An analysis of fear of crime using multimodal measurement

Biomedical Signal Processing and Control 41 (2018) 186–197

Contents lists available at ScienceDirect

Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc

An analysis of fear of crime using multimodal measurement Seul-Kee Kim, Hang-Bong Kang ∗ Department of Digital Media, Catholic University of Korea, Wonmi-gu, Bucheon-si, Gyeonggi-do, Republic of Korea

a r t i c l e

i n f o

Article history: Received 24 January 2017 Received in revised form 3 November 2017 Accepted 2 December 2017 Keywords: Fear of crime EEG ECG GSR

a b s t r a c t Fear of crime, which may be present without experiencing an actual crime, can restrict one’s daily physical and mental activities and reduce quality of life. In previous research, fear of crime was measured by regional surveys. Though useful for confirming group characteristics, regional surveys cannot measure in real-time, assess individual characteristics, or provide an objective measure of anxiety. Since the causes and effects of fear of crime are highly individualized, we have developed a protocol to measure physiological signals in concert with existing surveys; this system can verify an individual’s fear of crime characteristics in real time. Subjects were shown 6 clips of actual pedestrian environments (day/night of a commercial street scene, day/night of a residential street scene, day/night of a natural street scene). To ease immersion of the subjects into the scenes, clips were produced from the subjects’ first-person point of view. Subjects were divided into two groups (Group1: N = 14, age, 22 ± 1.66 years; Group2: N = 13, age 21 ± 1.35 years) based on their fear intensity as reported on our pre-recording survey; electroencephalographic (EEG), electrocardiographic (ECG), and galvanic skin response (GSR) signals were compared between groups. They were then assessed via video for comparative purposes. Our results demonstrated that the physiological signals were dependent on how conscious an individual was of his or her own fear of crime. We found significant differences between the two groups for all video clips except for daytime commercial street and nighttime natural street; these data suggest that individual characteristics are important in measuring fear of crime. © 2017 Elsevier Ltd. All rights reserved.

1. Introduction With improving quality of life, social demand for security against crime also rises. Despite not personally experiencing crimes, disasters, and accidents, people experience such situations vicariously through the media and their social circles, leading to a rising fear of crime in individuals. To address this, existing research has been used to establish a crime prevention map, based on areas where crime has occurred in the past and other environmental factors [1,2]. However, because this map only deals with data on past crimes, it cannot be used to examine the fear of crime that related to individual characteristics of people in areas with little or no previously recorded crimes. Objectively, the fear of crime and actual crime rate have no direct correlation [3,7,8]. Yet many studies have indicated that the fear of crime affects an individual’s mental health and physical activity. Lorenc et al. [3,4] examined the effects that the fear of crime had on physical and social environments, and determined that there

∗ corresponding author. E-mail addresses: [email protected] (S.-K. Kim), [email protected] (H.-B. Kang). https://doi.org/10.1016/j.bspc.2017.12.003 1746-8094/© 2017 Elsevier Ltd. All rights reserved.

should be more focus on how the fear of crime can potentially mediate the effects of community-level environmental factors related to health and well-being. In addition, Lorenc’s group developed a putative framework for the relationship between fear of crime, environment, and individuals’ mental health and well-being. In this framework, he has argued that both crime and fear of crime have significant effects on well-being. In fact, they found significant effects of fear of crime on the walking behavior of residents. By establishing a crime map that better reflects this information, improvements in both individual and community well-being can be attained [3–6]. Research that measures fear of crime is mostly carried out by surveying a large population of a selected region. This can identify general results for the region, but can only rarely identify individual factors. Moreover, differences in populations and between regions can produce significantly different results. For example, when analyzing the effect of fear of crime on performing a physical activity, Roman et al. [7] found that performance of an AfricanAmerican population living in urban public housing was unaffected by fear of crime. However, many studies on other populations have demonstrated significant effects of fear of crime on the population’s activity [3–6]. Such wide differences imply that fear of crime

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

needs to be analyzed on smaller groups or in individuals as well. In addition, Lim et al. [8] found that responses differed based on how the particular survey for a given population was written, which limits the validity of comparisons. In addition, when measuring fear of crime, many studies have argued that emotional aspects and perceived risk of crime should be distinguished [9–11], but Lim and colleagues noted that the two factors could not be clearly differentiated through a survey. To overcome the survey-related limitations, this study used physiological signals as a more objective measure of emotional state (i.e., fear/anxiety), and to enable identification of individual characteristics. Physiological signals used for measuring emotion include pupillary reflex, electrocardiogram (ECG), electromyogram (EMG), electroencephalogram (EEG), galvanic skin response (GSR), temperature, and heart rate (HR) [12–16,19,40–41]. These measures can continuously analyze real-time emotional state, unlike the surveys, which involve remembering past memories. EEG in humans uses scalp electrodes to quantify regional electrical activity in the cerebral cortex; the most commonly studied EEG frequency bands are delta (␦: 1–4 Hz), theta (␪: 4–8 Hz), alpha (␣: 8–13 Hz), beta (␤: 13–30 Hz), and gamma (ϒ: 30–50 Hz). Current studies propose that frontal EEG power asymmetry represents emotional processing [17,18,33,36,37]. The study of EEG power has been important in brain asymmetry (inter-hemispheric activity differences) and emotion. Broadly speaking, the left regions are specialized for the expression of positive emotion, whereas the right regions are involved in the expression of negative emotions. Davidson et al.’s research has shown that left frontal EEG activity was high for positive emotions, while right frontal EEG activity was high for negative emotions [21,22]. Baumgartner et al. has argued that compared to negative emotions, a positive emotion resulted in the increase of EEG activity in the left hemisphere [20]. The parietal cortex is believed to modulate affective processing and emotionrelated arousal [34–36]. Sarlo et al. suggested that disgust related to alpha power changes in the parietal lobe [37], while Shutter et al. noted a highly significant relationship between right parietal beta EEG activity and attentional responses to an angry face [34]. Balconi et al. has argued that right parietal gamma spectral changes are involved in evaluating emotions as an index of consciousness [38]. Li et al. has shown gamma-band activity to be appropriate for emotion classification [23]. In addition, Nie et al. suggested that left frontal and right temporal lobe gamma activity served as a marker for emotion recognition [24]. ECG measures the electrical activity of the heart in terms of the variation of the cardiac electrical potential over time. The most commonly used cardiac measure is heart rate variability (HRV) [12,25]. It can be used as a factor for examining emotional states [26] and is measured by RR interval variation. R is a peak of the QRS complex, a major signal in the ECG wave. Folino et al. measured the QRS complex during mental stress and found a positive correlation between the energy of the QRS complex and task difficulty [27]. Liu et al. suggested that HRV is a valuable physiological indicator, which could measure the expression of fear memories in mice [39]. GSR measures skin resistance towards electrical conduction between two electrodes, which increases with increasing stimulation (and anxiety/stress) due to increased sweating. GSR in the palm and sole is sensitive to mental stimulation and environmental conditions [28]. This is a particularly useful method for measuring degree of arousal. Liu et al. has proposed GSR as a method for evaluating the emotional intensity of happiness and grief, testing its feasibility in real-life affective computing applications [38]. Vijaya et al. used GSR to categorize emotions in a study utilizing video clips of the emotions of fear, disgust, happiness, and surprise. The 10 subjects showed the highest percentage classification of arousal with fear (80.65%), among the 4 emotions tested [29].

