Evaluation of detection and discrimination ability of peripheral vision on notification information based on large displays

Displays 41 (2016) 50–60

Contents lists available at ScienceDirect

Displays journal homepage: www.elsevier.com/locate/displa

Evaluation of detection and discrimination ability of peripheral vision on notification information based on large displays Kuo Hao Tang ⇑, Yueh Hua Lee Feng Chia University, 100 Wenhua Road, Taichung, Taiwan, ROC

a r t i c l e

i n f o

Article history: Received 17 February 2015 Received in revised form 16 November 2015 Accepted 3 December 2015 Available online 8 December 2015 Keywords: Notification information Detection Discrimination Large display Peripheral vision

a b s t r a c t Large displays enable users to perform several tasks simultaneously. Under such circumstances, notification information provided through the concept of ambient displays plays a vital role in assisting users to switch among tasks. This paper presents the experimental results of a notification system design in the peripheral region of large displays. The aim is to provide guidance for notification information design by investigating detection and discrimination performance of human observers when visual notification information is presented away from the foveal region and viewed using peripheral vision. The proposed notification system was designed using an array of glyphs. Each glyph is a small gray square with a fixed size of 60
60 pixels. By changing the gray levels of adjacent glyphs dynamically, a glyph array presents a particular dynamic pattern. The experiments involved testing factors that comprised the visual angle, size and shape of glyph arrays, frequency of temporal modulation, phase shift of each pattern, and number of stimuli. The results show that glyph arrays are detected accurately if they are larger, even at wide viewing angles, and that the number of glyphs in a glyph array affects the performance more than the shapes of glyph arrays do. Furthermore, the discrimination performance is higher when both the frequency and phase are manipulated simultaneously (multidimensional design), compared with the case when each of these dimensions is varied separately (single-dimensional design). When the number of stimuli is set at 8, for example, users can maintain an accuracy rate of 70% for the multidimensional design, whereas the accuracy rate is only approximately 60% for the single-dimensional design. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction Large high-resolution displays are leading the growth in the global display market. NPD Display Search indicated that in January 2005, the average desktop monitor size for personal workstations was 16.400 and by 2013, the corresponding average measurement was 20.900 . For the professional graphics market, the largest market share by size in 2013 was 2700 displays, and curved displays with a size of up to 4000 are currently available. A larger display provides more space to present information, thereby supporting various tasks for users simultaneously and in a more detailed view than on a smartphone or tablet [1]. In addition to the trend of increased display size, many people currently enjoy maintaining awareness of information such as news, the weather, entertainment, and other personally relevant information when interacting with a computer [2,3] or a smart device [4,5]. Such information is typically provided by a notification system that transmits current and timely information effi⇑ Corresponding author. E-mail address: [email protected] (K.H. Tang). http://dx.doi.org/10.1016/j.displa.2015.12.002 0141-9382/Ó 2015 Elsevier B.V. All rights reserved.

ciently and effectively without causing unwanted distraction to a user’s ongoing tasks [6]. A notification system can be used for several purposes including (1) receiving news [7], (2) interacting with social groups [8,9], and (3) delivering information through notifications such as time-sensitive data [10]. According to the priority of the information being conveyed, the display of notification systems can be divided into two categories: ambient and alert. An ambient display shows low-priority information and requires divided human attention, whereas an alert display shows prioritized information demanding focused attention [10]. Notification information can be transmitted differently through human modalities such as visual, auditory, tactile, olfactory, and multimodal [11–14]. Arroyo et al. [15] compared five notification modalities—heat, smell, sound, vibration, and light—from the aspect of disruption. The results indicated no considerable differences among the five transmission methods regarding disruption. Warnock et al. [14] compared eight delivery methods categorized into four groups: visual, auditory, tactile, and olfactory. The visual group comprised text, pictograms, and abstract visual stimuli; the auditory group comprised voice, earcons, and auditory icons; the tactile group comprised tactons; and the olfactory group

