Research Article A System for True and False Memory ...

Research ArticleA System for True and False Memory Prediction Based on2D and 3D Educational Contents and EEG Brain Signals

Saeed Bamatraf1 Muhammad Hussain1 Hatim Aboalsamh1

Emad-Ul-Haq Qazi1 Amir Saeed Malik2 Hafeez Ullah Amin2 Hassan Mathkour1

Ghulam Muhammad3 and Hafiz Muhammad Imran4

1Department of Computer Science College of Computer and Information Sciences King Saud University Riyadh 11543 Saudi Arabia2Centre for Intelligent Signal and Imaging Research (CISIR) Department of Electrical amp Electronics EngineeringUniversiti Teknologi PETRONAS Bandar Seri Iskandar 32610 Tronoh Perak Malaysia3Department of Computer Engineering College of Computer and Information Sciences King Saud UniversityRiyadh 11543 Saudi Arabia4TeleNoc Riyadh Saudi Arabia

Correspondence should be addressed to Muhammad Hussain mhussainksuedusa

Received 14 September 2015 Revised 9 November 2015 Accepted 12 November 2015

Academic Editor Hasan Ayaz

Copyright copy 2016 Saeed Bamatraf et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

We studied the impact of 2D and 3D educational contents on learning and memory recall using electroencephalography (EEG)brain signals For this purpose we adopted a classification approach that predicts true and false memories in case of both shortterm memory (STM) and long term memory (LTM) and helps to decide whether there is a difference between the impact of 2Dand 3D educational contents In this approach EEG brain signals are converted into topomaps and then discriminative features areextracted from them and finally support vector machine (SVM) which is employed to predict brain states For data collection halfof sixty-eight healthy individuals watched the learning material in 2D format whereas the rest watched the same material in 3Dformat After learning task memory recall tasks were performed after 30 minutes (STM) and two months (LTM) and EEG signalswere recorded In case of STM 975 prediction accuracy was achieved for 3D and 966 for 2D and in case of LTM it was 100for both 2D and 3D The statistical analysis of the results suggested that for learning and memory recall both 2D and 3D materialsdo not have much difference in case of STM and LTM

1 Introduction

Themultimedia technology has widely taken over the educa-tion systembecause itmakes the concepts easy to understandThe multimedia principle by Mayer states that people learnmore deeply from words and pictures than from words alone[1] Currently the dominantmultimedia resources aremainlyin 2D form With recent advances in science and technology3D technology is becoming common and soon it will beavailable for educational purposes As such the educationalcontents can be presented to the students either in 2Dor in 3Dformat In 2D format 3D objects are visualized by projectingthem on 2D space With the advent of 3D technologynowadays 3D objects can be visualized as 3Dwith the help of

3D devices These advances in technology have now made uswonder whether the 3Dmaterial ismore effective than the 2Dcontent in terms of learning andmemory retention and recall

Electroencephalography (EEG) is an imaging techniquethat captures brain activation patterns detected as electricalactivity on the scalp [2 3] EEG signals reflect the brainbehavior during mental activities Since the procedure ofEEG is pain-free and noninvasive it has been widely usedin normal adults and children to study brain activationpatterns associated with tasks like memory retentionrecallperception attention and emotions

Some researchers studied the effects of 2D and stereo-scopic 3D on tasks like viewerrsquos experience of the movie con-tent [4] spatial cognition [5ndash7] deepening the understanding

Hindawi Publishing CorporationComputational Intelligence and NeuroscienceVolume 2016 Article ID 8491046 11 pageshttpdxdoiorg10115520168491046

2 Computational Intelligence and Neuroscience

of PC hardware [8] understanding and knowledge acqui-sition [9 10] spatial visualization skills [11] and educa-tion learning processes [12] These studies used subjectiveapproach without using EEG brain signals Also no oneexplicitly studied the effects of 2D and 3D educational con-tents on learning and memory recall Though the subjectiveapproach based on the statistical analysis of the answers ofquestions can be used to assess the effects of 2D and 3D edu-cational contents EEG signals give direct insight into brainstates and can lead to more accurate conclusions Intuitivelythe brain states are different while giving answers based onlearned information and guess

EEG technology has been employed by a number ofresearchers to study brain activations during different taskssuch as cognitive tasks [13] playing video games on largescreens [14] and playing video games on small and largedisplays [15] Some researchers studied the EEG brain signalsto locate brain regions and the components responsible formemory functions [16ndash19] Also some studies have been donerecently on attention learning andmemory using EEG brainactivations [20ndash23] To the best of our knowledge there donot exist any studies so far which focused on studying theimpact of 2D and 3D educational contents on learning andmemory recall using direct brain behavior through EEGtechnology

We studied the effects of 2D and 3D educational contentson memory retention and recall with the aid of brainimages (topomaps) captured using EEG technology For thispurpose we developed pattern recognition systems to pre-dict true memory (rememberedrecalled) and false memory(forgotten) using direct brain responses via topomaps for 2Dand 3Deducational contents and thenused them for assessingtheir effects on learning and memory retention and recallFor developing the systems the data was collected from sixty-eight healthy individuals Half of them watched the learningmaterial in 2D format whereas the second half watched thesame material in 3D format After learning task memoryretention and recall tasks were performed in the form ofmul-tiple choice questions (MCQs) related to the watched con-tents after 30 minutes (for STM) and two months (for LTM)and EEG signals were recorded

The proposed pattern recognition systems comprisefeature extraction and classification stages Features areextracted from EEG signals corresponding to correct answer(true memory) and wrong answer (false memory) and SVMis used for prediction For feature extraction a simple androbust technique has been introduced that first of all createstopomaps from EEG signals and removes redundancy usingcity-block distance One system was developed for 2D and3D each to predict true and false memories which can beused to predict the memory retention and recall abilities ofindividuals Each system encodes the brain states and theirperformance has been analyzed to decide whether there isa difference between the impacts of 2D and 3D educationalcontents onmemory retention and recall In case of STM themean prediction accuracy is 965 for 2D and 975 for 3Dwhereas it is 100 for both 3D and 2D for LTM Further statis-tical analysis of the results revealed that there is no significantdifference in using 3D and 2D materials as far as memory

retention and recall is concerned The initial results of thisstudy were presented in [24] this paper extends the study toLTM and gives thorough analysis of the proposed methods

The main contributions of the paper are

(i) a pattern recognition system for predicting the stateof memory whether the memory is true or false and

(ii) assessment of the effects of 2D and 3D educationalcontents on memory retention and recall