187

Most of the above studies used protocols where the distinction between negative and positive emotions was clear. Fear of crime is a negative emotion, and represents the strongest arousal level among the negative emotions [29]. In this study, we began with a survey to identify subjects as low-intensity or high-intensity fear of crime. We divided them into two groups based on the fear intensity and combined electrophysiological recordings with their perceived fear of crime. We approached fear of crime from an individual perspective to identify those characteristics that would not be found in regional surveys. Our aim was to provide a “safe route” that would ultimately reduce fear of crime in each pedestrian, thereby reducing stress, improving their mental health, and enhancing their quality of life. 2. Materials and methods 2.1. Subjects We enrolled 27 participants (10 men, mean age 22.3 ± 1.70 years; 17 women, mean age 21.05 ± 1.34 years). We deliberately restricted the age to approximately 20 ∼ 25 years, because the focus herein was on the effects of environmental factors on fear of crime, and limiting the age range allowed us to exclude other factors that may have affected the physiological signals. Because this study analyzed physiological signals, some of our previously-used methods from earlier quantitative (survey-based) research were incompatible with this study’s protocols. First, since the time required for the experiment exceeds 2 h, the number of experiments that can be performed per day was limited. Second, the cost of equipment used to measure the physiological signals is significant, especially when compared to studies based on surveys alone. Finally, because of the nature of the signals being analyzed, interference by extraneous signals not related the experimental stimulus needed to be limited, so it was necessary to select the most physically healthy subjects of age 20–25 years. To meet these conditions, many studies will typically include 10–30 subjects; therefore, we concluded that the statistical analysis of 27 subjects in this experiment would provide enough power. When recruiting participants, we accepted all participants meeting the age limit and health requirements, and divided them into two groups (LFG and HFG) according to their responses on the survey of fear of crime. However, any subject for whom the experiment was interrupted due to sleepiness, or for whom the data were corrupted by excessive movement artifacts, was removed from the group prior to data analysis. The study was done with the approval of the IRB of the Catholic University of Korea. All subjects were briefed on the study and provided informed written consent (in accord with the Helsinki Declaration). Fig. 1 depicts the overall study protocols. After completion, participants were paid a small remuneration. 2.2. Survey and behavioral testing All enrolled participants first filled out a survey on fear of crime. After completing the survey, a subject was given a description of the physiological signal measurement apparatus. The subject was then prepared for ECG and GSR measurements by applying the appropriate electrodes. Lastly, EEG electrodes were attached, and the video clips used to assess fear of crime were played. Because accurate measurement of the physiological signals was our top priority, participants could not actually walk without creating movement artifacts, so we used real-time video clips of walking down a street. In order to best approximate the actual conditions, we instructed each subject to think of the clip as a first-person point of view (his/her own point of view), and created an environment in which

188

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

Fig. 1. Overview of the study protocol. “Fear of crime” was measured by fear of crime survey. “Physiological Signals” was contained ECG, GSR and EEG signals.

Fig. 2. Behavioral testing protocol.

they could concentrate on the video. The study procedure is as shown in Fig. 2.

2.3. Video clip format Our aim herein was to measure fear of crime in our daily lives, not fear when given a real crime situation. Thus, our video clips presented an environment that anyone could experience freely. We considered two main factors in clip selection. First, we considered brightness of the environment, dividing clips into Day and Night. Second, we considered location in terms of the surrounding environment, using three places (a commercial street, residential street and natural street) whose characteristics vary according to the surrounding population and vehicle traffic. These clips allowed us to examine individual differences in fear of crime in relatively common environments. For the first 30 s, a subject watched an “intro” clip that contained precautions for viewing the clips. Each test clip was 4 min long, with 30 s of black screen. Clips were presented in the order Commercial Street − Day, Residential Street − Day, and Natural Street − Day, followed by the same sequence of Night clips. The clips were produced with a Canon DSLR 500d by walking through actual streets. The locations were selected from Yukgok-2-

dong, Wonmi-gu, Bucheon, Korea. Samples of each clip are shown in Fig. 3. Before viewing the clips, the subjects were told to view the scene as though the clips’ perspectives were their own. Subjects were warned to minimize movements to prevent electrophysiological noise. During viewing, subjects were isolated and the room darkened. The researchers observed the subjects from a separate room. Videos were displayed on a 27-inch HP monitor with a 1920 × 1080 resolution. The viewing distance was set at 45–50 cm. The study environment is depicted in Fig. 4. 2.4. Survey on fear of crime The survey was based on the fear of crime survey proposed by Kim [30]. This survey examines the subject’s level of fear of crime (10 questions), perceived threat (24 questions), experience with crime (21 questions), perceived vulnerability to crime (11 questions), perceived lack of order in the area of their home (7 questions), and perceived crime prevention in the introduced area (10 questions). It consisted of 87 total questions requiring answers on a 4 point scale. Question examples are as shown in Table 1. All questions can be found in S1 file.