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

comprised aromacons regarding disruptiveness and effectiveness. The results indicated that the users demonstrated considerably higher accurate responses to the notification via visual and audio transmissions than they did via tactile and olfactory transmissions. Regarding the response time, the users had the shortest response time to the visual cues and the longest response time to the olfactory cue. From the aspect of disruption, tactile and olfactory transmissions introduced more disruption to the users than visual and auditory cues did. These studies have suggested that, compared with other human modalities, visual and auditory cues provide optimal transmission to users regarding disruption, response time, and accuracy. However, auditory cues are designed to prompt immediate action, whereas visual cues are not designed for vigilant types of tasks [16]. This suggests that auditory cues are more suitable for alert displays and visual cues are more suitable for ambient displays. 1.1. Visual cue Because this study focused on nonemergent information types, the information transmission was designed on the basis of visual cues. Different forms of visual cues, such as text, patterns, pictograms, shapes, and colors, can be used for transmitting information. The transmission method can be either static or dynamic and the presentation of the transmitted information can be abstract or concrete. Numerous peripheral awareness systems have been created to support abstract presentation [17–20]. Tarasewich et al. [11] combined color and position on three LEDs and conveyed 27 messages with high recognition accuracy for users. One of the peripheral awareness systems, ambient media, comprised physical devices, such as money color [21], breakaway [22], and daylight displays [23], that were placed in a person’s environment. Hung and Ostovari [24] designed an assistive interface in which hints (e.g., changing the color of a cursor) are provided to attract a user’s attention to a notification that is initially displayed in the peripheral region outside the user’s field of view. An example of concrete presentation involves text information. Plaue and Stasko [2] compared different peripheral display configurations for text information. McCrickard et al. [25] compared three animation notification systems (i.e., blast, fade, and ticker) with no animation regarding the correct rate, hit rate, and false alarm rate. The results showed that the blast and fade animations resulted in considerably faster monitoring times than the ticker did. The hit rate for the ticker was higher than that for the fade and blast. 1.2. Design for human peripheral vision When interacting with large displays, users generally separate the focal region from the peripheral region depending on the priorities of tasks, and they can take advantage of peripheral vision to monitor applications of lower relevance by placing them in the peripheral areas of the display [26]. Therefore, demand is increasing for using peripheral displays in maintaining awareness [3,7,27], in which users tend to glance at or use peripheral vision to view low-priority information. Anderson et al. [28] measured spatial contrast sensitivity functions at retinal locations from 0° to 55° along the nasotemporal meridian for a single eye and found that contrast sensitivity functions for peripheral vision are shaped similarly to those observed foveally, but are shifted to lower spatial frequencies. In particular, there is a clear nasotemporal asymmetry in contrast sensitivity in the far peripheral visual field. Stimuli imaged on the nasal retina are detected with higher sensitivity than those imaged on the temporal retina.