The remainder of the paper is organized as follows Theproposed methodology is elaborated in Section 2The resultsare presented and discussed in Section 3 and Section 4 givesthe statistical analysis Section 5 concludes the paper

2 Methods and Materials

We developed one pattern recognition system for 2D and 3Deach for assessing the brain states corresponding to the trueand false answers of MCQs that is true and false memoriesEach system was trained with the labeled EEG brain signalsthat encode the brain behavior during true answer (truememory) and false answer (false memory) As both thesystems have the same architecture and configuration andhave been trained in the sameway the trained systems encodethe brain behaviors corresponding to 2D and 3D contents andtheir prediction accuracies can be used to assess the effective-ness of 2D and 3D for learning and memory recall To assessthe effects of 2D and 3D we developed the hypothesis thatperformance of the system associated with 3D is better thanthat of 2D Precisely the hypothesis is modeled as follows

Null Hypothesis H0 Performance of the system asso-ciated with 3D is better than that of 2DAlternate Hypothesis H1 There is no difference in theperformance of the systems associated with 2D and3D

Using hypothesis testing we conclude whether the nullhypothesis holds or not

Our hypothesis about the difference between 2D and 3Dgroups is based on the following observations

(i) For data collection the subjects were selected in sucha way that the age IQ levels and the backgroundknowledge of the subjects in each group were same

(ii) Each group watched the same educational contentsand responded to the questionnaires containing thesame questions The only difference was that of 2Dand 3D formats

(iii) The brain states of each group are modeled as aclassification system which is based on the same fea-ture extraction technique and the same classificationmethod

(iv) Each system was implemented on the same platformusing the same environment and was learned in thesame way and was evaluated using the same proce-dure

Computational Intelligence and Neuroscience 3

EEG recording Preprocessing Topomaps

creation

Topomaps selection

Feature extraction and

selection

SVM classifier

SVM model

Learning

TestingPrediction

EEG signals

minus10minus8minus6minus4minus20246810

Figure 1 The proposed system for predicting true and false memories

(v) There is complete similarity from data collection tolearning and evaluating the systems If there is anydifference between 2D and 3D educational contentsin respect of memory retention and recall it shouldappear in the performance of the systems

In the rest of this section first we give a general overviewof the pattern recognition system Then we zoom into eachcomponent of the system Feature extraction is an importantcomponent of the system and discriminative features areneeded to represent the brain behavior reliably correspondingto the answer of a questionWe introduce a simple and robustmethod for feature extraction

21 Proposed System An overview of the proposed system isshown in Figure 1 First subjects are selected to participatein the learning and memory recall tasks EEG signals arerecorded from the subjects during the tasks These signalsare preprocessed to remove the artifacts and noisy signalsTopomaps are then created from the clean signals and afterthat the most representative topomaps are selected from thehuge number of topomaps per sample (question) Differentfeature extraction approaches are used to extract the featuresfrom topomaps Feature selection is needed to reduce thehigher dimensionality of feature spaces and selects the mostrepresentative features which subsequently are passed tothe classifier to produce the classification model Using thetrain classification model we assess true and false memoriesand afterwards analyze the effects of 2D and 3D educationalcontents on learning and memory retention and recall

In the following subsections we give details of each stepin the proposed system

211 Collection of Data The details related to the datacollection process are presented in this section

Selection of Participants A total number of volunteers whotook part in the experiments were 68 and their ages were inthe range of 18 to 30 yearsThe volunteers were having normalor close to normal eye sight and were not suffering from anyform of neurological disorders which could affect the resultsTwo groups were formed for 2D and 3D based on metrics ofage and their knowledge

Learning Contents and Tasks Related to Experiments Thesame learning contents for both 2D and 3D related to

the experiments in this study were acquired from Eureka3D system [25] The contents were of biology in nature andthe volunteers had no existing knowledge in biological areaLearning and memory recall were two building blocks of theexperiments performed In learning phase the participantswatched 2D or 3D learning contents depending upon theirgroup for the time spans of 8 to 10 minutes In the phase ofmemory recall two retention periods were employed that isone was for STM in which the retention period was of 30minutes and the other was for LTM in which the retentionperiodwas of twomonths In this process twentyMCQswereinquired of the participants and each MCQ had four choicesfor answers The same MCQs were asked to participants ofboth the groups that is 2D and 3D While performing therecall experiment EEG signals were recorded for study The41-inch TV screens were used for the display of learningcontents andMCQs of the recall phase and those were placedat the distance of 15m from the eyes of the participants Theparticipants had to answer the questions in the max time of30 seconds In the case of right answers ldquo1rdquo was assigned andin the opposite case ldquo0rdquo was assigned for the wrong ones

Recording of EEG Signals The recording of EEG signals wasdone in the period of 30 seconds which started when theparticipants were shown the questions and ended when theychose answers from the four choices for the answer Thestarting and ending time points of the time period specifiedfor answering the questionwere recorded in a file whichwereused eventually for the extraction of the part of EEG signalrelated to the question Sampling rates of 250 samplessec and128-channel HydroCel Geodesic Net (see Figure 2) were usedfor the recording of EEGdataTheEEG signals are spatiotem-poral data which are characterized by the temporal evolutionand the adjacent spatial potential distributions These signalsare represented with 119883(119896) isin R119872times119879 where 119896 is the questionidentifier119872 is the number of channels and 119879 is the numberof sampled data points along temporal evolution The selec-tion of electrodes was performed based on their location onthe scalp that is the outermost electrodes were omitted andin total 93 were chosen out of 128 electrodes for furtheranalysis

212 Preprocessing The noise in the form of artifacts that iseye blinking and so forth is present in the recorded EEGdatawhich adversely affects the performance of feature extraction


123

4

5

6

7

89

10

12

13

14

15

16

17

18

19

20

21

22Fp1

23

24F3

43

44

45T3

4647

38

39

41

42

32

34

35

37

25

26

27

28

29

30

31

48

49

50 51 52P3

5354

55

56 57LM

58T5

61

COM

6364 65

66

67

68

71

72

73

74

75Oz

76

77

78

79

80

81

82

84

86

87

88

89

90

92P4

93

SN

94

95

96T6

97

98

99

100RM

101

102

103

105106

107

108T4

110111112

113

115

116117118

114

119120121

122F8124

F4

125

126127

128 99Fp2

12333F7

36C3

40

5960

69

70O1 83

O2

8591

104C4REF

Cz

109

11Fz

62Pz

5

Figure 2 128-channel HydroCel Geodesic Net

and eventually the prediction accuracy It is utmost importantto remove such noise for the overall performance of thesystem In preprocessing phase such noise is removed Thenoise is present in various forms in the recorded EEG datawhich includes the EEG activity that is not the result ofresponse of stimuli noise due to the variability in ERP com-ponents is a result of neural and cognitive activity variationsanother common source of noise is the presence of bioelectricactivities like movement of eyes blinking movement ofmuscles and so forth and the final source of noise isdue to the electric equipment like display devices and so forth

The artifacts in the raw EEG data were detected using aband pass filter (1ndash48Hz) and then exported into MATLABfiles (mat) format using Net Station software of EGI Ocular

artifacts were removed by using Gratton and Coles method[26] and visual inspection

213 Creation of Topomaps One possibility for featureextraction is EEGpower spectral density but it is computed inFourier domainwhere time information is lostWe know thatbrain is a nonlinear dynamic system [27] and to keep track ofthe evolution of brain states over time is important As suchwe used EEG potential because it represents the changes ofbrain states over time Recent studies have shown that timedomain analysis plays an important role in understandingthe EEG signals [28ndash31] These studies have applied manytime analysis methods on EEG recorded during performingvarious tasks (such as problem solving [29]) and brain states


minus40

minus30

minus20

minus10

0

10

20

30

40

(a)minus30

minus20

minus10

0

10

20

30

(b)

Figure 3 Two sample topomaps (a) correct answer (b) incorrect answer

(eg emotional states [30] and epileptic seizure [32]) Hencewe have focused on the EEG potential rather than spectralanalysis

EEG signal at a particular time point 119905 is a collectionof the potentials at all electrode positions that is 119909119905 =(119909119905

1 119909119905

2 119909119905

3 119909

119905

119872) and is represented as a topomap (a digital

image describing brain activation at 119905) two such topomapsare shown in Figure 3 EEG signals corresponding to theanswer of each question are converted into topomaps Fea-tures can be extracted from the topomaps using image pro-cessing and analysis techniquesTheEEG signals are recordedfrom each subject while answering the questions and savedin a MATLAB raw file (mat) and the time points at whichthe subject starts answering each question until finishinganswering are saved in an event file Using mat file togetherwith the event file we create the topomaps correspondingto the EEG recordings of each question We used EEGLABtoolbox [33] for this step

The quantity of topomaps depends upon the time whichis taken by the participants for answering the questions Thenumber of topomaps corresponding to a question can be upto 7500 in total as the maximum time allocated for answeringa question is 30 sec and sampling rate is 250

214 Selection of Topomaps Topomapswhich occur consec-utively in time contain redundant information which do notaid in achieving good prediction accuracy The more the dis-criminative information is the more the prediction accuracymay be achieved In order to have maximum discriminativeinformation the consecutive topomaps are analyzed using adistancemetric and as a result themost discriminant ones areselected A number of distance metrics have been proposedin the research community but city-block distance metric iswell known for its simplicity and effectiveness and thereforewe chose it for the selection ofmost discriminative topomapsUsing city-block distance metric the distance between thepairs of topomaps is calculated and put in descending order

and finally the topomaps which are highly dissimilar areselected as shown in Figure 4

215 Feature Extraction The next step after removing theredundant information is to extract discriminative featuresfrom the selected topomaps First order statistics are used forthe extraction of features from topomaps

In order to discriminate the structures in images textureplays a major role Texture information from the selectedtopomaps is extracted using first order statistics meanstandard deviation entropy skewness and kurtosis Firstorder statistics calculated from a topomap are global featuresand less discriminative because the localization informationof the texture micropatterns is lost For capturing localizedtexture information each topomap is divided into a numberof blocks and then the statistical features are extractedfrom each block and merged into a vector that forms therepresentation of the topomap being processed This processis shown in Figure 5 For our experiments we tested differentnumbers of blocks but we found the acceptable number to be8 times 8 and 16 times 16 blocksWe examined different combinationsof five statistical features

Similarly we compute the feature vectors for each selectedtopomap and combine them into one feature vector thatrepresents the brain state while answering a question Wecan concatenate the vectors corresponding to all selectedtopomaps to form the feature vector but in this case thedimension of the feature space will be excessively largecausing curse of dimensionality problem For instance if 100topomaps are selected each topomap is divided into 16 times 16blocks and 5 features are calculated from each block and thenthe dimension of the feature space will be 100 times 16 times 16 times 5= 128000 We employ a different approach that reduces thedimension of the feature space significantly while keepingthe most discriminative information and not depending onthe number of selected topomaps Using this approach thedimension of feature space reduces to 4 times 16 times 16 times 5 = 5120for the above case details are given below


Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12

3 middot middotmiddotmiddot middot

middot

9091

92

93

94

95

96

97

98

99

10022 23 24

Figure 4 Selection of 119871 topomaps with the highest dissimilarity

After forming the vectors 1198811 1198812 119881119871 (each of dimen-sion 119904) which contain features from the 119871most discriminantselected topomaps related to one question the feature matrix(FM) of order (119904times119871) is built where the vectors1198811 1198812 119881119871form FM columns Equation (1a) shows FM for the case whenonly one feature (mean 120583) is computed from each block Nowfrom each row of this matrix the discriminative informationis captured by computing four first order statistics (meanstandard deviation skewness and kurtosis) see (1b) andthese are concatenated to form the feature vector as shownin (1a)

In (1a) FM of order 119904 times 119871 is the feature matrix for alltopomaps corresponding to a question and 119881 is the featurevector computed from matrix FM in (1b) from the 119894th rowof the matrix FM four statistical features are extracted

FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832

1sdot sdot sdot 120583

119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832

119904minus1sdot sdot sdot 120583119871

119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)

216 Feature Selection If a topomap is divided into 16 times 16blocks and five statistical features are computed from eachblock then each topomap will be represented by (5 times 16 times16 = 1280) features and the total number of features will be1280times 4 = 5120 per question which is highHigh dimensionalfeature space has a number of disadvantages like highcomputational cost and affects the classification accuracy It isutmost important to employ some feature selection method-ology in order to reduce the high dimension of the feature

ViB1

B2B3

B16

middot middot middot

The ith topomap

Computationof first order

statistics fromeach block

120583i1

120575i1

ki1

ski1

ei1

ei16

Figure 5 Features extraction from 4 times 4 blocks of topomaps

space by reducing the redundancy and selecting the mostdiscriminant features In the proposed approach ROC curveis used to select the most important features The receiveroperating characteristic (ROC) curve measures the class sep-arability of a certain feature It quantifies the overlap betweenthe distributions of the feature in two classes and is computedas an area between two curves This area is zero for completeoverlap and is 05 for complete separation A feature is dis-criminatory if the area for this feature is close to 05 for detailssee [34]

217 Classification Learning and memory recall processincorporates two classes of data that is correct and incorrectanswers For this problemmany learningmodels can be usedwhich include but are not limited to artificial neural networksdecision trees and support vector machines (SVMs) Out ofthese support vector machines are regarded as the-state-of-the-art classification models which can achieve outstandingaccuracies This is a linear classifier which achieves themaximum margin between the classes of data and can workfor the very few training samples and at the same time worksfor classifying nonlinear data as well by taking the data tohigher dimensional space using kernel trick Various kinds ofkernels can be usedwith SVM like radial basis function (RBF)and polynomial As RBF kernel is known to give good resultswe choose it for the classification purposesThe parameters ofthe kernel are learned using grid-searchmethod and the well-known LIBSVM [35] library is used for the implementationof SVM Note that SVM is useful for applications where thenumber of features ismuch larger than the number of samples[36ndash39]


3 Results and Discussion

In this section the results of the proposed approach arepresented and subsequently discussed First we describe theevaluation policy that we used to evaluate the systems Thenwe present an overview of the subjects and their answers tothe questions in both STM and LTM sessions After that wepresent the result of the methods we used on STM and LTMIn the next section we give the statistical analysis of the resultof STM and LTM for 2D and 3D to see if there is a differencebetween them or not

31 Evaluation Policy In this section the details about thedataset used for experiments are presented along with theevaluationmethodology and the metrics used to measure theprediction accuracies are discussed

For the experiments performed the dataset is selectedto make sure that the samples used for training and testingphases fairly represent the two classes The selected setcomprises 200 questions out of which 100 are from correctclassThe sameprocedure is used for data selection in both 2Dand 3D learning contents for both the cases of STM and LTM

Classification is performed between subjects so that thesystem is general and does not depend on a particular subjectEach system is learned and tested using the data acrossdifferent subjects of the same group

In order to estimate the prediction accuracy of the classi-fiers we used 10-fold cross validation [40] Cross validationis a renowned validation technique that is employed for suchpurpose In 10-fold cross validation data is randomly parti-tioned into 10 parts in which training is performed on nineparts and the left-over fold is used for testing purposes Thisprocedure is applied 10 times and in each turn different fold isused for testing purposes while the rest are used for trainingAt the end to estimate the prediction accuracy of the systemmean and standard deviation of 10 iterations are calculatedThis is the-state-of-the-art methodology for estimating theprediction accuracy of a classifier and ensures there is nooverfitting because every time the system is trained withdifferent data set and tested with different data set not usedin training

In order to measure the performance of the systemsthe well-known metrics of accuracy and the AUC wereused As we collected the data corresponding to 2D and 3Deducational contents in the same way and also used the sameclassification models which were trained and tested in thesame way the prediction performance of the classifiers isgoing to help in investigating the effectiveness of 2D or 3Dlearning contents for learning and memory recall

32 Results for Long Term Memory (LTM) We modeledtwo systems one for 2D educational content and one for3D case We trained and tested each system using themethod described in Section 31 In this case each systeminvolves three parameters which include number of blocksthe extracted features from each block and the number ofselected topomaps

Initially we selected 20 topomaps and partitioned eachtopomap into 8 times 8 and 16 times 16 blocks The results are shown

Table 1 Results with 8 times 8 blocks and 20 selected topomaps

features Accuracy AUC2D 3D 2D 3D

1 74 plusmn 108 81 plusmn 94 073 plusmn 013 084 plusmn 0112 915 plusmn 58 905 plusmn 64 090 plusmn 009 089 plusmn 0113 90 plusmn 47 89 plusmn 57 093 plusmn 007 090 plusmn 0054 895 plusmn 37 895 plusmn 6 090 plusmn 008 091 plusmn 0075 915 plusmn 47 885 plusmn 47 093 plusmn 006 093 plusmn 005










in Tables 1 and 2 The best accuracy obtained in case of 8 times 8is 905 for 3D with two features and 915 for 2D with fivefeatures and it is 915 for 3D with two features and 955for 2D with five features in case of 16 times 16 blocks In case of16 times 16 blocks there is little increase in the accuracy but theoverall trend is the same that is the best accuracy for 3D caseis obtainedwith 2 features and that for 2D case is with featuresper block

Next the number of selected topomaps was increased to30 and features are extracted from 8 times 8 and 16 times 16 blocksThe results are shown in Tables 3 and 4 The best accuracyobtained in case of 8 times 8 is 965 for 3D with three featuresand 925 for 2D with three features and it is 96 for 3Dwith two features and 975 for 2D with four features in caseof 16 times 16 blocks It indicates that by increasing the numberof selected topomaps the accuracy increases




1 945 plusmn 5 97 plusmn 42 094 plusmn 007 097 plusmn 0042 985 plusmn 24 100 099 plusmn 002 13 985 plusmn 24 995 plusmn 16 098 plusmn 004 099 plusmn 0014 99 plusmn 21 99 plusmn 21 099 plusmn 004 099 plusmn 0035 99 plusmn 21 99 plusmn 21 099 plusmn 002 1



1 100 100 1 12 995 plusmn 16 100 099 plusmn 001 13 995 plusmn 16 100 1 14 995 plusmn 16 995 plusmn 16 1 15 99 plusmn 21 995 plusmn 16 099 plusmn 002 1

In view of this trend we increased the number of selectedtopomaps to 50 100 120 and 150 With 120 and 150 selectedtopomaps we got the maximum accuracy The results with150 topomaps are shown in Tables 5 and 6 The best meanaccuracy of 100 was achieved with 16 times 16 blocks and onefeature (ie mean) per block for both 2D and 3D cases

Generally the more the number of topomaps and thenumber of blocks the better the accuracy for both 2D and 3Dcases as shown in Figures 6 and 7 The best average accuracy(100) was achieved by 120 and 150 topomaps The numberof selected topomaps which gives the best accuracy is muchlower than the total number of topomaps per question (less orequal 7500) It indicates that the discriminative informationabout the brain behavior for true or false memory is embed-ded in few topomaps Further the best performing systemsselect 150 discriminative topomaps divide each topomap into16times 16 blocks and compute one feature (ie mean) from eachblockTheother performancemetric that is AUC also showsthe similar performance for both 2D and 3D cases