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

189

Fig. 3. Screen shots of scene clips. (a) Commercial Street − Day; (b) Commercial Street − Night; (c)Residential Street − Day; (d) Residential Street − Night; (e) Natural Street − Day; (f) Natural Street − Night.

Table 1 Sample survey on fear of crime. Category

Question

fear of crime

I feel fear when I walk alone at night around my home. I am afraid of meeting a mugger outside. I am afraid of being sexually assaulted by an acquaintance. ... What do you think is the probability of your home being invaded by a robber? During the night, I try to have someone accompany me. I installed security lights (lighting) around my home. . . .. I often see the reports of crimes through newspaper. I have had the experience of burglary while being outside the home. Among those I know, someone has had their home invaded. ... I think that I am more vulnerable to property crime than others are. If I become a victim of crime, I would have a more difficult time recovering mentally/psychologically than others. If someone attacks me I would have a hard time holding that person back . . .. Our neighborhood has a lot of garbage on the streets and is dirty. Our neighborhood has dark and obscure places. I can easily see drunken people at night in our neighborhood. ... Our neighborhood police perform regular patrols. Our neighborhood police have friendly gatherings with members of the community. If I am being assaulted by someone, the neighbor would help me. ...

perceived threat

experience as victim of crime

perceived vulnerability

perceived disorderness

crime control

190

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

Fig. 4. Experimental Environment. Subjects were isolated and the room darkened. The viewing distance was set at 45–50 cm.

2.5. EEG/ECG/GSR recording and preprocessing Because the physiological signals are characteristically sensitive to noise, we provided some instructions before the experiment. First, when participants were watching the clip, they were asked to refrain from yawning, sneezing, and coughing. Also, they were instructed not to move their hands or feet. To also minimize noise, we set impedance to below 5 k before recording, and isolated the subjects in a darkened room to allow them to concentrate on the video clip. We also removed eye movement and facial muscle artifacts. 2.5.1. EEG Using a NeuroScan system, EEG was recorded and analyzed, and power spectra of each frequency range and electrode location were compared. To minimize noise, impedance was set below 5 k before recording. The placement of electrodes was according to the 10–20 system, including 64 channels sampled at a rate of 1 kHz. For analysis, we used Curry 7 software (NeuroScan, http://compumedicsneuroscan.com/). A downsampling to 250 Hz was carried out and a 100 Hz low-pass filter and 60 Hz notch filter applied; the baseline of the data was set as the constant. During the analysis, electrodes that had impedance of 10 k or more between the device and the subject’s scalp were interpolated through kNN (k = 4). Motion artifacts and noise were removed manually, and artifacts due to eye movement were removed for the VEO ch, except those from 0 ∼ 60 ␮V. Data were overlapped using a 1 s window and the average time identified through a bandpass filter of 1 Hz ∼ 50 Hz. In addition, a Fast Fourier transform (FFT) was used to analyze power, and the average frequency interval calculated for the alpha (␣: 8–13 Hz), beta (␤: 13–30 Hz) and gamma (ϒ: 30–50 Hz) bands. EEG power asymmetry was measured for each band for 11 electrode site pairs (F1-F2, F3-F4, F5-F6, F7-F8, FT7FT8, T7-T8, TP7-TP8, P1-P2, P3-P4, P5-P6, P7-P8) (see Fig. 5). We calculated right (R) hemisphere vs. left (L) hemisphere asymmetry indices (R-L) with the formula [(R-L)/(R + L)] [32]. 2.5.2. ECG To measure ECG, we used shimmer3 (shimmer3, www. shimmersensing.com). We attached electrodes to the surface of the body using a standard 12 lead system. For data analysis, MatlabR2013a was used. To identify the R peak of the QRS wave, FFT was carried out and the time domain signal changed to a frequency range. Then a bandpass filter of 10 ∼ 40 Hz was applied, and the result converted back to the time domain via inverse FFT (ifft). Each clip’s R peak was identified, and the value of the RR interval was compared (see Fig. 6). We calculated heart rate (HR) in beat/min, and used the mean value of the RR interval to identify HRV in the time domain.

Fig. 5. Electrodes Used to Measure Asymmetry power. Frontal lobe includes 8 electrode(F1,F2,F3,F4,F5,F6,F7,F8), Temporal lobe includes 6 electrode(FT7,FT8, T7,T8,TP7,TP8) and Parietal lobe includes 8 electrode(P1, P2, P3, P4, P5, P6, P7, P8).

Fig. 6. ECG Waveform and Components.LFG is low score group and HFG is high score group in survey on fear of crime. In fear of crime, perceived crime, Experience as victim of crime and perceived vulnerability, the difference between LFG and HFG is highly significant(**: p value < 0.001).

2.5.3. GSR For measuring GSR, we used NeuLog (NeuLog, www.neulog. com). Electrodes were attached to the left middle and index fingers, and GSR sampled at 10 Hz. For data analysis, MatlabR2013a was used. The obtained data were normalized to a value between 0 and 1 in order to match the different baselines for each subject. The average GSR from each clip was compared to the average GSR for the intro clip to calculate amplitude changes related to viewing a specific clip. 2.6. Assessment of clips A subjective report of the clips was collected after all clips were viewed. The clip assessment included one question (on a 5-point scale) each regarding perspective, threat of crime, familiarity of the area, disorderness, and security (total of 5 questions) [31]. 2.7. Statistical analysis For statistical analysis, we used IBM SPSS Statistics 23. We conducted the Kolmogorov-Smirnov test to identify normality. If the data had normality, we used independent t-tests or ANOVAs. If not,

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

191

3.1. EEG, ECG and GSR features analysis

Fig. 7. Group scores from a survey on fear of crime. LFG is a low-scoring group and HFG is a high-scoring group in a survey on fear of crime. In fear of crime, perceived crime, experience as victim of crime, and perceived vulnerability, the difference between LFG and HFG is statistically significant(** : p value <.001). Table 2 Statistical results of group scores from survey on fear of crime We used a Kdistribution (K = 2) to divide subjects into LFG(low fear group) and HFG(high fear group). **: p < 0.01.

