In this study we have worked on the classification of EEG signals produced by the exposure of primary colours (RGB). The main goal of this study was to perform an offline analysis and classification of color information obtained from EEG signals recorded in response to individual RGB colours presentation in order to verify our hypothesis, if the observation of different colors can be detected or not by selecting different frequency bands. We have also performed an offline analysis of EEG signals produced by the colour imagination to observe similarities in EEG signals between actual color exposure and their corresponding imagination in order to find a Way-In to further establish our argument for developing future BCI applications that utilizes colour information from EEG signals unlike the Wadsworth and Graz noninvasive BCI applications that utilizes sensory motor rhythm. It was seen that it is possible to detect the information, not only of actual colour exposure but also the information of colours imagination, from EEG signals. It was also seen that the colour information obtained through the imagination of colours was similar to the actual colour exposure in some subjects. The experiment was designed in a way to expose the colours to the subjects in random order of presentation and also their corresponding imaginations. Different features are extracted and analyzed. The EEG signals have to be classified into Red, Green and Blue classes. We have used Support Vector Machines with event-related spectral perturbation as features for the classification task using three different kernels, linear, polynomial and RBF which came out with the average classification accuracy of 84% with linear, 89% with polynomial and 97% with RBF kernel for real exposure of colors whereas for imagination of colors accuracy was 64%, 70% and 76% respectively. As an alternative, we have also performed extreme energy ratio (EER) and extreme energy difference (EED) criterions to extract energy features using only linear kernel with SVM. The classification was performed on three different groups of colors i.e. (Blue, Green), (Red, Green) and (Red, Blue). The accuracies found with both of EER and EED are (79%, 78% and 80%) and (82%, 83% and 84%) respectively for real exposure of colors and for imagination of colors are (72%, 70% and 73%) and (73%, 75% and 72%) respectively. EED performed better than EER. Another experiment was performed with different shapes of colors and the EEG data was categorized as four different groups for classification. In group1, the classification accuracies for circle, square and triangle are found to be (88%, 52%, 94%), (84%, 47%, 89%) and (84%, 49%, 94%) respectively as triplet (linear, polynomial, RBF). In group 2, 3 and 4 classification accuracies achieved are (71%, 50%, 94%), (60%, 48%, 92%) and (57%, 29%, 94%) respectively as triplet of (linear, polynomial, RBF) kernels. After the successful classification of colour information from EEG signals we are planning to work for online classification in order to implement with any possible future Brain-Computer Interface applications. We believe that this study could further be extended to find out the possibilities for e.g. simulating a scenario of traffic light signals in virtual environment or to identify and explore any possibility of analyzing the EEG signals and developing BCI applications for color blind and/or blind people. Since such applications are quite novel in their fields of BCI therefore requires extensive collaborative research work in different domains.

RECOGNITION OF PRIMARY COLOURS IN ELECTROENCEPHALOGRAPH SIGNALS USING SUPPORT VECTOR MACHINES / S. Rasheed ; tutor: Daniele Marini ; correlatore: Alessandro Rizzi ; direttore della Scuola di Dottorato in Informatica: Ernesto Damiani. DIPARTIMENTO DI INFORMATICA E COMUNICAZIONE, 2011 Mar 25. 23. ciclo, Anno Accademico 2010. [10.13130/rasheed-saim_phd2011-03-25].

RECOGNITION OF PRIMARY COLOURS IN ELECTROENCEPHALOGRAPH SIGNALS USING SUPPORT VECTOR MACHINES

S. Rasheed
2011

Abstract

In this study we have worked on the classification of EEG signals produced by the exposure of primary colours (RGB). The main goal of this study was to perform an offline analysis and classification of color information obtained from EEG signals recorded in response to individual RGB colours presentation in order to verify our hypothesis, if the observation of different colors can be detected or not by selecting different frequency bands. We have also performed an offline analysis of EEG signals produced by the colour imagination to observe similarities in EEG signals between actual color exposure and their corresponding imagination in order to find a Way-In to further establish our argument for developing future BCI applications that utilizes colour information from EEG signals unlike the Wadsworth and Graz noninvasive BCI applications that utilizes sensory motor rhythm. It was seen that it is possible to detect the information, not only of actual colour exposure but also the information of colours imagination, from EEG signals. It was also seen that the colour information obtained through the imagination of colours was similar to the actual colour exposure in some subjects. The experiment was designed in a way to expose the colours to the subjects in random order of presentation and also their corresponding imaginations. Different features are extracted and analyzed. The EEG signals have to be classified into Red, Green and Blue classes. We have used Support Vector Machines with event-related spectral perturbation as features for the classification task using three different kernels, linear, polynomial and RBF which came out with the average classification accuracy of 84% with linear, 89% with polynomial and 97% with RBF kernel for real exposure of colors whereas for imagination of colors accuracy was 64%, 70% and 76% respectively. As an alternative, we have also performed extreme energy ratio (EER) and extreme energy difference (EED) criterions to extract energy features using only linear kernel with SVM. The classification was performed on three different groups of colors i.e. (Blue, Green), (Red, Green) and (Red, Blue). The accuracies found with both of EER and EED are (79%, 78% and 80%) and (82%, 83% and 84%) respectively for real exposure of colors and for imagination of colors are (72%, 70% and 73%) and (73%, 75% and 72%) respectively. EED performed better than EER. Another experiment was performed with different shapes of colors and the EEG data was categorized as four different groups for classification. In group1, the classification accuracies for circle, square and triangle are found to be (88%, 52%, 94%), (84%, 47%, 89%) and (84%, 49%, 94%) respectively as triplet (linear, polynomial, RBF). In group 2, 3 and 4 classification accuracies achieved are (71%, 50%, 94%), (60%, 48%, 92%) and (57%, 29%, 94%) respectively as triplet of (linear, polynomial, RBF) kernels. After the successful classification of colour information from EEG signals we are planning to work for online classification in order to implement with any possible future Brain-Computer Interface applications. We believe that this study could further be extended to find out the possibilities for e.g. simulating a scenario of traffic light signals in virtual environment or to identify and explore any possibility of analyzing the EEG signals and developing BCI applications for color blind and/or blind people. Since such applications are quite novel in their fields of BCI therefore requires extensive collaborative research work in different domains.
25-mar-2011
Settore INF/01 - Informatica
EEG signals ; Electroencephalogram ; support vector machine ; Primary colors ; RGB colors ; Brain-Computer Interfaces
MARINI, DANIELE LUIGI ROBERTO
DAMIANI, ERNESTO
Doctoral Thesis
RECOGNITION OF PRIMARY COLOURS IN ELECTROENCEPHALOGRAPH SIGNALS USING SUPPORT VECTOR MACHINES / S. Rasheed ; tutor: Daniele Marini ; correlatore: Alessandro Rizzi ; direttore della Scuola di Dottorato in Informatica: Ernesto Damiani. DIPARTIMENTO DI INFORMATICA E COMUNICAZIONE, 2011 Mar 25. 23. ciclo, Anno Accademico 2010. [10.13130/rasheed-saim_phd2011-03-25].
File in questo prodotto:
File Dimensione Formato  
phd_unimi_R07932.pdf

accesso aperto

Tipologia: Tesi di dottorato completa
Dimensione 12.21 MB
Formato Adobe PDF
12.21 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/155486
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact