Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

presentation

No description
by

reham awadlah

on 7 July 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of presentation

Image De-noising
Image De-noising
Two main image de-noising techniques will be discussed:


Spatial domain filters.

Transformed domain filters.

I. Mean Filter
Replace each pixel value in an image with the mean value of its neighbors, including itself .

It is based on a Kernel (mask), which represents the shape and size of the neighborhood to be sampled when calculating the mean; for example:

II. Gaussian Filter

It is similar to the mean filter, but it uses a different Kernel that represents the shape of a Gaussian hump.
Mask of the Gaussian filter [5 * 5]

The Original Image and Salt & Pepper Noise


To solve the problem of misalignment in acquired images due to (camera movement – object movement – different resolutions - …).

Clinicians seek to integrate the complementary information provided by different modalities.

This leads to improved diagnostic accuracy and treatment effectiveness
The Need For Image Registration
Aims of The Present Work
To evaluate the performance of different image de-noising techniques using simulated data set.

To compare different image

registration techniques.

To apply the best registration
technique to a selected set of
retinal images.

To process and register
retinal images.

PROCESSING AND REGISTRATION OF RETINAL IMAGES
Supervisor
Assis. Prof. Hossam El-Din Moustafa

Mansoura University
Faculty of Engineering
Dept. of Electronics and
Communications Engineering

Project Outline
chapter 1

chapter 2
Chapter 3
chapter 4
chapter 5
gives an introduction to the project.

describes the basic
idea of digital simulation.
describes image denoising techniques
describes the image registration problem, main components of an image registration system, and the techniques used to solve the registration problem
describes color processing and color models.
Chapter 6
introduce retinal Images processing and registration.
Chapter 7
summarizes the concluded remarks and the future work.
Digital Simulation
Four different artifacts will be simulated;


Gaussian noise, salt & pepper noise,

speckle noise, and composite noise,


which is a scaled version of the first three types.

Noise Simulation
Real images usually contain departures from ideality which are referred to as noise or artifacts.

Digital Simulation
Digital Image Simulation
The original image is cameraman image of size (256×256) .

White noise is a random signal with flat power spectral density which is normally distributed.

Gaussian noise with zero mean and 0.01 variance was added to the original image.

Gaussian Noise
Gaussian noise is an idealized form of white noise, which is caused by random flactuations in the signal.

The Original Image & Gaussian Noise
The Original Image and Gaussian Noise

Speckle Noise
It is modeled with a multiplicative or nonlinear model using the following equation :
Where I is the pure image,
J is the output image, and n is a uniformly distributed random noise with zero-mean
The Original Image and Speckle Noise
The Original Image and Speckle Noise

During the imaging process all types of artifacts are present and affect the image quality.

Composite Noise
Composite noise is a scaled version of the first three types of noise.

The Original Image and Composite Noise

The Original Image and Composite Noise

Salt & Pepper
Salt & pepper noise can be represented as on/off pixels.

The noise is usually quantified by the percentage of pixels which are corrupted.

This degradation can be caused by sharp, sudden disturbances in the image signal.


The Original Image and Salt & Pepper Noise


Results of Mean Filter
The Output of The Gaussian Filter
Disk Filter
An average filter but Its mask is circular

Non-Linear Filters
Removing Speckle Noise Using Spatial Domain Techniques
Removing Salt & Pepper Noise Using Spatial Domain Techniques
Removing Gaussian Noise Using Spatial Domain Techniques
Non-Linear Filters
A nonlinear filter is an image processing technique whose output is not a linear function of its input.
Nonlinear filter locates and removes data that is recognized as noise.
The algorithm is nonlinear because it looks at each data point and decides if that data is noise or valid information.

Removing Composite Noise Using Spatial Domain Techniques
I. Wiener Filters
The Wiener filter is: non Linear, adaptive, and time-invariant.
It minimizes the mean-squared value of the error (MSE); that is defined as the difference between desired response and the actual filter output.

II. Median Filter
The median is calculated by first sorting all
the pixel values from the surrounding neighborhood into numerical order and then replacing
the pixel being considered with the middle pixel value.