3.2. Statistical analysis

K-distribution statistical results

Fear of crime Perceived Ccime Experience as victim of crime Perceived vulnerability Perceived disorderness Crime control

We analyzed alpha, beta and gamma asymmetry power in the frontal, temporal and parietal lobes. In the LFG, right asymmetry was more active in the black clip than the street clips; however, in the HFG it was less active in the black clip than the street clips. Overall, the HFG appeared to have more right asymmetry power than the LFG in beta and gamma bands (Fig. 8). We analyzed the ECG for HRV and RR interval times. The HFG and LFG had higher values for the black clip than the street clips, whereas the HFG had a higher HR value than the LFG in the street clips. Mean RR intervals in both groups were shorter viewing the black clip than the street clips. Moreover, the HFG had a shorter RR interval than the LFG overall. (Fig. 9). We also compared the variation of GSR between groups, noting that the LFG had higher GSRs while viewing the black clip than while viewing the street clips. Otherwise, the HFG had lower GSRs while viewing the black clip than while viewing the street clips, as well as higher values for night clips than day clips. Changes in GSR amplitude for each clip are given in Fig. 9.

F

p

33.70198 36.9881 15.35595 15.0404 0.094044 0.424528

.000** .000** .001** .001** 0.762 0.521

Mann-Whitney U tests and Kruskal-Wallis tests were used. To compare the differences of fear of crime between groups (LFG and HFG), independent t-tests or Mann-Whitney U tests were used. To compare the difference in fear of crime among the 6 clips, ANOVAs or Kruskal-Wallis tests were used. To analyze variation for each clip statistically, the differences in results for the street clip and a black clip were analyzed just prior to determining the variation of each clip. The statistical significance threshold was set to p < 0.05.

To analyze the difference between groups for each clip, we conducted statistical analyses using the difference between a given street clip and the data just prior to showing the black clip again. After testing for normality, Mann-Whitney U tests were used for group-wise comparisons if the distributions were not normal; for those with normal distributions, independent t-tests were used. The results of the tests for normality are shown in Table 3. In the differences between LFG and HFG groups for each clip, we did not find significant differences in any feature for Commercial Street (Day) or Natural Street (Night). However, we found significant differences in Residential Street (Day) in the ECG (both HR and RR interval). We also found significant differences in Natural Street (Day) in ECG (RR interval) and GSR. In Night clips, Commercial Street demonstrated significant differences in Gamma Frontal EEG and HR in ECG; moreover, the Residential Street clip demonstrated significant results in Beta Frontal and Gamma Frontal EEG. Specific statistical values are shown in Tables 4 and 5.

3. Results 3.3. Clip assessment To analyze environmental factors, we compared location (commercial street − residential street − natural street) and time period (day − night). Based upon the scores from the fear of crime survey, we used a K-distribution (K = 2) to divide them into low (LFG-low fear group) and high (HFG-high fear group) fear of crime groups. We intended to identify the influence of environment as a factor affecting fear of crime on an individual level. Thus, we used the survey of fear of crime suggested by Kim et al. [30]., and divided the LFG and HFG groups through the K-distribution. By comparing the resulting EEG, ECG and GSR data for each group across all the street clips, we confirmed that the environment was a factor that affected both groups. As a result, there were 14 people in LFG (10 men, 4 women, average age of 22 ± 1.66 years), and 13 people in HFG (13 women, average age of 21 ± 1.35 years). Scores for each category are shown in Fig. 7. Through ANOVA statistical analysis, ‘fear of crime’, ‘perceived threat’, and ‘crime victim experience’ categories had significant differences between the two groups. Specific statistical values are shown in Table 2. We confirmed the difference between groups (LFG and HFG) using 12 features of the EEG (9 features; spectral powers for Alpha Frontal, Alpha Temporal, Alpha Parietal, Beta Frontal, Beta Temporal, Beta Parietal, Gamma Frontal, Gamma Temporal, and Gamma Parietal); ECG (2 features; HR and RR interval); and GSR (1 feature; amplitude).

The subjective assessments were high for all clips except the nature − night clip; both groups gave similar responses in all categories. Both groups answered that there was a higher threat of crime present in the night clips than the day clips. In addition, the commercial street clip scored high in familiarity, whereas the natural street clip scored low. The commercial street clip had high disorderness scores, while the natural street clip had low disorderness scores. Furthermore, day clips were considered to have higher security than night clips; the commercial street clips had the highest security scores, and the residential street and natural street − night clips had low stability scores. However, there were no significant differences between groups. More specific values are shown in Fig. 10. 4. Discussion In this paper, we aimed to provide physiological evidence for establishing a new crime prevention map that would consider pedestrians’ fear of crime, thereby reducing stress and improving quality of life in the community. We measured fear of crime by survey, then analyzed electrophysiological signals in individuals to attempt to isolate characteristics that would be missed by largegroup surveys. We found that perceived threat and experiences as

192

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

Fig. 8. Asymmetry in alpha, Beta and Gamma band power represented as time sequence. Blue line is low fear of crime group (LFG) and Red dotted line is high fear of crime group (HFG). (a) Frontal lobe is mean of 4 pairs asymmetry electrode power(F1-F2, F3-F4, F5-F6, F7-F8); (b) Temporal lobe is mean of 3 pairs asymmetry electrode power(FT7-FT8, T7-T8, TP7-TP8); (c) Parietal lobe is mean of 4 pairs asymmetry electrode power(P1-P2, P3-P4, P5-P6, P7-P8);. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 9. The ECG and GSR results represented as time sequence. Blue line is low fear of crime group (LFG) and Red dotted line is high fear of crime group (HFG). (a) HR; (b) average of R–R interval; (c) GSR. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

a victim of crime differentiated individuals between the high and low fear of crime groups. However, like a previous study [5], we could not identify whether such elements directly gave rise to fear of crime. To compare the emotional responses between the high and low fear of crime groups, we analyzed EEG, ECG, and GSR signals in

response to different scenes. According to our results, we found significant differences between LFG and HFG groups in ECG and GSR features in the Day clips. We also identified significant differences between LFG and HFG groups in high-frequency bands (beta and gamma) of Frontal EEG in Night clips. Because people perceived the fear of crime more in the evening than morning, caution related to

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

193

Table 3 Results of the Kolmogorov-Smirnov test for normality. Feature

group

Statistic Alpha Frontal Alpha Temperal Alpha Parietal Beta Frontal Beta Temporal Beta Parietal Gamma Frontal Gamma Temporal Gamma Parietal HR RRinterval GSR

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

Residential Street Day

Commercial Street Day

0.117 0.190 0.127 0.107 0.185 0.266 0.171 0.159 0.206 0.160 0.153 0.159 0.130 0.147 0.090 0.170 0.113 0.162 0.308 0.163 0.265 0.172 0.179 0.407

Sig.