51

Legge et al. [29] linked the spatial and temporal properties of letter recognition to reading speed for text viewed using central or peripheral vision. They found that the size of the visual span decreased from at least 10 letters in central vision to 1.7 letters at 15° eccentricity, concluding that the retinal position, exposure time, and relative position within a character string are key factors that limit letter-recognition accuracy. Chung et al. [30] compared the effects of central and peripheral vision on the spatialfrequency characteristics of letter identification, determining that the spatial frequency tuning and scaling properties for letter identification were similar between the fovea and periphery. In addition to letter-like stimuli, human peripheral vision in texture segregation and contour integration also has crucial implications in pattern and object recognition. Joffe and Scialfa [31] investigated texture segmentation as a function of eccentricity and concluded that optimal texture segregation does not peak in foveal vision but does so in the near periphery. Experimental evidence suggests that contour integration is mainly present in foveal vision [32,33]. However, recently Kuai and Yu [34] demonstrated that for contour stimuli such as circles and ellipses, which bear favorable Gestalt properties, contour integration for shape detection and discrimination was nearly constant from the fovea to up to 35° of visual periphery. The cones and rods in the human retina provide different ocular capabilities. The cones are efficient for visual acuity, visual resolution, and color recognition, and the rods are effective for motion detection. The cell density is a function of the retinal angle; where away from fovea, the retina is composed primarily of rod receptors with extremely few cones [35]. Although this may imply that human peripheral vision is sensitive to motion and is relatively ineffective for color discrimination, it is now well known that peripheral color vision is similar to foveal vision if the target is sufficiently large. Gordon and Abramov [36] measured the spectral hue and saturation functions of the nasal retina both at and 45° from the fovea. Using large and small targets in the fovea (1.5° and 50 ) and periphery (6.5° and 1.5°), they found that the quality of color vision in the periphery depends crucially on stimulus size. A sufficiently large stimulus enables detecting a complete range of well-saturated hues. 1.2.1. Dynamic design for peripheral vision Notification information is widely presented using animation [2,25] because dynamic presentations obviously attract more attention than static presentations do; the efficiency of such presentations is generally evaluated according to glanceability [37] instead of studying peripheral vision directly. Research on the perception and recognition ability of peripheral vision has largely focused on static information (e.g., a fixed color or word); only a few studies have focused on the effects of dynamic information. Bartram et al. [38] reported that motion cues draw more attention than do static representations, and some motion types (e.g., traveling motions) are more distracting and irritating than other types (e.g., anchored motions). They suggested that traveling motion requires more attention because in addition to detection, a cognitive act of tracking is involved. Park and Nam [39] used card sorting skills to extract four dynamic design elements: tempo, direction, rhythm, and volume. They suggested that in the case of presenting information, complex information tends to require more design elements or coding dimensions compared with simple information. Yamada et al. [40] developed an information notification method called peripheral cognition technology. They applied the phenomenon of visual field narrowing (VFN), in which the human visual field narrows considerably during a difficult task, to design a peripheral agent. When a new message arrives, this agent appears in a peripheral visual area outside the visual scope of the primary task. Because of VFN, users may not notice the onset of this agent

52

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

when they are focused on the main task. The information notification is perceived only when the attention to the main task is lowered somewhat and the phenomenon of VFN disappears, thus enabling the attention to be switched to other regions of the display.

(255, 255, 255) (128, 128, 128) 0s

1s (0, 0, 0)

1.3. Objective The objective of this study was to design a novel notification system for presenting nonemergent information by using the peripheral areas of a large display while minimizing the distraction and mental workload on users. In other words, by applying the concept of calm technology, users can perceive information presented on a certain part of a display by using peripheral attention to focus on it as desired [41]. To prevent unexpected interruptions, current notification systems generally present notification information along the border of a display. Users generally glance briefly at the information or capture it by using their peripheral vision. However, when the display size increases, more space is potentially available to present information. This raises several questions. According to the concept of calm technology, can human eyes capture notification information by using peripheral vision when the information is presented away from the foveal region? If so, then what types of peripheral display designs can facilitate human detection and discrimination? This study performed two experiments to answer these questions. In the first experiment, detectability was measured to verify the design of the proposed notification system and to determine the parameters for the design. In the second experiment, discriminability was measured to determine the performance of the proposed design.

2. Design of notification system using a glyph array The proposed notification system was designed using an array of glyphs. Each glyph is a small gray square with a fixed size of 60
60 pixels. The behaviors of the notification system were controlled by applying four parameters. 2.1. Shape and size The size of the notification system is determined according to the number of columns and rows of the array of glyphs. Four sizes in width (1, 2, 4, and 8 columns) and four sizes in length (2, 4, 8, and 16 rows) produce 16 (4
4) shapes in the proposed notification system. 2.2. Rhythm and sine wave In the proposed design, the luminance of each glyph changes dynamically with time to attract attention. However, to prevent abrupt interruptions during the onset of information notification, the dynamic change in the luminance of a glyph is controlled by a sine wave function. A sine wave function along time is mapped to an RGB value from (0, 0, 0) to (255, 255, 255) to create a smooth gray-level transition for each glyph (Fig. 1).

Fig. 1. Gray level of a glyph shown in RGB values changes with time and is controlled with a sine wave function at a frequency of 1 Hz.

