collection of images simultaneously
and thereby reducing the time complexity in detecting
the disease.
Thank you

Presentation Slides

Uploaded by

Copyright:

Available Formats

Presentation Slides

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Presentation Slides

Uploaded by

Copyright:

Available Formats

A Semi supervised

computer vision, computer graphics, medical imaging

WHAT IS SEMISUPERVISED MODEL?

unlabeled data for training.( i.e small amount of

WHY SEMISUPERVISED MODEL IS

amount of labeled data, produces higher accuracy.

image in to regions based on similar attributes.

considering the geometric properties(i.e. by

considering the photometric property(i.e. images are

parameters during the initial stage.

applied to each unlabeled image so that its color

which aims to partition n observations into k

STEPS OF K-MEANS CLUSTERING

or matrices are built-in to the MATLAB environment.

ORIGINAL RETINAL IMAGE

semiautomatic optimization model for segmentation of

several collection of images, since segmenting

You might also like