Statistic ∗

.200 .200∗ .200∗ .200∗ .200* 0.012 .200∗ .200∗ .109 .200∗ .200∗ .200∗ .200∗ .200∗ .200∗ .200∗ .200∗ .200∗ 0.001 .200* 0.009 .200* .200* 0.000

0.200 0.132 0.142 0.167 0.218 0.192 0.191 0.173 0.148 0.186 0.177 0.176 0.324 0.321 0.161 0.177 0.231 0.313 0.128 0.184 0.156 0.170 0.260 0.243

Natural Street Day Sig. .136 .200∗ .200∗ .200∗ .071 .200∗ .178 .200∗ .200∗ .200∗ .200∗ .200∗ 0.000 0.001 .200∗ .200∗ 0.041 0.001 .200∗ .200∗ .200∗ .200∗ 0.011 0.035

Statistic 0.178 0.134 0.115 0.223 0.157 0.267 0.130 0.163 0.219 0.269 0.121 0.326 0.135 0.188 0.187 0.243 0.226 0.384 0.130 0.140 0.154 0.111 0.132 0.172

Commercial Street Night Sig.

Statistic ∗

.200 .200∗ .200∗ .075 .200* 0.012 .200∗ .200∗ 0.066 0.011 .200* 0.000 .200∗ .200∗ .200* 0.034 0.051 0.000 .200∗ .200∗ .200∗ .200∗ .200∗ .200∗

0.289 0.150 0.249 0.140 0.189 0.156 0.302 0.205 0.177 0.286 0.354 0.234 0.169 0.250 0.268 0.251 0.275 0.347 0.147 0.182 0.196 0.207 0.297 0.092

Residential Street Night Sig. 0.002 .200* 0.019 .200* .189 .200∗ 0.001 0.139 .200* 0.005 0.000 0.050 .200* 0.026 0.008 0.025 0.005 0.000 .200∗ .200∗ .152 .133 0.002 .200*

Statistic 0.102 0.242 0.180 0.148 0.118 0.319 0.297 0.157 0.189 0.119 0.127 0.177 0.247 0.125 0.230 0.183 0.270 0.208 0.245 0.294 0.267 0.232 0.237 0.145

Natural Street Night Sig. *

.200 0.036 .200∗ .200∗ .200* 0.001 0.002 .200* .189 .200∗ .200∗ .200∗ 0.021 .200* 0.043 .200* 0.007 0.129 0.023 0.003 0.008 0.055 0.032 .200*

Statistic

Sig.

0.242 0.114 0.191 0.105 0.196 0.217 0.150 0.224 0.142 0.278 0.150 0.271 0.120 0.113 0.268 0.202 0.220 0.299 0.193 0.192 0.126 0.121 0.177 0.229

0.026 .200* .176 .200∗ .153 .095 .200∗ .074 .200* 0.007 .200* 0.010 .200∗ .200∗ 0.008 0.150 0.065 0.002 .166 .200∗ .200∗ .200∗ .200∗ .060

Degree of Freedom: LFG = 14; HFG = 13; *:This is a lower bound of the true significance. Significance: P > 0.05: normal distribution, P < 0.05: non-normal distribution.

Fig. 10. Results of clip assessments according to each street clip. Blue line is low fear of crime group (LFG) and Red dotted line is high fear of crime group (HFG). (a) clip’s perspective; (b) threat of crime; (c) familiarity of the area; (d) disorderness; (d) security. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

194

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

Table 4 Statistical results of features using independent t-tests.

Commercial Street Day

Features

Group

mean

Std. Dev.

Std. Err

t

Alpha Frontal

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

−0.027 −0.039 0.003 0.001 −0.013 −0.015 0.0008 −0.009 0.0002 0.013 −0.008 −0.004 −0.005 −0.003 0.001 0.0009

0.090 0.075 0.076 0.051 0.028 0.033 0.042 0.042 0.014 0.027 0.023 0.011 0.032 0.017 0.005 0.011

0.024 0.020 0.020 0.014 0.007 0.009 0.011 0.011 0.003 0.007 0.006 0.003 0.008 0.004 0.001 0.003

0.387

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

0.022 0.038 −0.019 0.015 −0.024 0.035 −0.024 −0.016 −0.039 −0.028 −0.014 −0.001 −0.021 −0.016 −3.803 0.076 31.741 −4.701

0.136 0.085 0.057 0.055 0.057 0.152 0.069 0.031 0.074 0.042 0.053 0.038 0.043 0.026 5.369 3.359 35.390 30.850

0.036 0.023 0.015 0.015 0.015 0.042 0.018 0.008 0.019 0.011 0.014 0.010 0.011 0.007 1.434 0.931 9.458 8.556

−0.377

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

0.013 −0.031 −0.024 0.019 0.013 −0.0003 0.008 −0.001 −5.214 −1.711 52.391 14.204 −0.034 0.087

0.115 0.095 0.077 0.121 0.051 0.024 0.022 0.008 6.096 3.413 60.681 28.358 0.124 0.053

0.030 0.026 0.020 0.033 0.013 0.006 0.006 0.002 1.629 0.946 16.217 7.865 0.033 0.014

1.095

LFG HFG LFG HFG LFG HFG

0.032 0.031 −5.857 −1.326 52.716 16.815

0.083 0.153 5.8026 4.838 59.382 47.891

0.022 0.042 1.550 1.341 15.870 13.282

0.023

LFG HFG LFG HFG LFG HFG

−0.021 0.012 0.014 −0.014 −0.003 0.008

0.082 0.065 0.059 0.037 0.022 0.023

0.021 0.018 0.015 0.010 0.006 0.006

−1.163

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

0.035 −0.0005 −0.047 0.023 −0.009 0.001 −0.006 −0.004 −4.392 0.692 44.658 10.722 0.040 0.042