If the neighborhood under consideration contains an even number of pixels, the average of the
two middle pixel values is used .

welCome
Transformed Techniques
Transformed domain filters process the digital image in any domain other than spatial domain. The domain may be frequency, wavelet, or any other .

Wavelet Filters
Wavelet Filters
The Wavelet transform is a signal analysis tool that provides a multi-resolution decomposition of an image and results in a non-redundant image representation

This basis consists of wavelets, which are functions generated from one single function, called mother wavelet, by dilations and translations.

Results of Wavelet Filter
Advantages of Wavelet Transform
Windowing technique with variable sized regions.
Window will be long time for low frequency.
Window will be short time for high frequency.
Multi-scale.
Shifting.
Variety of bases functions, which will be selected to fit the application.
Mother Function
Haar.

Dubechies

Coiflets

Morlet
Meyer

Biorthogonal

Symlets

Removing Gaussian Noise Using Wavelet Filter
Removing Salt & Pepper Noise Using Wavelet Filter
Removing Speckle Noise Using Wavelet Filter
Removing Composite Noise Using Wavelet Filter
Eye collects light, focuses on retina,
forms images Refraction of light by the cornea
Parallel lights from far distance must be
bent by refraction
Air to aqueous humor change
makes refraction on the surface
of cornea

Focal distance depends on
the curvature of cornea

The Structure of the Eye Retina
Retinal Anatomy

Retinal Anatomy
Image Formation by the Eye

Image Formation by the Eye
Anatomy
Medical Background
Image Formation by the Eye

Intra subject
Inter subject
Atlas
A. Monomodal
B. Multimodal
C. Modality to model
D. Patient to modality
The parameters that make up the registration transformation can be either
directly computed or searched for.
Directly computed

parameters are determined in an explicit fashion from the available data.

The searching optimization
methods are determined by finding an optimum
of some function defined
on the parameter space.

Automatic
Semi-automatic
Interaction
A transformation is called
global
if it applies to the entire image.

It is called
local
if subsections of the image each have their own transformation defined.



Extrinsic methods rely on artificial
objects attached to the patient.



A. Spatial dimensions only:

Image Registration :

Image registration is the process of geometrically aligning two or more images corresponding to the same scene but taken under different imaging conditions.
In other words, image registration is the process of superimposing two images to find the best
transform to make them match.

8.Subject
The object of the registered image may be related to the
head
; brain, skull, eye, or dental.

It can be
thorax
; cardiac or breast.

It may also be
abdominal
; kidney or liver.

It can be related to
limbs
; femur, humerus, or hand.

9. Object

7. Modalities involved
6. Optimization Procedure
5. Interaction
4. Domain of The Transformation
2. Nature of Registration Basis
1. Dimensionality
Importance of Image Registration
Examples of Transformation Type
This figure shows the concept of image registration.


Face detection.

Estimating intensity differences between images.

Biomedical image registration.

Image fusion.

Taxonomy Of Medical Image Registration Methodologies
Dimensionality
Nature of registration basis
Nature of transformation
Domain of transformation
Interaction
Optimization procedure
Modalities involved
Subject
Object
2D/2D
2D/3D
3D/3D
B. Time series (more than two images):

2D/2D
2D/3D
3D/3D
A. Extrinsic
B. Intrinsic

Intrinsic methods rely on patient
generated image content only.

3. Nature of The
Transformation
Nature of transformation

Rigid
Affine
Projective
Curved
An image coordinate transformation
is called
rigid
, when only translations and rotations
are allowed .

If the transformation maps parallel lines onto parallel lines it is called
affine.

If it maps lines onto lines, it is called projective.

If it maps lines onto curves, it is called curved or
elastic.

Nature of The Transformation
Automatic,
The user only supplies the algorithm with the data and possibly information on the image acquisition

Semi-automatic
:
The interaction required can be of two different natures:
the user needs to initialize the algorithm, e.g., by segmenting the data, or steer the algorithm, e.g., by rejecting or accepting suggested registration hypotheses.

Interactive:
The user does the registration himself, assisted by software supplying a visual or numerical impression of the current transformation, and possibly
an initial transformation guess.