2.4. Sequence and phase shift Although the luminance of each glyph changes according to the rhythm and tempo, when there is no difference between adjacent glyphs, the notification system displays a large gray rectangle with changing shades because all glyphs follow the same rhythm and tempo in the proposed design. To capture a user’s attention through only his or her peripheral vision, the dynamic interactions between adjacent glyphs should create an overall pattern that can elicit awareness. The phase-shift parameter is introduced to provide this function. The phase shift in our case is defined as the difference (expressed in degrees) between two adjacent glyphs following the same rhythm and tempo and referenced to the same time point. Given xij ðtÞ, which represents the corresponding sinusoidal function for the glyph at the ith row and jth column of the notification system at time t, a single parameter (£) can be used to control all vertically and horizontally adjacent glyphs, as shown in Eq. (1), and to generate an overall dynamic pattern.

 xij ðtÞ ¼ sin 2pft þ ði þ j  2Þ£

p 

ð1Þ

180

where i is the length (number of rows), j represents the width (number of columns), f is the tempo (frequency), and £ is the phase shift. Fig. 2 depicts a notification system of size 2 glyphs
2 glyphs with £ ¼ 36 of change along the time axis. 3. Experiment I: detectability of glyph arrays This experiment was conducted to determine whether peripheral vision can be used to detect dynamic patterns generated by glyph arrays presented away from the central focal area of a display. The just-noticeable difference (JND) of the phase shift (£) was measured as a dependent variable to evaluate the design. 3.1. Participants and settings This study recruited 10 paid participants, five females and five males, from the College of Engineering at Feng Chia University. Informed consent was obtained from all the participants prior to participation. All collected data during the experiments were depersonalized before statistical analyses were performed. The participants had at least 3 years of experience using computers. The average age of the participants was 20 years, with a standard deviation of 1.1 years. All the participants had normal or corrected-to-normal eyesight.

2.3. Tempo/Frequency The tempo of the changing gray level is determined by the frequency (f) of the sine wave function. Fig. 1 illustrates an example of a sine wave function at a frequency of 1 Hz.

Time 

Fig. 2. Example of a notification system of size 2
2 with £ ¼ 36 .

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

The viewing angles for presenting the notification information (i.e., glyph array) were tested at two levels: 25° and 45° shifts from the center of the display along the nasotemporal meridian. In Bi and Balakrishnan [42], participants interacted with a tiled display at distances of 2–2.5 m; the width and height of the tiled display were 4.9 and 1.8 m, respectively. They concluded that 81% of mouse events occurred within a central square area ranging from 1.4 to 3.5 m wide and less than 1.2 m high. They determined that in this central square area, primary tasks were performed within a width of 2.1 m (3.5–1.4 m = 2.1 m) and secondary tasks were performed outside this area. When this area is used as a baseline, the viewing angle from 2.25 m (the average distance from a participant to the display) in front of the center of the tiled display to the border between the primary and secondary areas can be calculated at approximately 25° (tan1(2.1/2)/2.25 = 24.99°), and the viewing angle from the same location to the border of the tiled display is nearly 47° (tan1(4.9/2)/2.25 = 47.44°). Thus, in the current study, these viewing angles were used as a reference, and the presentation of the notification information was tested at both 25° and 45°. Three 2300 displays with 1920
1080 resolution and a 42-cd/m2 mean luminance (RGB = 128) were placed side by side as the tiled display for this experiment. The width and height of each 2300 display were 51 and 29 cm, respectively. Therefore, the total width and height of the tiled display were 153 and 29 cm, respectively, and the total resolution was 5760
1080. The distance between the participants and the display was set at 70 cm, and the environment luminance was controlled as 320 ± 30 lux. The tiled display was arranged in the layout shown in Fig. 3. The three 2300 displays were placed side by side on a surface, as shown in Fig. 3(a); when measured at 25° in periphery, the angle between the line of sight and the front panel was 65°, and the distance was approximately 78 cm. When measured at 45° in periphery, the angle between the middle display and the remaining two displays was 160° so that both the viewing distance and angle between the line of sight and the front panel were the same as those measured at 25°. With such a layout, glyph arrays of the same size at either 25° or 45° would generate the same visual angle and luminance level for a participant. 3.2. Experimental design and task Four factors were used as independent variables in this experiment and are outlined as follows: Length: the length of the glyph array at four levels; that is, 2, 4, 8, and 16 glyphs, respectively corresponding to 3.2, 6.4, 12.8, and 25.6 cm on the display. Width: the width of the glyph array at four levels; that is, 1, 2, 4, and 8 glyphs, respectively corresponding to 1.6, 3.2, 6.4, and 12.8 cm on the display. Frequency: the rhythm of a glyph at three levels; that is, 0.25, 0.5, and 1 Hz. Viewing angle: the position of the glyph array centered at 25° and 45° from the center of the display along the nasotemporal meridian. The glyph arrays of the same width presented on the display at two viewing angles subtend the same visual angles, as shown in Fig. 3. Given a participant sitting at 70 cm in front of the display, when viewed at either 25° or 45°, the visual angles corresponding to 1, 2, 4, and 8 glyphs in width are 1.0°, 2.0°, 4.0°, and 8.0° respectively. The dotted lines in Fig. 3(a) and (b) represent the visual angle of the largest glyph arrays (i.e., 8.0°) and the corresponding locations on the two display layouts for two viewing angles. All the four factors were treated as within-subject factors and repeated twice for measuring the phase-shift JND. The frequency and viewing angle are block factors, and each participant interacted with six blocks (3
2) in random order. Under each block,