0.125 0.086 0.120 0.084 0.040 0.018 0.035 0.014 5.164 9.105 56.978 50.063 0.123 0.050

0.033 0.024 0.032 0.023 0.010 0.005 0.009 0.003 1.380 2.525 15.228 13.885 0.033 0.013

0.852

Alpha Temp Beta Frontal Beta Temoral Beta Parietal Gamma Frontal Gamma Temporal Gamma Parietal Home Street Day

Alpha Frontal Alpha Temporal Alpha Parietal Beta Frontal Beta Temporal Beta Parietal Gamma Temporal HR RRinterval

Nature Street Day

Alpha Frontal Alpha Temporal Beta Frontal Gamma Frontal HR RRinterval GSR

Commercial Street Night

Alpha Parietal HR RRinterval

Home Street Night

Alpha Temporal Beta Temporal Beta Parietal

Nature Street Night

Alpha Temporal Alpha Parietal Beta Frontal Gamma Frontal HR RRinterval GSR

Statistical significance in differences in EEG, ECG and GSR; data were normally distributed. *: p < 0.05; **: p < 0.01 Std. Dev: standard deviation; Std. Err: standard error.

0.082 0.116 0.634 −1.529 −0.604 −0.173 0.044

−1.625 −1.370 −0.379 −0.455 −0.722 −0.385 −2.230* 2.842**

−1.121 0.882 1.472 −1.822 2.067* −3.262**

−2.194* 1.721

1.487 −1.376

−1.764 −0.918 −0.258 −1.802 1.639 −0.049

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

195

Table 5 Statistical results in features using Mann-Whitney U tests for non-normally distributed data.

Commercial Street Day

Features

Group

N

Mean

Std. Dev.

Mean Rank

Mann-Whitney’s U

Z

p

Alpha Parietal

LFG HFG LFG HFG LFG HFG LFG HFG

27

0.022

0.060

67.5

−1.141

0.254

27

−1.120

4.753

88.5

−0.122

0.903

27

9.514

47.609

78

−0.631

0.528

27

0.033

0.217

12.32 15.81 14.18 13.81 13.07 15 15.71 12.15

67

−1.165

0.244

LFG HFG LFG HFG LFG HFG

27

−0.007

0.043

89.000

−0.097

0.923

27

−0.006

0.024

86.500

−0.218

0.827

27

0.003

0.152

13.86 14.15 14.32 13.65 11.71 16.46

59.000

−1.553

0.120

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

27

−0.0009

0.109

63.000

−1.359

0.174

27

−0.003

0.072

80.000

−0.534

0.593

27

−0.001

0.072

60.500

−1.480

0.139

27

−0.002

0.034

69.500

−1.043

0.297

27

−0.001

0.033

12.00 16.15 13.21 14.85 11.82 16.35 12.46 15.65 12.75 15.35

73.500

−0.850

0.396

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

27

−0.016

0.131

85.000

−0.291

0.771

27

−0.028

0.119

80.000

−0.534

0.593

27

0.006

0.047

52.500

−1.869

0.062

27

−0.006

0.107

89.000

−0.097

0.923

27

0.011

0.103

86.500

−0.218

0.827

27

0.009

0.038

48.500

−2.063

0.039

27

0.014

0.053

73.500

−0.849

0.396

27

−0.001

0.026

85.000

−0.291

0.771

27

0.093

0.141

14.43 13.54 13.21 14.85 16.75 11.04 14.14 13.85 14.32 13.65 17.04 10.73 15.25 12.65 13.57 14.46 13.86 14.15

89.000

−0.097

0.923

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

27

−0.015

0.055

56.000

−1.699

0.089

27

0.008

0.090

78.000

−0.631

0.528

27

0.004

0.055

45.000

−2.232

0.026

27

0.008

0.043

47.500

−2.111

0.035

27

0.001

0.034

67.000

−1.165

0.244

27

0.001

0.012

86.000

−0.243

0.808

27

−4.796

6.284

60.000

−1.505

0.132

27

49.541

59.355

59.000

−1.553

0.120

27

−0.020

0.150

11.50 16.69 14.93 13.00 17.29 10.46 17.11 10.65 15.71 12.15 13.64 14.38 11.79 16.38 16.29 11.54 9.50 18.85

28.000

−3.057

0.002

LFG HFG LFG HFG LFG HFG LFG HFG LFG HFG

27

0.016

0.098

75.000

−0.776

0.438

27

0.011

0.065

90.000

−0.049

0.961

27

−0.004

0.044

83.000

−0.388

0.698

27

0.008

0.040

78.000

−0.631

0.528

27

0.001

0.018

12.86 15.23 13.93 14.08 14.57 13.38 13.07 15.00 15.00 12.92

77.000

−0.680

0.497

HR RRinterval GSR Home Street Day

Gamma Frontal Gamma Parietal GSR

Nature Street Day

Alpha Parietal Beta Temporal Beta Parietal Gamma Temporal Gamma Parietal

Commercial Street Night

Alpha Frontal Alpha Temporal Beta Frontal Beta Temporal Beta Parietal Gamma Frontal Gamma Temporal Gamma Parietal GSR

Home Street Night

Alpha Frontal Alpha Parietal Beta Frontal Gamma Frontal Gamma Temporal Gamma Parietal HR RRinterval GSR

