**●○●
EEG/ERP Lab**

**●○●
Lab Equipment**

BIOPAC Systems (MP150 Model)
: 16 Channels
Real-Time Recorder

For recording EEG, EMG, ECG, EOG, EGG, Blood Pressure,
etc. Compatibility with MATLAB, Labview, VC++

EEG
electrode caps (small, medium) 10/20 International System

Disposable EMG electrodes

Impedance meter and other accessories

AcqKnowledge 4.1, Data acquisition software

**●○●
Main Research
Topics**

Example of experiments
on EEG study

**○ **
*Cognitive Science
*(decision making, working memory, perception, intelligence,
emotion, behavioral response, neuro-based studies)

**○ ***
Physiological** Signal Processing* (biofeedback, brain-machine
interface, EEG signal classification, ERP, ERD/ERS)

**○ **
*Fractal Analysis and Chaos Theory*
(fractal features, dynamical systems and chaos in EEG)

**○ **
**Speech Processing** (speech
enhancement, speech recognition, speech segmentation)

**○ **
*Image Processing* (fractal
dimension of texture image, object recognition, real-time image
processing)

**●○●
Example of Research
Interests**

**○****
****Differential Box-Counting
Method for Estimation of Gray-Level Image Fractal Dimension**

Fractal dimension (FD) of image studies originated with the principle
found in the
differential box-counting (DBC) method [1]. This study is to
implement simple DBC on the basis of the concept of self-similarity found
in gray images. The FD of a bounded set A in
Euclidean n-space is given by N_{r} α (1/r)^{D} where N_{r} is the
least number of distinct copies of A in the scale r.

Estimated FDs by DBC method (Left) Texture image D04 (Right) Texture
image D33

3-D
space (Left) Texture image D04 (Right) Texture image D33

If we consider an image of size MxM
pixels as a 3-D space with (x,y) representing 2-D and the third
coordinate (z) representing the gray intensity of image, the (x,y) space
is partitioned into grids of size sxs by the DBC method where 1 < s
<= M/2 [3]. In this experiment, we set s = 2, 4, ..., M/2.

Comparison of results of FDs of Brodatz
texture image [2]

Brodatz Texture Images |
Implemented Method |
Conventional Method [1] |

FD |
MSE |
FD |
MSE |

D03 |
2.5801 |
0.0349 |
2.60 |
0.032 |

D04 |
2.6467 |
0.0247 |
2.66 |
0.026 |

D05 |
2.3795 |
0.0305 |
2.45 |
0.032 |

D33 |
2.3474 |
0.0087 |
2.23 |
0.007 |

D84 |
2.6216 |
0.0435 |
2.60 |
0.029 |

D92 |
2.5217 |
0.0340 |
2.50 |
0.023 |

D68 |
2.5306 |
0.0221 |
2.52 |
0.024 |

D55 |
2.4670 |
0.0304 |
2.48 |
0.031 |

References

[1] N. Sarker and B.B. Chaudhuri, "An Effective Differential
Box-Counting approach to Compute Fractal Dimension of Image", IEEE
Trans. Syst. Man Cyb., vol.24, no.1, pp.115-120, 1994.

[2] Brodatz Texture Image Database, SAMP Lab, Ohio State University,
Http://sampl.ece.ohio-state.edu/database.htm.

[3] J. Li, C. Sun, and Q. Du, "A New Box-Counting Method for
Estimation of Image Fractal Dimension", IEEE Int. Conf. Img. Proc.
(ICIP'06), pp.3029-3032, 2006.

**○****
****Nondestructive Maturity
Classification of Durian Based on Fractal Dimension Analysis**

The
objective of this research was to determine maturity levels of durian
by using fractal analysis. Based on the computation of classical
methods, it is difficult to apply theses methods to extract feature
patterns of two different classes. Therefore, in this research, we used the automatic knocking machine
[1] to
knock the durian in which the knocked sound was analyzed in terms of
fractal concepts, and the fractal dimension (FD) values would be
presented as a feature. The fractal algorithm, namely, Higuchi’s
method [2] was selected to evaluate FD of the knocked sound. To show the
FD changes in waveform with respect to time, the time-dependent FD
(TDFD) was proposed. Probability distribution of TDFDs indicates
that the relationship of immaturity and maturity levels of durian
has proven that FD-based features can be employed.

Classical methods (Left) Probability distribution function of two
classes (Right) Corresponding normalized entropy

Comparison of average TDFD waveforms of maturity and immaturity (Monthong
durian weight 3.0 – 3.5 kg.)

Attractors in phase space

The
proposed method achieves the average accuracy rates of 89.94% and
88.20% for the "Monthong" durians weighting 3.5–4.0 kg and 4.0–4.5
kg, respectively. The experimental results also show that FD is
capable of classifying maturity levels of durian effectively.

References

[1] T. Paritwanon,
P. Somboonyod, W. Bundit, and M. Phothisonothai, "Non-Destructive
Fruit Maturity Determination by using Knock Sound Processing" *The
10th Annual Conf. TSAE (in Thai)*, pp.344-349, Nakhon Ratchasima,
Thailand, April 2009.

[2] T. Higuchi, “Approach to An Irregular Time Series on the Basis
of the Fractal Theory,” *Physica D*, vol.31, 277–83, 1988.