53

the 16 (4
4) combinations of glyph arrays were presented to the participants in random order. There were 192 trials (4
4
3
2
2). A testing platform developed by Microsoft Visual Basic 6.0 was used to control the experiment. In each trial, the participants were presented with a glyph array of predefined size with a 0° phase shift at the beginning. For example, one trial may have involved a glyph array of size 16
2 (960
120 pixels) changing at 1 Hz and viewed from 25°. At the beginning with a 0° phase shift, this glyph array is simply a large gray rectangle with changing shades following a sine wave function at 1 Hz. The experiment administrator subsequently pressed the Up key to increase the phase shift by one unit. When the administrator released the Up key, the top-left glyph of the glyph array was maintained at the same gray level, and the gray levels of the remaining glyphs changed according to Eq. (1) and their corresponding positions in the glyph array. The participants were requested to answer the question ‘‘Is this pattern different from the previous one?” If a participant answered ‘‘No,” the administrator continued to increase the phase shift until the participant detected a difference. The JND was recorded in a database as a dependent variable for analysis. The participants were allowed breaks between trials to prevent fatigue. Throughout the experiment, to ensure that the participants used only peripheral vision to detect the glyph array, we explicitly instructed them to focus on the center of the tiled display where a pseudo primary task was provided (staring at a fixed dot on the display), and an eye tracker was used to monitor their eye movements. The image captured by the eye tracker was projected to another display in real time and was monitored by an experiment administrator. If it was determined that a participant did not fixate on the center of the display during the experiment, the results from the affected trials were deleted from the collected data set and the participant had to undergo the deleted trials again at the end of each block. 3.3. Results 3.3.1. Phase shift JND of the glyph array Table 1 lists the means and standard deviations of the measured JND of the phase shifts for the four factors, length, width, frequency, and viewing angle. The results indicated that the participants’ JNDs of the phase shift at a viewing angle of 45° were higher than those at 25°. The four-way repeated measures ANOVA results revealed three significant main effects: viewing angle: F(1, 9) = 31.724, p < 0.001; length: F(3, 27) = 41.594, p < 0.001; and width: F(3, 27) = 54.482, p < 0.001. When the viewing angle was increased, the participants’ JNDs of the phase shift significantly increased (p < 0.001). The LSD test results indicated that the participants’ JNDs of the phase shift were significantly different among the four levels of length (p < 0.001) and four levels of width (p < 0.001). The ANOVA results indicated three significant two-way interaction terms: length
viewing angle F(3, 27) = 14.999, p < 0.001; width
viewing angle F(3, 27) = 23.828, p < 0.001; and length
width F(3, 27) = 23.323, p < 0.001. Fig. 4 shows these three two-way interaction terms. Fig. 4(a) shows the relationship between the viewing angle and length, indicating that despite the JNDs of the phase shift at a viewing angle of 45° being higher than those at 25° for all length levels, the differences in the JNDs between the 45° and 25° viewing angles decreased when the length was increased. The LSD test revealed that when the length was 2, the participants’ JNDs of the phase shift at the 45° viewing angle was significantly higher than that at the 25° viewing angle (p < 0.001). However, the difference became insignificant when the length was equal to or greater than 4 glyphs (p > 0.05). The second significant relationship was between the viewing angle and