Nature Street Night

Alpha Frontal Beta Temporal Beta Parietal Gamma Temporal Gamma Parietal

*: p < 0.05; **: p < 0.01

196

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197

such perception likely generated the EEG response in the Night clip. In the Day clip, which is a relatively safe condition, the ECG and GSR were likely caused by general nervousness. Thus, we conclude that the significant results allow us to group levels of fear of crime in each environment by using EEG, ECG, and GSR features. First, we measured the EEG asymmetry power of alpha, beta, and gamma frequency bands in the frontal, temporal, and parietal lobes. EEG data will generally include other cognitive and mental elements that could obscure emotional processing. We measured frontal gamma band activity, which Davidson [21,22] and Baumgartner [20] have shown to be preferentially right-sided (in beta band as well) in response to negative emotions. Like Balconi et al. [38] and Li et al. [24], we could evaluate emotional states of high arousal in right parietal, temporal and frontal regions using gamma bands. We found stronger positive values in the HFG than LFG. We also found the asymmetry power of the LFG was stronger for street clips than the black clip in beta and gamma bands for all three described regions. Nevertheless, the asymmetry power of the HFG was stronger than that of the LFG. In particular, the asymmetry power for the street clips was smaller than that for the black clip, but only for the LFG. The results suggest that the LFG felt stronger fear relating to the experimental environment during the black clips (isolated, dark, and unfamiliar) than during the street clips; however, this effect was lost for the HFG. Like EEG, ECG has similar issues of signal contamination, since general emotional arousal and valence elements greatly influence HRV. We found that the HFG had higher HR and a shorter mean RR interval for street clips than the LFG. The higher HR/shorter RR intervals suggest the HFG is more fearful than the LFG. As we found for the EEG data, ECG data indicated that participants in both groups experienced greater fear viewing the black clips than the street clips. GSR is a validated measure of general emotional arousal. As seen in previous work [28], there was a stronger change in amplitude with more strongly felt stimulation. Like the other results, the amplitude of GSR was higher (i.e., greater fear during viewing) for black clips than for street clips in the LFG. On the other hand, the amplitude of GSR gradually increased in the HFG, suggesting that the level of fear changes with time, regardless of the physical place. We compared the groups using differences in EEG, ECG and GSR signals between street clips and black clips. All clips except commercial street − day and natural street − night were found to be significantly different between groups in all physiological measures; this difference was not found in the results from the clip assessment survey. The survey did reveal that the commercial street − day was considered the most stable and familiar, suggesting that viewers of this clip felt no fear about crime occurring there. On the other hand, the natural street − night scene scored lowest for both stability and familiarity, so both groups were relatively fearful while viewing the clip. Since the physiological signals were not significantly different in this respect, we believe that the relative lack of reality of the simulation could not stimulate fear of crime sufficiently. The limitations of our study include the need to have subjects watching video clips rather than actually walking, which may not model the situation very well. Second, we could only divide the degree of fear of crime into two general groups, but with more subjects, it would be possible to further divide the degree of fear of crime in an individual by gender/age group. Finally, the environmental factors were divided into time (day/night), and population and vehicle movement (commercial/residential/natural). In future studies it will be necessary to combine the various elements (CCTV, streetlight, big building, graffiti, etc.) that can modulate fear of crime.

5. Conclusion Much research on fear of crime was conducted by survey according to demographic characteristics as quantitative research. However, fear of crime is affected by previous experience, recognition of crime, usual walking habits and various other individual factors. Thus, we focused on the individual factors to identify fear of crime in a novel way. In our study, the baseline level of fear of crime was important, so we used 6 categories to measure the fear of crime, and then 4 categories (fear of crime, perceived crime, experience as victim of crime and perceived vulnerability) to classify the fear of crime group. Using this system, we found features of the physiological signals measured that could divide the level of fear of crime more objectively. Because of the need for approaching fear of crime on a more individual basis, we attempted to confirm the influence of environmental and individual factors on fear of crime, specifically in relation to the degree of fear of crime. If the real-time changes in physiological signals represent the influence of environmental factors on an individual’s fear of crime as revealed by the survey of fear of crime, recognizing these factors will aid the development of safe walking routes for individuals that reduce such fear, thus improving our daily quality of life. In our study, we used the physiological signals to confirm the influence of environment as a factor during walking according to individual fear of crime levels in real-time. We propose that not only was fear of crime influenced by environment according to an individual’s level of fear of crime, but also that the physiological signals we analyzed represent a significant method to measure the fear of crime objectively, and thus overcome the limitations of previous studies. We suggest a new direction of study, using physiological signals and a survey of previous studies to identify the characteristics of fear of crime on an individual level. Future studies will use wearable devices in order to study real pedestrian environments instead of simulations. Furthermore, the study will be carried out with the CPTED (Crime Prevention Through Environmental Design) element as well as individual elements rather than environmental elements.

Acknowledgments This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and future Planning (No. 2015R1A2A1A10056304).

References [1] S. Chainey, L. Tompson, S. Uhlig, The utility of hotspot mapping for predicting spatial patterns of crime, Secur. J. 21 (1-2) (2008) 4–28. [2] Y. Xue, D.E. Brown, Spatial analysis with preference specification of latent decision makers for criminal event prediction, Decis. Support Syst. 41 (3) (2006) 560–573. [3] T. Lorenc, M. Petticrew, M. Whitehead, D. Neary, S. Clayton, K. Wright, et al., Fear of crime and the environment: systematic review of UK qualitative evidence, BMC Public Health 13 (1) (2013) 1. [4] T. Lorenc, S. Clayton, D. Neary, M. Whitehead, M. Petticrew, H. Thomson, et al., Crime, fear of crime, environment, and mental health and wellbeing: mapping review of theories and causal pathways, Health Place 18 (4) (2012) 757–765. [5] J. Burke, P. O’Campo, C. Salmon, R. Walker, Pathways connecting neighborhood influences and mental well-being: socioeconomic position and gender differences, Soc. Sci. Med. 68 (7) (2009) 1294–1304. [6] P. O’Campo, C. Salmon, J. Burke, Neighbourhoods and mental well-being: what are the pathways? Health Place 15 (1) (2009) 56–68. [7] C.G. Roman, C.R. Knight, A. Chalfin, S.J. Popkin, The relation of the perceived environment to fear, physical activity, and health in public housing developments: evidence from Chicago, J. Public Health Policy 30 (1) (2009) S286–S308. [8] H. Lim, Y. Chun, The limitations and advancements in measuring fear of crime, J. Public Admin. Govern. 5 (2) (2015) 140–148.