The reason that the average accuracy of 100 was reachedin case of LTM is that after two months of retention subjectseither still remember the answers or forgot them at all that isclear separation of correct and incorrect answers

33 Results for Short Term Memory (STM) The results forSTM were presented in [24] the details can be found thereWe tried 20 30 50 100 120 and 150 topomaps with 8 times 8 and16 times 16 blocks We present here only the results of the bestcase which were obtained using 150 topomaps The resultsare shown in Tables 7 and 8 The best accuracy obtained incase of 8 times 8 is 975 for 3D with three features and 955 for2D with two features and it is 975 for 3D with two featuresand 965 for 2D with also two features in case of 16 times 16blocks In this case the configuration of the best performingsystems for the prediction of true and false memories is 150most discriminative topomaps 16 times 16 block division andtwo features (mean and standard deviation) from each block

2D accuracy for different number of topomaps of different block divisions using 1 statistical feature

60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150

16 times 168 times 8

Number of blocks

Figure 6 The best case accuracies in case of LTM for differentselected topomaps for 2D educational material


70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks


It was observed that prediction accuracy is directly pro-portional to the number of selected topomaps and reached toits peak for 150 topomaps The primary factor for this behav-ior is that when higher numbers of topomaps are selectedwe get more discriminative information The results of thebest case for all the selected topomaps with two differentblock divisions have been shown in a comprehensive way inFigures 8 and 9 Generally themore the number of topomapsand the number of blocks the more the accuracy of 2D and3D

34 STM + LTM Now we merged the 200 samples forSTM with the 200 samples for LTM together and extract the


Table 7 Results of 8 times 8 blocks using 150 selected topomaps






statistical features The average prediction accuracy is stillexcellent we got an average accuracy of 97 for both 2Dand 3D with the use of five features This indicates that theproposed system is robust and is not affected by the size ofthe data

4 Which Is Better 2D or 3D

In order to come to any conclusion about the effectivenessof 2D or 3D educational contents for learning and mem-ory retentionrecall we tested the hypothesis described inSection 3 For this purpose we need enough samples ofprediction accuracy values To achieve this we executed thesystems five times with 10-fold cross validation using thebest parameters and every time randomizing the datasetsso that we can get average accuracy values for 50 sampleshaving different combinations of true and false memoriesAfter getting 50 accuracy values for the systems developedfor 2D and 3D we used SPSS for hypothesis testing

Using SPSS software an independent 119905-test was applieddepending on the assumption that normality is acceptableThe results for 2D and 3D groups are approximately samethat is in case of 2D M is 966 and SD is 37 while in 3D caseM is 972 and SD is 33The value of 119905 is minus0847 and 119901 is 0399which is greater than 005 Therefore the null hypothesisis rejected and it can be said that statistically 2D and 3Dcontents are same for learning and memory recall in the caseof STM Similar tests were performed in the case of LTM andsame can also be said that 2D and 3D contents are statisticallythe same for learning and memory recall In the case of 2DM is 994 and SD is 16 and for 3D M is 993 and SD is 20while the value of 119901 is 0787 and 119905 is 0271

The observed results did not contradict with the resultsbased on the answers to the questions of the 2D and 3Dcontent for both STM and LTM If we analyze the results wefind that in case of STM the difference between the correct

203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy

2D accuracy for different number of topomaps of different block divisions using 2 statistical features

Figure 8 The best case accuracies in case of STM for differentselected topomaps for 2D educational material


203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks


answers is 23 for 2D and 3D group which is equal to 3 of allanswers whereas this difference is only 3 (ie 0468) in caseof LTMThis percentage is not significantly different if we takeinto account that some correct answers may be due to guesssimply and not true memory This supports our finding that2D and 3D educational contents are approximately same forlearning andmemory retention and recallThese findings aresimilar to the previous subjective findings on the effectivenessof 2D and 3D contents on learning and knowledge acquisition[11 14] and education learning processes [21] without usingEEG technology


5 Conclusion

We studied the effectiveness of 2D and 3D educationalcontents on learning andmemory retentionrecall using EEGbrain signals For this purpose we proposed pattern recog-nition systems for predicting true and false memories andthen used them for assessing the effects of 2D and 3D edu-cational contents on learning and memory retentionrecallby analyzing the brain behavior during learning andmemoryrecall tasks using EEG signals

For modeling pattern recognition systems first we col-lected the data Sixty-eight healthy volunteers participated indata collection Two groups corresponding to 2Dand 3Dwereformed which were based on the participantsrsquo ages and theirknowledgeThedata collectionwas done in twophases learn-ing andmemory recall During the learning phase dependingupon the group the participants watched 2D or 3D contentsfor durations of 8 to 10 minutes In order to check memoryrecall two retention periods were used that is 30-minuteperiod in case of STM and 2-month period in case of LTMEach participant answered 20 MCQs which were same forboth 2D and 3D groups and were about the learned contentsDuring the recall process EEG signals were recorded

For the pattern recognition systems we introduced a sim-ple and robust feature extraction technique that first convertsthe EEG signals (corresponding to the response of a question)into topomaps selects the most discriminative topomapsand then captures the localized texture information from theselected topomaps For classification we employed SVMwithRBF kernel The proposed systems can reliably predict trueand false memories employing the direct brain activations incase of STM and LTM We developed separate systems withthe same architecture and configuration for 2Dand 3Deduca-tionalmaterials As each system encodes the brain activationsduringmemory retentionrecall tasks the performance of thesystem can be analyzed to assess the impact of 2D and 3Deducational materials Statistical analysis reveals that thereis no significant difference between the effects of 2D and3D educational materials on learning and memory reten-tionrecall Our findings are in accordance with the previousstudies (without using direct brain activations) on the effectsof 2D and 3D materials However our approach is based onthe pattern recognition systems As such our research has theadded advantage of proposing pattern recognition systemswhich can be used to predict true and falsememories and canbe adopted to assess the learning levels of students Thoughthere is no difference in the effectiveness of 2D and 3D edu-cational contents on learning and memory retentionrecallfrom the performance perspective different brain regionscan be activated for 2D and 3D educational material Toinvestigate which brain regions are activated for 2D and 3Deducational materials are the subject for future work

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

This project was supported by NSTIP strategic technologiesprograms Grant no 12-INF2582-02 in the Kingdomof SaudiArabia

References

[1] R E Mayer and R Moreno ldquoA cognitive theory of multimedialearning implications for design principlesrdquo Journal of Educa-tional Psychology vol 91 no 2 pp 358ndash368 1998

[2] M Teplan ldquoFundamentals of EEGmeasurementrdquoMeasurementScience Review vol 2 no 2 pp 1ndash11 2002

[3] E Niedermeyer and F L da Silva Electroencephalography BasicPrinciples Clinical Applications and Related Fields LippincottWilliams ampWilkins 2005

[4] L M Carrier S S Rab L D Rosen L Vasquez and N ACheever ldquoPathways for learning from 3D technologyrdquo Interna-tional Journal of Environmentalamp Science Education vol 7 no 1pp 53ndash69 2012

[5] A Cockburn and B McKenzie ldquoEvaluating spatial memory intwo and three dimensionsrdquo International Journal of Human-Computer Studies vol 61 no 3 pp 359ndash373 2004

[6] A Price and H-S Lee ldquoThe effect of two-dimensional andstereoscopic presentation on middle school studentsrsquo perfor-mance of spatial cognition tasksrdquo Journal of Science Educationand Technology vol 19 no 1 pp 90ndash103 2010

[7] M Tavanti and M Lind ldquo2D vs 3D implications on spatialmemoryrdquo in Proceedings of the IEEE Symposium on InformationVisualization (INFOVIS rsquo01) pp 139ndash145 IEEE San DiegoCalif USA October 2001

[8] A Mukai Y Yamagishi M J Hirayama T Tsuruoka and TYamamoto ldquoEffects of stereoscopic 3D contents on the processof learning to build a handmade PCrdquo Knowledge Managementamp E-Learning vol 3 no 3 pp 491ndash506 2011

[9] T Huk and C Floto ldquoComputer-animations in education theimpact of graphical quality (3D2D) and signalsrdquo in Proceedingsof the World Conference on E-Learning in Corporate Govern-ment Healthcare and Higher Education pp 1036ndash1037 2003

[10] R M Rias andW Khadija Yusof ldquoAnimation and prior knowl-edge in amultimedia application a case study on undergraduatecomputer science students in learningrdquo in Proceedings of the2nd International Conference on Digital Information and Com-munication Technology and itrsquos Applications (DICTAP rsquo12) pp447ndash451 Bangkok Thailand May 2012

[11] H-C Wang C-Y Chang and T-Y Li ldquoThe comparativeefficacy of 2D-versus 3D-based media design for influencingspatial visualization skillsrdquo Computers in Human Behavior vol23 no 4 pp 1943ndash1957 2007

[12] F L Perez S I C Vanegas and P E C Cortes ldquoEvaluation oflearning processes applying 3D models for constructive pro-cesses from solutions to problemsrdquo in Proceedings of the 2ndInternational Conference on Education Technology and Com-puter (ICETC rsquo10) pp V3-337ndashV3-341 IEEE Shanghai ChinaJune 2010

[13] H U Amin A SMalik N Badruddin andW-T Chooi ldquoBrainactivation during cognitive tasks an overview of EEG and fMRIstudiesrdquo in Proceedings of the 2nd IEEE-EMBS Conference onBiomedical Engineering and Sciences (IECBES rsquo12) pp 950ndash953IEEE Langkawi Malaysia December 2012

[14] A S Malik D A Osman A A Pauzi and R N H R Khairud-din ldquoInvestigating brain activationwith respect to playing video


games on large screensrdquo in Proceedings of the 4th InternationalConference on Intelligent and Advanced Systems (ICIAS rsquo12) pp86ndash90 IEEE June 2012

[15] A S Malik A A Pauzi D A Osman and R N H RKhairuddin ldquoDisparity in brain dynamics for video gamesplayed on small and large displaysrdquo in Proceedings of the IEEESymposium on Humanities Science and Engineering Research(SHUSER rsquo12) pp 1067ndash1071 IEEE Kuala Lumpur MalaysiaJune 2012

[16] C Babiloni F Babiloni F Carducci et al ldquoHuman corti-cal responses during one-bit short-term memory A high-resolution EEG study on delayed choice reaction time tasksrdquoClinical Neurophysiology vol 115 no 1 pp 161ndash170 2004

[17] W Klimesch ldquoEEG alpha and theta oscillations reflect cogni-tive and memory performance a review and analysisrdquo BrainResearch Reviews vol 29 no 2-3 pp 169ndash195 1999

[18] X-L Chen B Yao and D-Y Wu ldquoApplication of EEG non-linear analysis in vision memory studyrdquo Chinese Journal ofRehabilitation Theory and Practice vol 6 p 12 2006

[19] H U Amin A S Malik N Badruddin and W-T ChooildquoBrain behavior in learning and memory recall process a high-resolution EEG analysisrdquo in The 15th International Conferenceon Biomedical Engineering vol 43 of IFMBE Proceedings pp683ndash686 Springer 2014

[20] B A Coffman V P Clark and R Parasuraman ldquoBattery pow-ered thought enhancement of attention learning andmemoryin healthy adults using transcranial direct current stimulationrdquoNeuroImage vol 85 pp 895ndash908 2014

[21] B-M Gu H van Rijn andW H Meck ldquoOscillatory multiplex-ing of neural population codes for interval timing and workingmemoryrdquo Neuroscience and Biobehavioral Reviews vol 48 pp160ndash185 2015

[22] E L Johnson and R T Knight ldquoIntracranial recordings andhuman memoryrdquo Current Opinion in Neurobiology vol 31 pp18ndash25 2015

[23] L-T Hsieh and C Ranganath ldquoFrontal midline theta oscil-lations during working memory maintenance and episodicencoding and retrievalrdquo NeuroImage vol 85 pp 721ndash729 2014

[24] S Bamatraf M Hussain H Aboalsamh et al ldquoA system basedon 3D and 2D educational contents for true and false memoryprediction using EEG signalsrdquo in Proceedings of the 7th Interna-tional IEEEEMBS Conference on Neural Engineering (NER rsquo15)pp 1096ndash1099 IEEE Montpellier France April 2015

[25] Eurekain httpswwwdesignmatecom[26] G GrattonM G H Coles and E Donchin ldquoA newmethod for

off-line removal of ocular artifactrdquo Electroencephalography andClinical Neurophysiology vol 55 no 4 pp 468ndash484 1983

[27] D P Subha P K Joseph R Acharya and C M Lim ldquoEEG sig-nal analysis a surveyrdquo Journal of Medical Systems vol 34 no 2pp 195ndash212 2010

[28] W Lian R Talmon H Zaveri L Carin and R CoifmanldquoMultivariate time-series analysis and diffusion mapsrdquo SignalProcessing vol 116 pp 13ndash28 2015

[29] M H Dıaz F M Cordova L Canete et al ldquoOrder and chaos inthe brain fractal time series analysis of the EEG activity duringa cognitive problem solving taskrdquo Procedia Computer Sciencevol 55 pp 1410ndash1419 2015

[30] S Akdemir Akar S Kara S Agambayev and V Bilgic ldquoNonlin-ear analysis of EEGs of patients with major depression duringdifferent emotional statesrdquo Computers in Biology and Medicinevol 67 pp 49ndash60 2015

[31] Z Li X Jin and X Zhao ldquoDrunk driving detection basedon classification of multivariate time seriesrdquo Journal of SafetyResearch vol 54 pp 61e29ndash64e29 2015

[32] U R Acharya F Molinari S V Sree S Chattopadhyay K-HNg and J S Suri ldquoAutomated diagnosis of epileptic EEG usingentropiesrdquo Biomedical Signal Processing and Control vol 7 no4 pp 401ndash408 2012

[33] A Delorme and S Makeig ldquoEEGLAB an open source toolboxfor analysis of single-trial EEG dynamics including indepen-dent component analysisrdquo Journal of NeuroscienceMethods vol134 no 1 pp 9ndash21 2004

[34] S Theodoridis and K Koutroumbas ldquoFeature selectionrdquo inPattern Recognition S Theodoridis and K Koutroumbas Edschapter 5 pp 261ndash322 Academic Press BostonMass USA 4thedition 2009

[35] C-C Chang and C-J Lin ldquoLIBSVM a Library for supportvector machinesrdquo ACM Transactions on Intelligent Systems andTechnology vol 2 no 3 article 27 2011

[36] C-W Hsu C-C Chang and C-J Lin ldquoA practical guide tosupport vector classificationrdquo Bioinformatics vol 1 no 1 2003

[37] T Joachims ldquoText categorizationwith support vectormachineslearning with many relevant featuresrdquo in Machine LearningECML-98 vol 1398 of Lecture Notes in Computer Science pp137ndash142 Springer Berlin Germany 1998

[38] O Chapelle P Haffner and V N Vapnik ldquoSupport vec-tor machines for histogram-based image classificationrdquo IEEETransactions on Neural Networks vol 10 no 5 pp 1055ndash10641999

[39] C Cortes and V Vapnik ldquoSupport-vector networksrdquo MachineLearning vol 20 no 3 pp 273ndash297 1995

[40] R Kohavi ldquoA study of cross-validation and bootstrap foraccuracy estimation and model selectionrdquo in Proceedings of the14th International Joint Conference on Artificial Intelligence(IJCAI rsquo95) vol 2 pp 1137ndash1143Montreal Canada August 1995

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



of PC hardware [8] understanding and knowledge acqui-sition [9 10] spatial visualization skills [11] and educa-tion learning processes [12] These studies used subjectiveapproach without using EEG brain signals Also no oneexplicitly studied the effects of 2D and 3D educational con-tents on learning and memory recall Though the subjectiveapproach based on the statistical analysis of the answers ofquestions can be used to assess the effects of 2D and 3D edu-cational contents EEG signals give direct insight into brainstates and can lead to more accurate conclusions Intuitivelythe brain states are different while giving answers based onlearned information and guess

EEG technology has been employed by a number ofresearchers to study brain activations during different taskssuch as cognitive tasks [13] playing video games on largescreens [14] and playing video games on small and largedisplays [15] Some researchers studied the EEG brain signalsto locate brain regions and the components responsible formemory functions [16ndash19] Also some studies have been donerecently on attention learning andmemory using EEG brainactivations [20ndash23] To the best of our knowledge there donot exist any studies so far which focused on studying theimpact of 2D and 3D educational contents on learning andmemory recall using direct brain behavior through EEGtechnology

We studied the effects of 2D and 3D educational contentson memory retention and recall with the aid of brainimages (topomaps) captured using EEG technology For thispurpose we developed pattern recognition systems to pre-dict true memory (rememberedrecalled) and false memory(forgotten) using direct brain responses via topomaps for 2Dand 3Deducational contents and thenused them for assessingtheir effects on learning and memory retention and recallFor developing the systems the data was collected from sixty-eight healthy individuals Half of them watched the learningmaterial in 2D format whereas the second half watched thesame material in 3D format After learning task memoryretention and recall tasks were performed in the form ofmul-tiple choice questions (MCQs) related to the watched con-tents after 30 minutes (for STM) and two months (for LTM)and EEG signals were recorded

The proposed pattern recognition systems comprisefeature extraction and classification stages Features areextracted from EEG signals corresponding to correct answer(true memory) and wrong answer (false memory) and SVMis used for prediction For feature extraction a simple androbust technique has been introduced that first of all createstopomaps from EEG signals and removes redundancy usingcity-block distance One system was developed for 2D and3D each to predict true and false memories which can beused to predict the memory retention and recall abilities ofindividuals Each system encodes the brain states and theirperformance has been analyzed to decide whether there isa difference between the impacts of 2D and 3D educationalcontents onmemory retention and recall In case of STM themean prediction accuracy is 965 for 2D and 975 for 3Dwhereas it is 100 for both 3D and 2D for LTM Further statis-tical analysis of the results revealed that there is no significantdifference in using 3D and 2D materials as far as memory

retention and recall is concerned The initial results of thisstudy were presented in [24] this paper extends the study toLTM and gives thorough analysis of the proposed methods

The main contributions of the paper are

(i) a pattern recognition system for predicting the stateof memory whether the memory is true or false and

(ii) assessment of the effects of 2D and 3D educationalcontents on memory retention and recall

The remainder of the paper is organized as follows Theproposed methodology is elaborated in Section 2The resultsare presented and discussed in Section 3 and Section 4 givesthe statistical analysis Section 5 concludes the paper

2 Methods and Materials

We developed one pattern recognition system for 2D and 3Deach for assessing the brain states corresponding to the trueand false answers of MCQs that is true and false memoriesEach system was trained with the labeled EEG brain signalsthat encode the brain behavior during true answer (truememory) and false answer (false memory) As both thesystems have the same architecture and configuration andhave been trained in the sameway the trained systems encodethe brain behaviors corresponding to 2D and 3D contents andtheir prediction accuracies can be used to assess the effective-ness of 2D and 3D for learning and memory recall To assessthe effects of 2D and 3D we developed the hypothesis thatperformance of the system associated with 3D is better thanthat of 2D Precisely the hypothesis is modeled as follows