54

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

Fig. 3. Three 2300 displays with 1920
1080 resolution were placed side by side for the tiled display in this experiment. The width and height of each 2300 display were 51 and 29 cm, respectively. Fig. 3(a) shows the top views of the three 2300 displays, placed side by side on a surface, when measured at 25° in periphery. The distance between the participants and the display was 70 cm. The angle between the line of sight and the front panel was 65° and the distance was approximately 78 cm. Fig. 3(b) shows that the angle between the middle display and the remaining two displays was 160°, so that when measured at 45° in periphery, both the viewing distance and angle between the line of sight and the front panel were the same as those measured at 25° in periphery.

Table 1 The just-noticeable difference (JND) of the phase shift was measured as a dependent variable to evaluate the design that involves using dynamic patterns generated by the glyph arrays presented at two levels, 25° and 45° shifts from the center of the display along the nasotemporal meridian. Means and standard deviations of the four main effects: length, width, frequency, and viewing angle for the JND of the phase shift were calculated for all participants.

Mean SD

Viewing Angle

Frequency

25

45

0.25

0.5

1

2

Length 4

8

16

1

2

4

8

2.186 1.528

3.322 2.890

2.843 2.333

2.667 2.504

2.750 2.301

4.490 3.443

2.874 1.981

2.077 1.176

1.573 0.808

4.163 3.069

3.214 3.069

2.095 2.578

1.536 1.352

width (p < 0.001). As illustrated in Fig. 4(b), these results were highly consistent with those of the first two-way interaction term. Further illustrated in Fig. 4(c), the marginal contribution to the JND of the phase shift decreased, regardless of whether the length or width was increased. This suggests that the size of the glyph arrays influenced the JND of the phase shift. This finding raised the question of how the size of a glyph array (i.e., the total number of glyphs in a glyph array) affects detection when peripheral vision is used and whether the shape of a glyph array affects the JND of the phase shift. 3.3.2. Size and shape of glyph arrays The same set of data was used for this analysis. The length and width factors were recoded into another factor, the glyph number,

Width

which represents the total number of glyphs contained in the corresponding glyph array. Thus, this model comprised three factors: the glyph number: (2, 4, 8, 16, 32, 64, and 128), frequency (0.25, 0.5, and 1 Hz), and viewing angle (25°, 45°). Table 2 lists the means and standard deviations of the measured JND of the phase shift for the three main effects. These results were consistent with our expectation that when the glyph number increased, the participants’ JNDs of the phase shift decreased. The three-way repeated measures ANOVA test results revealed two significant main effects: glyph number: F(3, 27) = 60.654, p < 0.001; and viewing angle: F(1, 9) = 35.254, p < 0.001. The LSD test results indicated that the participants’ JNDs of the phase shift were significantly different among all seven levels of the glyph number (p < 0.01). As shown in Fig. 5(a), one two-way interaction

55

(a)

6

Phase-shift Threshold (deg)

(a)

Phase-shift Threshold (deg)

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

4 45° 25°

2

0 2

4

8

16

45° 25°

4

2

4

8

16

32

64

128

Number of Glyphs 4

(b)

45° 25°

2

1

2

4

8

Width of Glyphcy Arrays

Phase-shift Threshold (deg)

6

2

0

(c)

8

0

6

Phase-shift Threshold (deg)

(b)

Phase-shift Threshold (deg)

Length of Glyph Arrays

10

10

8

Width of Glyph Arrays 2*1

2 8

6

2*2

4

4*1

8*1 4*2 2*4

16*1 8*2 4*4 2*8

16*2 8*4 4*8

2 0

4 16

2

4

8

8

16

32

16*4 8*8

64

16*8

128

Number of Glyphs Width of Glyph Arrays 1 4 8 16

6

Fig. 5. Instead of using different sizes of glyph arrays, the total number of glyphs contained in a glyph array was analyzed. Fig. 5(a) shows the interaction term of the glyph number and viewing angle. Fig. 5(b) shows the shape effect of the glyph arrays; the four symbols represent JNDs for four levels of glyph array length.