S.-K. Kim, H.-B. Kang / Biomedical Signal Processing and Control 41 (2018) 186–197 [9] E.K. Asencio, M. Merrill, M. Steiner, Self[U + 2010]Esteem, the Fear of Crime, and the Decision to Protect Oneself From Victimization, Wiley Online Library Sociological Forum, 2014. [10] K.F. Ferraro, Fear of Crime: Interpreting Victimization Risk, SUNY press, 1995. [11] K.F. Ferraro, R.L. Grange, The measurement of fear of crime, Sociol. Inquiry 57 (1) (1987) 70–97. [12] F. Agrafioti, D. Hatzinakos, A.K. Anderson, ECG pattern analysis for emotion detection, IEEE Trans. Affective Comput. 3 (1) (2012) 102–115. [13] R.W. Picard, Affective computing: challenges, Int. J. Hum. Comput. Stud. 59 (1) (2003) 55–64. [14] P. Ekman, R.W. Levenson, W.V. Friesen, Autonomic nervous system activity distinguishes among emotions, Science 221 (September (4616)) (1983) 1208–1210. [15] F. Honig, A. Batliner, E. Noth, Real-time recognition of the affective user state with physiological signals, Proceedings of the Doctoral Consortium, Affective Computing and Intelligent Interaction (2007). [16] C.M. Jones, T. Troen, Biometric Valence and Arousal Recognition. Proceedings of the 19th Australasian Conference on Computer-Human Interaction: Entertaining User Interfaces, ACM, 2007. [17] J.A. Coan, Allen J.J. Frontal, EEG asymmetry as a moderator and mediator of emotion, Biol. Psychol. 67 (1) (2004) 7–50. [18] P.C. Petrantonakis, L.J. Hadjileontiadis, A novel emotion elicitation index using frontal brain asymmetry for enhanced EEG-based emotion recognition, IEEE Trans. Inf. Technol. Biomed. 15 (5) (2011) 737–746. [19] X. Wang, D. Nie, B. Lu, Emotional state classification from EEG data using machine learning approach, Neurocomputing 129 (2014) 94–106. [20] T. Baumgartner, M. Esslen, L. Jancke, From emotion perception to emotion experience: emotions evoked by pictures and classical music, Int. J. Psychophysiol. 60 (1) (2006) 34–43. [21] R.J. Davidson, N.A. Fox, Asymmetrical brain activity discriminates between positive and negative affective stimuli in human infants, Science 218 (4578) (1982 Dec 17) 1235–1237. [22] R.J. Davidson, Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums, Psychophysiology 35 (05) (1998) 607–614. [23] M. Li, B. Lu, Emotion classification based on gamma-band EEG, in: Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE 2009, 2009. [24] D. Nie, X. Wang, L. Shi, B. Lu, EEG-based emotion recognition during watching movies. Neural Engineering (NER), in: 5th International IEEE/EMBS Conference on, IEEE; 2011, 2011. [25] H. Ue, I. Masuda, Y. Yoshitake, T. Inazumi, T. Moritani, Assessment of cardiac autonomic nervous activities by means of ECG R[U + 2010]R interval power spectral analysis and cardiac depolarization[U + 2010]repolarization process, Ann. Noninvasive Electrocardiol. 5 (4) (2000) 336–345.

197

[26] S. Jerritta, M. Murugappan, K. Wan, S. Yaacob, Emotion detection from QRS complex of ECG signals using Hurst Exponent for different age groups. Affective Computing and Intelligent Interaction (ACII), in: Humaine Association Conference on, IEEE; 2013, 2013. [27] A.F. Folino, G. Buja, P. Turrini, L. Oselladore, A. Nava, The effects of sympathetic stimulation induced by mental stress on signal-averaged electrocardiogram, Int. J. Cardiol. 48 (3) (1995) 279–285. [28] F. Liu, G. Liu, X. Lai, Emotional intensity evaluation method based on galvanic skin response signal. computational intelligence and design (ISCID), in: Seventh International Symposium on, IEEE; 2014, 2014. [29] P. Vijaya, G. Shivakumar, Galvanic skin response: a physiological sensor system for affective computing, Int. J. Machine Learn. Comput. 3 (1) (2013) 31. [30] Y. Kim, A Study on the Causes of and Reaction to Fear of Crime, Graduate School of Dongguk University of Korea, 2010. [31] P. Salesses, K. Schechtner, C.A. Hidalgo, The collaborative image of the city: mapping the inequality of urban perception, PLoS One 8 (7) (2013) e68400. [32] R. Yuvaraj, M. Murugappan, N.M. Ibrahim, M.I. Omar, K. Sundaraj, K. Mohamad, et al., On the analysis of EEG power, frequency and asymmetry in Parkinson’s disease during emotion processing, Behav. Brain Funct. 10 (1) (2014) 1. [33] E. Harmon Jones, Clarifying the emotive functions of asymmetrical frontal cortical activity, Psychophysiology 40 (6) (2003) 838–848. [34] D.J. Schutter, P. Putman, E. Hermans, J. van Honk, Parietal electroencephalogram beta asymmetry and selective attention to angry facial expressions in healthy human subjects, Neurosci. Lett. 314 (1) (2001) 13–16. [35] B.D. Smith, M. Meyers, R. Kline, A. Bozman, Hemispheric asymmetry and emotion: lateralized parietal processing of affect and cognition, Biol. Psychol. 25 (3) (1987) 247–260. [36] W. Heller, Neuropsychological mechanisms of individual differences in emotion, personality, and arousal, Neuropsychology 7 (4) (1993) 476. [37] M. Sarlo, G. Buodo, S. Poli, D. Palomba, Changes in EEG alpha power to different disgust elicitors: the specificity of mutilations, Neurosci. Lett. 382 (3) (2005) 291–296. [38] M. Balconi, C. Lucchiari, Consciousness and arousal effects on emotional face processing as revealed by brain oscillations. A gamma band analysis, Int. J. Psychophysiol. 67 (1) (2008) 41–46. [39] J. Liu, W. Wei, H. Kuang, F. Zhao, J.Z. Tsien, Changes in heart rate variability are associated with expression of short-term and long-term contextual and cued fear memories, PLoS One 8 (5) (2013) e63590. [40] W. Wu, J. Lee, Improvement of HRV methodology for positive/negative emotion assessment. collaborative computing: networking, applications and worksharing, in: CollaborateCom 2009. 5th International Conference on, IEEE; 2009, 2009. [41] N.D. Ahuja, A.K. Agarwal, N.M. Mahajan, N.H. Mehta, H.N. Kapadia, GSR and HRV: its application in clinical diagnosis. Computer-Based Medical Systems, in: Proceedings. 16th IEEE Symposium, IEEE; 2003, 2003.