You are on page 1of 15

A NOVEL APPROACH FOR THE DETECTION

OF NEW VESSELS IN THE RETINAL IMAGES


FOR SCREENING DIABETIC RETINOPATHY

Under the guidance of Presented By


Dr.S.KAJA MOHIDEEN SARANYA.K
Head of the Department(ECE) M.Tech(COS)
B.S Abdur Rahman University
Chennai .
OBJECTIVE

 To detect abnormal blood vessels in the


retinal images to diagnose Diabetic
Retinopathy.
ABSTRACT:

 Proliferative Diabetic Retinopathy causes


sudden loss of vision.
 Characterized by the development of abnormal
new retinal vessels .
 Preprocessing is done to normalize and enhance
the contrast of an image
 Vessel like segments are detected and segmented
using Fuzzy C-means Clustering.
CONTD...

 Features based on shape, position, orientation,


brightness, contrast and line density are
extracted.
 Based on those features, segment is classified as
normal or abnormal using K-Nearest Neighbour
(KNN) Classifier.
 Comparision of KNN Classifiers and Support
Vector Machine (SVM) Classifiers are done.
BLOCK DIAGRAM
TOP HAT &BOTTOM HAT
TRANSFORM
of SEGMENTATION
ADAPTIVE HISTOGRAM
RETINAL
EQUALIZATION
IMAGE
Pre-Processing FEATURE
EXTRACTION

CLASSIFICATION

ABNORMAL
NORMAL VESSELS
VESSELS
ABNORMAL NEW BLOOD VESSELS
EXTISTING METHOD

 Watershed transforms and ridge strength


measurement are used for segmentation.
 Normal and abnormal blood vessels are categorized
by a Support Vector Machine(SVM)
Classifier.
Disadvantages:
 Over Segmentation.

 Lack of Smoothness.
PROPOSED METHOD

 Preprocessing operation by
Adaptive Histogram Equalization
Top-hat and Bottom-hat transformation.
 Segmentation is done by Fuzzy C means
Clustering
 Classification is performed using K- Nearest
Neighbour Classifier
IMAGE ACQUISITION:

 Process of acquiring an image.


 Images are collected from STARE, DRIVE,
MESSIDOR DATABASE .
PREPROCESSING:

RGB IMAGE

GREEN COMPONENT OF AN
IMAGE

MEDIAN FILTERING

ADAPTIVE HISTOGRAM
EQUALIZATION

TOPHAT AND BOTTOM HAT


TRANSFORMATION

PREPROCESSED
IMAGE
PREPROCESSING:
GREEN COMPONENT OF IMAGE:
 Green Components are extracted from RGB
image.
 It shows best Contrast between the vessels and
Background retina.
MEDIAN FILTER:
 To reduce noise in an image.

 Works by sorting pixels according to the gray


values and centre pixel is replaced by median of
all pixels.
ADAPTIVE HISTOGRAM EQUALIZATION:

 To normalize and enhance the contrast of an


image.
 More effective for detecting small blood vessels
characterized by low contrast levels.
 It is applied to the noised removed green
component of an image.
 Works by partitioning the image into regions and
histogram equalization is applied to each region.
 So the hidden features of the image are more
visible.
TOPHAT AND BOTTOMHAT
TRANSFORM:
 To correct uneven illumination and to enhance the blood vessels.
TOPHAT TRANSFORM:
That(f) = f – f O b
Where O denotes the opening operation.
BOTTOM HAT TRANSFORM:
Bhat(f) = f b - f
Where  denotes the closing operation.
 Add original image to top hat transformed image and then subtract

the bottom transformed image.


PREPROCESSING OUTPUT:
green component of an image noise filtered image
input retinal image

contrast enhanced image vessel enhanced image

You might also like