A. Mono modal applications:

The images to be registered belong to the same modality.
CT - CT
MR - MR
PET - PET
B. Multimodal registration
The images to be registered stem from two different modalities:

CT-MR
CT-PET
PET-MR
PET-US
C. Modality to model
D. Patient to modality

2. Maximization of Mutual Information Registration
Image Registration Techniques
1. Control Points Selection Registration
1. Control Points Selection Registration
Control points selection based on selecting control point pairs from two images , input image ad base image.

The control point pairs can be selected either automatically or by the user on the monitor.
Visual selection is a general
and safe method.

2. Maximization of Mutual Information Registration
MI is a measure of the amount of information that one random variable contains about another random variable.
MI will indicate the best match between a reference image and an input image.

commonly used definition:
I(A,B) = H(A) + H(B) - H(A,B) Advantage in using mutual info over joint entropy is it includes the individual input’s entropy

2. Maximization of Mutual Information Registration
1. Calculate (MI) of A and B for each rotation.

2. For each rotation, calculate (MI) for all possible translations of the input image to the reference image.

3. Find the overall maximum MI value. The coordinates of this maximum value refers to the rotation and translation.

Two different techniques were implemented:

Registration using control points.
Registration by maximization of mutual information (MI).
1. Control Points Selection Registration
1. Control Points Selection Registration
1. Control Points Selection Registration
Performance Criteria
1. Pixel difference-based measures as mean square error.
2. Correlation-based measures as (NCCC).
3. Edge-based measures as edge measure.
4. Spectral distance-based measures as spectral phase error.
5. Context-based measures as rate distortion measure.
6. Human Visual System (HVS)-based measures
as absolute norm.

Normalized Cross Correlation Coefficient (NCCC)
It can be described by the following equation:

Weighted Peak Signal-to-Noise Ratio
(WPSNR)

PSNR=24.6, WPSNR=26.4

PSNR=24.6, WPSNR=27.9

PSNR=24.6, WPSNR=29.3

Image Scaled and Rotated With Different Values

Registration Using Control Points
(Rotated image)

Image Rotated With Different Values


Image Scaled With Different Values

Registration Using Control Points
(Scaled image)

Registration Using MI Maximization
(Scaled Image)

Registration Using MI Maximization
(Scaled and Rotated Image)

Comparison Between Registration Techniques

Registration Using Control Points
(Scaled and Rotated image)

Data Simulation
Registration Using MI Maximization
(Rotated Image)

Processing of Colour Images
Perceptual Aspects of Colour
Colour Models
is a method to specifying colours in some standard way.
The human visual system is attuned to two things :

Edges
Color


What is colour?
Color study consists of:

1)The physical properties of light which give rise to colour.

2)The nature of the human eye and the ways in which it detects colour.

3)The nature of the human vision centre in the brain, and the ways in which messages from the eye are perceived as colour.

Physical Aspects of Colour
1. RGB
2. HSV
Each colour is represented as 3 values (R,G,B) indicate the amount of red,green and blue which make the colour

HSV stands for :
Hue the true colour attribute (red,green,blue,yellow,………)
Saturation the amount of colour which diluted with white
Value the degree of brightness

3.YIQ
The first technique depends on selecting control points from both the reference and the input images.
The second technique is based on maximization of mutual information between the two images.
The two registration techniques were evaluated and compared using both the Weighted Peak Signal to Noise Ratio (WPSNR) and the Normalized Cross Correlation Coefficient (NCCC).
MI maximization has given the best results in all cases.
It does not require any assumptions about the nature of the imaging modalities. It is robust with respect to variations of illumination.

In order to obtain better performance, low quality retinal images have been pre-processed using a set of color image processing techniques.

These include histogram equalization, histogram stretching, unsharp masking, and edge detection.

The retinal images have been improved significantly after the preprocessing step and this have led to more accurate registration results.

Future Work
Future work includes trying new techniques for image processing and registration.
Fast registration algorithms can be implemented based on edge images.
The presented techniques can be applied to more biomedical image sets.
Processing on HSV color model.

Conclusion

The present work introduced the image restoration techniques both in spatial domain and in transformation domain.
Image denoising was performed using average, disk, Gaussian, median, and Wiener filters.
Wavelet filters have been implemented.
Performance of different image denoising techniques was evaluated based on SNR
The present work introduced the image restoration techniques both in spatial domain and in transformation domain.