**○**** **
**Thai Speech
Processing Based on Fractal Theory**

In the past, many
methods have been developed to analyze Thai speech signals [1]–[3],
e.g., energy, frequency, entropy, wavelet, etc. In this present study,
for the
first time, we try to investigate Thai speech signals based on the fractal
concept in order to show whether a complexity of vowels and consonants
can be distinguished. To determine the quantity’s complexity, fractal
dimension (FD) value is proposed due to it begin one of the indicative parameters most
widely used.

Example of Thai
sentence "ฉุกเฉิน /Chook/Chan/" means
"emergency" (click
this to hear it)

(Upper) Waveform in time domain (Lower) Time-frequency domain

The
method to estimate FD that we used in this study is the modified
zero-crossing rate (MZCR) method [4]. The main principle of MZCR is
the
assumption that high complexity can be easily found by obtaining the
highest rate of the
zero-crossing point. It means that we can directly compute the
complexity of speech signal on the basis of the zero-crossing rate
function. The FD can be defined by: FD = 2 + m where m is a scaling
parameter. In the MZCR method, there are three main steps to determine
the scaling parameter of FD. This computation is repeated over all
possible interval lengths (in practice, we suggest minimum length be 2^{4}-point
and maximum length be 2^{n-1}-point).

FD waveform in fractal
domain estimated by the MZCR method

By
applying the MZCR method with the windowing
process, we can show the trajectory of FD
values from the original speech signal with respect to the number of
windows (small segmentation).

Probability distribution of two separated syllables
are that "/ฉุก/เฉิน/
/Chook/Chan/"

(Left) "ฉะ/Ch/" consonant is marked in blue,
"อุก/ook/" vowel is marked in red

(Right) "ฉะ/Ch/" consonant is marked in blue,
"เออ/aa/" vowel is marked in red,
"น/n/" final consonant is marked in green

In Thai language, a
syllable consists of a consonant, vowel, tone, and final consonant.
This study shows that we can simply separate the components of Thai syllables
by using the FD values.

References

[1] N. Jittiwarangkul, et al., "Thai Syllable Segmentation for
Connected Speech Based on Energy", *IEEE APCCAS*, pp.169-172,
Nov. 1998.

[2] N. Satravaha, et al., "Tone Classification of Syllable-Segmented
Thai Speech on Multilayer Perceptron", *Proceedings of the 35th
Southeastern Symposium on System Theory*, pp.392- 396, Mar. 2003.

[3] K. Chamnongthai, at al., "Final Consonant Segmentation for Thai
Syllable by using Vowel Characteristics and Wavelet Packet
Transform", *ECTI Trans. Comp & Info Theo.* vol.1,no.1,
pp.50-62, May 2005.

[4] M. Phothisonothai and M. Nakagawa, "A Complexity Measure Based
on Modified Zero-Crossing Rate Function for Biomedical Signal
Processing", *
The 13th Int. Conf. on Biomed. Eng.
(ICBME2008)*, vol.23, pp. 240-244, Singapore 2008.

**○**** **
**Algorithm
Development for
Measuring Complexity of Time-Series Data**

A complexity
measure is a mathematical tool for analyzing time-series data in
many research fields. Various measures of complexity were developed
to compare time series and distinguish whether input time-series
data are regular, chaotic, or random behavior. In the field of
physiological signal analysis, complexity of heart and brain data
can distinguish emotion, imagination, movement, etc. This study
proposes a simple technique to measure quantity’s complexity on the
basis of the rate values of a zero-crossing point. The conventional
method, namely, Higuchi’s algorithm, has been selected for comparison
in this study. The obtained results show that this proposed method
is able to measure the complexity of time-series data by
estimating the Hurst exponent which presents as a negative value.

**○
Electroencephalogram Signal
Classification for Brain-Machine Interface**

Brain-Machine
Interface (BMI) enables the user to operate with devices
through electroencephalogram (EEG) signals. Research and
development on BMI technology has led to effective applications
in the real world, improving quality of life and reducing social
costs. Many BMI applications in the real world allow a user to
operate, e.g., virtual keyboards, cursors, wheelchairs, speech
synthesizers, and assistance appliances. It also gives the user
access to Internet, characters classifier, computer games and
brain-controlled robots. Moreover, we are also interested in
researching rehabilitation and assistive technology using
other biomedical signals (EKG, MEG, etc.), for examples.

**○
EEG
Signal Analysis Based on Fractal Concepts**

The objective
of this study is to analyze the spontaneous EEG signal
corresponding to body parts movement imagery tasks in terms of
fractal properties. Six algorithms of fractal dimension (FD)
estimators: box-counting algorithm, Higuchi algorithm, variance
fractal algorithm, detrended fluctuation analysis, power
spectral density analysis, and critical exponent analysis are
proposed in this experiment.

**○
Speech Enhancement
Using Intelligent Algorithms**

Background
noises interfere with communication devices such as mobile
telephones, digital hearing aids, etc. Therefore noise reduction
(NR) for limiting the effect of these noises is important.
The study proposes a noise reduction method based on soft
decision-making by the fuzzy inference system (FIS). The
different characteristics of noises frequently occurring are
used for creating the fuzzy decision rule base of the FIS. The
FIS has two input parameters: the average energy and the
difference of the average energy. The analysis of the FIS is
done in the domain of the perceptual wavelet packet transform (PWPT)
that is the human’s psychoacoustic model. The output of the FIS
is used to modify the PWPT coefficients in such
a way that it is more likely that the noise components are
reduced while the speech signal is enhanced. The enhanced speech
signal is the result of the inverse perceptual wavelet packet
transform (IPWPT) of the modified coefficients.