waveletbasedfeatureextractionschemeofeegwaveform-13432076010536-phpapp02-120725041401-phpapp02

7/27/2019 waveletbasedfeatureextractionschemeofeegwaveform-13432076010536-phpapp02-120725041401-phpapp02

1/42

ANNA UNIVERSITY: CHENNAI 600 025

MAY 2012

DEPARTMENT OF ELECTRONICS AND

COMMUNICATION ENGINEERING

PROJECT VIVAVOCE


2/42

WAVELET BASED FEATURE

EXTRACTION SCHEME OFELECTROENCEPHALOGRAPHY

PRESENTED BY

E.ARUNA-12708106004

M.S.R.PUNEETHA CHOWDARI-12708106043B.SASI KALA-12708106050

N.SHANTHA PRIYA-12708106052

UNDER THE GUIDANCE OFMR.C.E.MOHAN KUMAR, M.E

ASSISTANT PROFESSOR

ECE DEPARTMENT


3/42

ABSTRACT

The Electroencephalogram (EEG) is a neuronal activity that represents the

electrical activity of the brain.

The specific features of EEG are used as input to Visual Evoked Potential

(VEP) based Brain-computer Interface (BCI) or self paced BCIs (SBCI)

for communication and control purposes.

This project proposes scheme to extract feature vectors using wavelet

transform as alternative to the commonly used Discrete Fourier Transform

(DFT).

The selection criterion for wavelets and methodology to implement

decomposition procedure, coefficient computation and reconstruction

methods are presented here using MATLAB software tool.


4/42

OBJECTIVES

To improve quality of life for those with severe neuromuscular disabilities

and aimed at restoring damaged hearing, sight and movement of muscles

by neuro-prosthetics applications based brain computer interface.

To investigate the feasibility of using different mental tasks as a wide

communication channel between neuro-diseased people and computer

systems.

To achieve the proper and efficient feature extraction algorithms can

improve the classification accuracy and to overcome the resolutionproblem and localization of artifact components in time and frequency

domain.


5/42

KEY WORDS

Electroencephalogram (EEG)

Brain-Computer interface (BCI)

Wavelet Transform (WT)

Continuous Wavelet Transform (CWT)

Discrete Wavelet Transform (DWT)

Visually Evoked Potential (VEP)

Discrete Fourier Transform (DFT)


6/42

INTRODUCTION

In human physiological system, Amyotrophic Lateral Sclerosis (ALS) is a

progressive neuronal-degenerative disease that affects nerve cells which

are responsible for controlling voluntary movement.

A Brain Computer Interface (BCI) or Brain Machine Interface (BMI) has

been proposed as an alternative communication pathway, bypassing the

normal cortical-muscular pathway.

BCI is a system that provides a neural interface to substitute for the loss of

normal neuronal-muscular outputs by enabling individuals to interact withtheir environment through brain signals rather than muscles.


7/42

BRAIN COMPUTER INTERFACE

Direct connection between the brain and a computer without using any of

the brains natural output pathways.

Neural activity of the brain cells are recorded and these signals are given

as drive to applications.

Read the electrical signals or other manifestations of brain activity and

translate them into a digital form.


8/42

BRAIN COMPUTER INTREFACE

WORKING

Blocks of Brain-Computer Interface

EEG Signal Acquisition

Signal Preprocessing

Feature Extraction

Signal Classification


9/42

LITERATURE REVIEW

The history of braincomputer interfaces (BCIs) starts with Hans Berger's

discovery of the electrical activity of human brain and the development of

electroencephalography (EEG).

Electroencephalography (EEG) is the most studied potential non-invasive

interface, mainly due to its fine temporal resolution, ease of use,

portability and low set-up cost.

Research on BCIs began in the 1970s at the University of California Los

Angeles (UCLA).

The field of BCI research and development has since focused primarily

on neuro-prosthetics applications that aim at restoring damaged hearing,

sight and movement.


10/42

LITERATURE REVIEW (CONT.)

Invasive BCIs: Implanted directly into the grey matter of the brain during

neurosurgery.