Processing Of Colour Images
Methods For Processing Of Colour Images
1.Contrast Enhancement
Histogram
is a graph indicating the number of times each grey level occurs in the image.

2.Unsharp Masking
The idea of unsharp masking is to subtract a scaled unsharp version of the image from the original image.

3.Edge Detection
Edges contain some of the most useful information in an image.
We may use edges to Measure the size of objects in an image; to isolate particular objects from their background; to recognize or classify objects.
We can use some filters for edge detection


1.First derivative
2.Second derivative

1: Contrast Enhancement


a) Histogram Stretching
b) Histogram Equalization

2: Unsharp Masking

3: Edge Detection

Processing of Color Images
1.Contrast Enhancement
1.Contrast Enhancement “ Bright”
1.Contrast Enhancement “Dark”
1.Contrast Enhancement
Histogram Equalization :

Histogram Stretching;

Application of Color Processing Techniques on Retinal Images (YIQ model)
Application of Color Processing Techniques on Retinal Images (RGB model)
Application of Color Processing Techniques on Retinal Images (RGB Model)
Application of Color Processing Techniques on Retinal Images (YIQ model)
Application of Color Processing Techniques on
Retinal Images (Gray model)

Application of Color Processing Techniques on Retinal Images (Gray model)
Conclusion
Registration by MI Maximization for RGB (Red Layer)
Registration by Control Points Technique for RGB
Registration by Control Points Technique for RGB
Tested Image

Registration by MI Maximization for YIQ (Q channel)
Original Image
Real Retinal images

Registration by Control Points Technique for RGB
Registration by Control Points Technique for RGB
Registration by Control Points Technique for Y Channel
Registration by Control Points Technique for I
Channel :

Registration by Control Points Technique for Q
Channel

Registration by Control Points Technique for YIQ

Registration by MI Maximization for RBG (Green Layer)
Registration by MI Maximization for RGB (Blue Layer)
Registration by MI Maximization for YIQ (Y channel)
Registration by MI Maximization for YIQ (I channel)
Retinal Images Registration

Blood vessels on the surface of
Retina


Optic disk :
A pale circular region
Gate for entering blood vessels
and Exiting optic nerve fibers
Blind spot
No photoreceptors present
Brain is deceiving you!

Macula (spot):
Central vision
Relative absence of blood vessels - improves the quality of central vision

Fovea (pit):
The centre of retina

The normal retina consists of 10 layers from outward inward:
Pigment Epithelium:
The RPE shields the retina from excess incoming light. the former for building photoreceptive membranes.

Ganglion Cells:
are types of nerve cells that receives signal inputs on average from about 100 rods and cones.

Muller Cells:
act as light collectors in the mammalian eye, collecting light to the rod and cone photoreceptors

Types of photoreceptors

Rods :
Have more disks and higher photo pigments concentration . More sensitive to light than cones
Mainly number of rods in the retina are about 120 million cell


Cones:
detect colors
Responsible for night vision and peripheral vision.
Mainly number of cones indide the retina are about 6 million cell

Accommodation by the Lens
Changing shape of lens allows for extra focusing power.

Important for focusing images of objects within 9m ranges .









Accommodations by
ciliary muscle contraction :
lose this function with age


The Pupillary Light Reflex :
Connections between retina and brain stem neurons that control muscle around pupil
Continuously adjusting to different ambient light levels
Pupil similar to the aperture of a camera
increase the depth of focus by the constriction of pupil

How to Capture Retina Image
There two type of capturing retina images:
Optical coherent temography

Fundus Cameras:
are used for fundus photography which is used by ophthalmologists for monitoring progression of a disease and diagnosis of a disease
Known Diseases
Glaucoma:

Increased fluid pressure in the eye (aqueous humor) leading to increased pressure inside the eye which damage the optical nerves and if left un-treated will lead to blindness in the affected eye.

Proliferative Diabetic Retinopathy
:

Caused by complications of diabetes lead to decreased effectiveness of the
vascular walls making
retinal blood vessels become
more permeable;
damaging the retina.
How to Capture Retina Image
Full transcript