Null Hypothesis H0 Performance of the system asso-ciated with 3D is better than that of 2DAlternate Hypothesis H1 There is no difference in theperformance of the systems associated with 2D and3D

Using hypothesis testing we conclude whether the nullhypothesis holds or not

Our hypothesis about the difference between 2D and 3Dgroups is based on the following observations

(i) For data collection the subjects were selected in sucha way that the age IQ levels and the backgroundknowledge of the subjects in each group were same

(ii) Each group watched the same educational contentsand responded to the questionnaires containing thesame questions The only difference was that of 2Dand 3D formats

(iii) The brain states of each group are modeled as aclassification system which is based on the same fea-ture extraction technique and the same classificationmethod

(iv) Each system was implemented on the same platformusing the same environment and was learned in thesame way and was evaluated using the same proce-dure



creation

Topomaps selection


selection

SVM classifier

SVM model

Learning

TestingPrediction

EEG signals














123

4

5

6

7

89

10

12

13

14

15

16

17

18

19

20

21

22Fp1

23

24F3

43

44

45T3

4647

38

39

41

42

32

34

35

37

25

26

27

28

29

30

31

48

49

50 51 52P3

5354

55

56 57LM

58T5

61

COM

6364 65

66

67

68

71

72

73

74

75Oz

76

77

78

79

80

81

82

84

86

87

88

89

90

92P4

93

SN

94

95

96T6

97

98

99

100RM

101

102

103

105106

107

108T4

110111112

113

115

116117118

114

119120121

122F8124

F4

125

126127

128 99Fp2

12333F7

36C3

40

5960

69

70O1 83

O2

8591

104C4REF

Cz

109

11Fz

62Pz

5







minus40

minus30

minus20

minus10

0

10

20

30

40

(a)minus30

minus20

minus10

0

10

20

30

(b)




1 119909119905

2 119909119905

3 119909

119905










Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12


middot

9091

92

93

94

95

96

97

98

99

10022 23 24




FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832


119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832


119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)


ViB1

B2B3

B16


The ith topomap



120583i1

120575i1

ki1

ski1

ei1

ei16








































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





creation

Topomaps selection


selection

SVM classifier

SVM model

Learning

TestingPrediction

EEG signals














123

4

5

6

7

89

10

12

13

14

15

16

17

18

19

20

21

22Fp1

23

24F3

43

44

45T3

4647

38

39

41

42

32

34

35

37

25

26

27

28

29

30

31

48

49

50 51 52P3

5354

55

56 57LM

58T5

61

COM

6364 65

66

67

68

71

72

73

74

75Oz

76

77

78

79

80

81

82

84

86

87

88

89

90

92P4

93

SN

94

95

96T6

97

98

99

100RM

101

102

103

105106

107

108T4

110111112

113

115

116117118

114

119120121

122F8124

F4

125

126127

128 99Fp2

12333F7

36C3

40

5960

69

70O1 83

O2

8591

104C4REF

Cz

109

11Fz

62Pz

5







minus40

minus30

minus20

minus10

0

10

20

30

40

(a)minus30

minus20

minus10

0

10

20

30

(b)




1 119909119905

2 119909119905

3 119909

119905










Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12


middot

9091

92

93

94

95

96

97

98

99

10022 23 24




FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832


119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832


119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)


ViB1

B2B3

B16


The ith topomap



120583i1

120575i1

ki1

ski1

ei1

ei16








































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




123

4

5

6

7

89

10

12

13

14

15

16

17

18

19

20

21

22Fp1

23

24F3

43

44

45T3

4647

38

39

41

42

32

34

35

37

25

26

27

28

29

30

31

48

49

50 51 52P3

5354

55

56 57LM

58T5

61

COM

6364 65

66

67

68

71

72

73

74

75Oz

76

77

78

79

80

81

82

84

86

87

88

89

90

92P4

93

SN

94

95

96T6

97

98

99

100RM

101

102

103

105106

107

108T4

110111112

113

115

116117118

114

119120121

122F8124

F4

125

126127

128 99Fp2

12333F7

36C3

40

5960

69

70O1 83

O2

8591

104C4REF

Cz

109

11Fz

62Pz

5







minus40

minus30

minus20

minus10

0

10

20

30

40

(a)minus30

minus20

minus10

0

10

20

30

(b)




1 119909119905

2 119909119905

3 119909

119905










Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12


middot

9091

92

93

94

95

96

97

98

99

10022 23 24




FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832


119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832


119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)


ViB1

B2B3

B16


The ith topomap



120583i1

120575i1

ki1

ski1

ei1

ei16








































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




minus40

minus30

minus20

minus10

0

10

20

30

40

(a)minus30

minus20

minus10

0

10

20

30

(b)




1 119909119905

2 119909119905

3 119909

119905










Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12


middot

9091

92

93

94

95

96

97

98

99

10022 23 24




FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832


119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832


119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)


ViB1

B2B3

B16


The ith topomap



120583i1

120575i1

ki1

ski1

ei1

ei16








































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Topomapscreation

Topomapsselection

L (L ≪ N)N

12

3

12


middot

9091

92

93

94

95

96

97

98

99

10022 23 24




FM

11988111198812sdot sdot sdot 119881

119871

(((((((((((((((

(

1205831

11205832


119871

1

1205831

21205832


119871

2

1205831

31205832


119871

3

sdot sdot sdot

sdot sdot sdot

sdot sdot sdot

1205831

119904minus11205832


119904minus1

1205831

1199041205832


119871

119904

)))))))))))))))

)119904times119871

997888rarr

119881

(((((((((((((((

(

1205831

1205751

1199041198961

1198961

120583119904

120575119904

119904119896119904

119896119904

)))))))))))))))

)4119904times1

(1a)

[1205831

119894 1205832

119894 120583

119871

119894] 997888rarr [120583

119894 120575119894 119904119896119894 119896119894]119879 (1b)


ViB1

B2B3

B16


The ith topomap



120583i1

120575i1

ki1

ski1

ei1

ei16








































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in







































60

65

70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks



70

75

80

85

90

95

100

Accu

racy

203050

100120150


Number of blocks
















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in















203050

100120150


Number of blocks

75

70

65

60

80

85

90

95

100

Accu

racy




203050

100120150

75

80

85

90

95

100

Accu

racy


Number of blocks




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




5 Conclusion






Acknowledgment


References


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



Research Article A System for True and False Memory ...

Documents

Transcript of Research Article A System for True and False Memory ...