Partially invasive BCIs: Devices are implanted inside the skull but rest

outside the brain rather than within the grey matter.

Non-invasive BCIs: Non-invasive neuro-imaging technologies as

interfaces.

Lawrence Farwell and Emanuel Donchin developed an EEG-based braincomputer interface in the 1980s.


11/42

FEATURE EXTRACTION

Due to stimulus in various sense organs , the responses is created in the

surface of the brain in the form of wavelets (evoked potentials).

These potentials is are the sum of the responses due to desired (EEG

waveforms) and undesired stimulus (EMG and EOG waveform).

From these responses a desired response is extracted which is called

feature. The whole process is called Feature Extraction.

This feature is given as a input or driving signal to the application to make

it work.


12/42

EXISTING SYSTEM

FOURIER TRANSFORM:

Breaks down a signal into constituent sinusoids of different frequencies.

Transform the view of the signal from time-base to frequency-base.

Only analyze the stationary signals but not the non stationary signals.

It can analyze the continuous signal with uniform frequency.

dtetfF tj


13/42

EXISITING SYSTEM

SHORT TIME FOURIER TRANSFORM

To analyze small section of a signal, Denis Gabor (1946), developed a

technique based on the FT and using windowing.

A compromise between time-based and frequency-based views of a signal.

Both time and frequency are represented in limited precision. The

precision is determined by the size of the window.

Window size is fixed.


14/42

DRAWBACKS OF EXISTING SYSTEM

Unchanged Window and frequency of the signal should be fixed.

Localization of artifact components and transients is not accurate.

Provides a signal which is localized only in frequency domain not in time

domain.

Signal is assumed to be stationary.

FT cannot locate drift, abrupt changes, beginning and ends of events

Does not provided Multi-resolution analysis.

Dilemma of Resolution

Wide window : poor time resolution

Narrow window : poor frequency resolution


15/42

PROPOSED SYSTEM

WAVELET TRANSFORM:

It is a mathematical tool for processing and analyzing the EEG signals

and to localize the artifact component in it.

An alternative approach to the Fourier transform to overcome the

resolution problem.

It is used to localize the spikes, spindles, ERPs.

It can analyze non-stationary signals.


16/42

PROPOSED SYSTEM

Basic Idea of DWT: To provide the time-frequency representation.

Wavelet

Small wave

Means the window function is finite length

Mother Wavelet

A prototype for generating the other window functions

All the used windows are its dilated or compressed and shifted versions.


17/42

MULTI RESOLUTION ANALYSES

It is a ability to disintegrate the signal components into fine and coarse

elements.

It is also defined as ability to extract the fine components from the signals.

Analyze the signal at different frequencies with different resolutions.

Good time resolution and poor frequency resolution at high frequencies.

Good frequency resolution and poor time resolution at low frequencies.

More suitable for short duration of higher frequency; and longer duration

of lower frequency components.


18/42

WAVELET TRANSFORM

ADVANTAGE OF WAVELET ANALYSIS:

It permits the accurate decomposition of neuro-electric waveforms like

EEG and ERP into a set of component waveforms called detail functions

and approximation coefficients.

It provides flexible control over the resolution with which neuro-electric

components and events can be localized in time, space and scale.

Wavelet transform can analyze the discontinuous signal with variable

frequencies.

It can analyze the non stationary waves.

It provides multi resolution.


19/42

WAVELET TRANSFORM

ADVANTAGE OF WAVELET ANALYSIS:

Wavelet representation can indicate the signal without information loss.

Through two pass filters, wavelet representation can reconstruct the

original signal efficiently.

Compared with Fourier transform, wavelet is localizable in both frequency

domain and space domain.

Wavelet representation provides a new way to compress or modify images.

For High frequencies it uses narrow window for better resolution and for

Low frequencies it uses wide window for bringing good resolution.


20/42

CONTINUOUS WAVELET TRANSFORM

The sum over the time of the signal convolved by the scaled and shifted

versions of the wavelet.

Its slow and generates way too much data. Its also hard to implement.

The continuous wavelet transform uses inner products to measure thesimilarity between a signal and an analyzing function.

dt

a

bt

a

tfttfbaC

*1)())(),(;,(


21/42


STEP 2:

Calculate a number, C, that represents how closely correlated the wavelet is

with this section of the signal. The higher C is, the more the similarity.

STEP 1: Take a Wavelet and compare it to a section at the start of the original signal.


22/42


STEP 3: Shift the wavelet to the right and repeat steps 1-2 until weve

covered the whole signal


23/42


STEP 4: Scale (stretch) the wavelet and repeat steps 1-3


24/42

DISCRETE WAVELET TRANSFORM

Wavelet transform decomposes a signal into a set of basis functions.

these basis functions are called wavelets.

Wavelets are obtained from a single prototype wavelet y(t) called mother

waveletby dilations and shifting:

where a is the dyadic scaling parameter and b is the dyadic shifting

parameter

)(1

)(,a

bt

a

tba


25/42

DISCRETE WAVELET ANALYSIS

(Cont.)

WAVELET CO-EFFICIENT:

At the large scale, the wavelet is aligned with the beginning of the EEG

waveform and the correlation of the wavelet shape with the shape of the

EEG waveform at that position is computed.

The same wavelet is then translated (moved) a small amount to a later

position in time, bringing a slightly different portion of the EEG waveform

a new wavelet coefficient is computed.

Whenever the wavelet shape matches the overall shape of the ERP, a large

wavelet coefficient is computed, with positive amplitude if the match is

normal and negative amplitude if the match is polarity inverted.


26/42

DISCRETE WAVELET ANALYSIS

(Cont.)

Conversely, when the shape match is poor, a small or zero wavelet

coefficient is computed.

At the small scale, the process of computing wavelet coefficients is thesame. The only difference is that the wavelet is contracted in time to bring

a different range of waveform fluctuations into the viewof the wavelet.


27/42

HAAR WAVELET

It is a type of Discrete Wavelet function and sequence of rescaled square

shaped functions.

Scaling function (father wavelet)

Wavelet (mother wavelet)

These two functions generate a family of functions that can be used to

break up or reconstruct a signal

The Haar Scaling Functions:

Translation

Dilation


28/42

MATCHING WAVELETS TO EEG

WAVEFORMS

The wavelet transform is free to use wavelets as its basis functions.

Wavelets have shapes that are as close as possible to the shapes of the

EEG events.

MATCHING PURSUIT:

To examine the spectral properties of a EEG waveform over segments of

different size and location.

To select a set of basis functions from a large dictionary of basis functions

that closely match the spectral properties of those regions of the EEG

waveform.


29/42

MATCHING WAVELETS TO EEG

WAVEFORMS (Cont.)

MATCHED MEYER WAVELETS

A method of directly designing a wavelet to match the shape of any signal

of interest.

The technique constructs a member of a flexible class of band-limited

wavelets, the Meyer wavelets, whose spectrum matches the spectrum of

any band-limited signal as closely as possible in a least squares sense.

An associated scaling function and high and low pass filters are then

derived that can be used to perform a DWT on any EEG waveform.


30/42

SIGNAL DECOMPOSITION

The decomposition of the signal led's to a set of Coefficients called

Wavelet Coefficients. Therefore the signals can be re-constructed as a

linear combination of wavelets functions weighed by the Wavelet

Coefficients.

Then the signal is sent through only two sub-band coders (which get the

approximation and the detail data from the signal).

High frequency and low scale components are know as Detail Coefficient

and Low frequency and low frequency components are known as

Approximation Coefficients.

Signal decomposed by

low pass and high pass

filters to get approx anddetail info.


31/42

SIGNAL DECOMPOSTION

The signal can be continuously

decomposed to get finer detail and more

general approximation, this is called

multi-level decomposition.

A signal can be decomposed as many

times as it can be divided in half.

Thus, we only have one approximation

signal at the end of the process.

Low Pass: Scaling Function, High Pass:

Wavelet Function.


32/42

32

SUB BAND CODING

h0(n)

h1(n)

2

2

2

2

g0(n)

g1(n)

+Analysis Synthesis

1( )y n

0 ( )y n

( )x n ( )x n

1 ( )H 1 ( )H

/ 2

Low band High band

0


33/42

SUB BAND CODING (Cont.)

Halves the Time Resolution: Only half number of samples resulted.

Doubles the Frequency Resolution: The spanned frequency band halved.

Filters h0(n) and h1(n) are half-band digital filters.

Their transfer characteristics H0-low pass filter, Output is an

approximation of x(n) and H1-high pass filter, output is the high frequency

or detail part of x(n).

Criteria: h0(n), h1(n), g0(n), g1(n) are selected to reconstruct the input

perfectly.


34/42

RECONSTRUCTION

A process Afterdecomposition or analysis iscalled synthesis.

Reconstruct the signal from the wavelet coefficients .

Where wavelet analysis involves filtering and down sampling, the wavelet

reconstruction process consists of up sampling and filtering.

For perfect reconstruction filter banks we have

In order to achieve perfect reconstruction the filters should satisfy

Thus if one filter is low pass, the other one will be high pass.

x x

0 0

1 1

[ ] [ ]

[ ] [ ]

g n h n

g n h n


35/42

IMPLEMENTATION BY MATLAB

MATLAB is high-performance interacting data-intensive software

environment for high-efficiency engineering and scientific numerical

calculations.

MATLAB is based on a high-level matrix array language with controlflow statements, functions, data structures, input/output, and object-

oriented programming features.

It integrates computation, visualization, and programming in an easy-to-

use environment where problems and solutions are expressed in familiar

mathematical notation.


36/42

RESULTS AND OUTPUTS

Outputs.docx
http://localhost/var/www/apps/conversion/tmp/scratch_4/Outputs.docxhttp://localhost/var/www/apps/conversion/tmp/scratch_4/Outputs.docx


37/42

SUMMARY

SUMMARY ON ARTIFACT REMOVAL SCHEME

The performance of the system deteriorates when the EOG and EMG

artifacts contaminate the EEG signal.

The goal of this thesis is to devise a scheme that achieves efficient artifact

removal from a composite EEG signal which in turn provides lower false

positive rates for SBCI systems.

The wavelet transform explores both time and frequency information, is

expected to be a more suitable feature extractor than those which work inthe time or frequency domain only The DWT is used main tool in this

scheme.


38/42

SUMMARY

SUMMARY ON MONTAGE SCHEME

The performance of the scheme was tested using the signal recorded from

13 monopolar EEG signals and from 18 bipolar EEG signals.

The performance of the system based on monopolar EEG electrodes was

weak and it resulted in high false positive rates.

Bipolar montage results in superior performance to those of the monopolar

montage.


39/42

SUMMARY

SUMMARY ON FEATURE EXTRACTION SCHEME

These results enable to describe the characteristics of various regions of

the brain for a specific stimulus.

The wavelet based scheme efficiently demarcates the Mu and Beta

rhythms and various other frequency bands and power associated with

each frequency band.

Bi-frequency stimulation produces more noise than single frequency

stimulation and both frequencies are not always elicited. A unique feature

vector is produced by single frequency stimulation from either

fundamental or harmonic component.


40/42

CONCLUSION

This project presents the use of wavelet transform for a given feature

extraction associated with electrode pair.

Mathematical basis of the wavelet transform has proved that EEG analysis

based on wavelet transform coefficients can be used very efficiently for

the estimation of EEG features.

Results of EEG feature extraction can be further improved by various

methods but one of the most important problems is in the right definition

of EEG features using both its frequency-domain and time-domainproperties.


41/42

FUTURE SCOPE

The proposed scheme was developed and implemented to address the

shortcomings in the design of Steady State Visual Evoked Potential

(SSVEP) based BCI systems.

SSVEP based BCI systems are assistive technology devices that allow

users to control objects in their environment using their brain signals only

and at their own pace.

This is done by measuring specific features of the brain signal that pertain

to intentional control (IC) commands issued by the user.


42/42

THANK YOU

waveletbasedfeatureextractionschemeofeegwaveform-13432076010536-phpapp02-120725041401-phpapp02

Documents