4 2 0

2

4

8

The results suggested that when the glyph number was greater than 4, the total number of glyphs in the array dominated the JND of the phase shift, whereas the shape of the array had little to no effect on human perception.

16

Length of Glyph Arrays Fig. 4. The three significant two-way interaction terms for JNDs of the phase shift obtained from Experiment I. Fig. 4(a) shows the relationship between the viewing angle and length of glyph arrays. Fig. 4(b) shows the interaction term between the angle and width of glyph arrays. Fig. 4(c) shows the interaction term between the length of glyph arrays and width of glyph arrays.

3.4. Discussion The measured JNDs of the phase shift suggest that the size of the glyph arrays exerts the greatest effect on the detectability, compared with the location or shape. Although the location (viewing angle) at which the glyph array is presented also plays a vital role in influencing the perception of human eyes on the JND of the phase shift, the location effect may decrease when the length or width increase, as illustrated in Fig. 4(c). This means that a user’s detection performance on the glyph array can be compensated by its size. The statistical analysis results shown in Fig. 5(a) indicate no significant difference between the two viewing angles (25° and 45°) when the number of glyphs is equal to or greater than 8. We further tested 16 shapes of glyph arrays, and the analysis results illustrated in Fig. 5(b) suggest that when the number of glyphs is equal to or greater than 8, the shape of a glyph array does

term, viewing angle
glyph number, was significant: F(6, 54) = 57.742 (p < 0.001). The JND between the 45° and 25° viewing angles clearly decreased when the glyph number increased. In addition to the total number of glyphs, we also sought to determine whether glyph arrays comprising the same number of glyphs but different shapes would generate similar JNDs. The four types of symbols indicated in Fig. 5(b) represent JNDs for four levels of glyph array length. Among all the contrast analyses based on the same number of glyphs, the participants’ JNDs of the phase shift in shape were significant only when the glyph number was 4, where 2
2 had a JND larger than 4
1 (p < 0.01). Other contrast pairs based on different lengths but the same glyph number (e.g., 8
2 vs. 4
4) showed no statistical significance.

Table 2 Instead of using different sizes of glyph arrays, the total number of glyphs contained in a glyph array was analyzed. Means and standard deviations for the three main effects: glyph number, frequency, and viewing angle for the JND of the phase shift were calculated for all participants.

Mean SD

Viewing Angle

Frequency

25

45

0.25

0.5

1

2

Glyph Number 4

8

16

32

64

128

2.184 1.608

3.323 3.023

2.841 2.473

2.670 2.584

2.750 2.400

7.166 4.306

4.887 3.281

3.117 1.858

2.301 1.138

1.633 0.793

1.315 0.632

1.031 0.455

56

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

not affect the JND; instead, only the number of glyphs affects the JND. Notably, when the stimulus size increases, given the stimuli centered at 45° and 25°, the stimuli do not overlap at the two measured eccentricities, as illustrated in Fig. 3. However, the stimulus size affects the viewing angle in periphery, as indicated by the dotted line in Fig. 3. The more central part of a large stimulus reduces the viewing angle in periphery and enhances the performance, whereas the more peripheral parts of a large stimulus would not contribute markedly to the performance. 4. Experiment II: discriminability of glyph arrays In Experiment I, we observed the detectability when participants interacted with the glyph array by using peripheral vision. If this glyph array is to be used as a notification system, it is essential to evaluate the amount of information that can be perceived or the amount of distinct stimuli that users can accurately identify. Thus, in Experiment II, the performance of the discriminability of human eyes on glyph arrays using peripheral vision was examined. First, the size and shape of the glyph array to be tested in Experiment II must be determined. Obviously, as the glyph number increased, the JND of the phase shift decreased. This implies that a high number of glyphs enhances the ability of human eyes to detect the changes in the dynamic pattern of glyph arrays. However, the marginal contribution to the JND of the phase shift clearly decreases as the size of the glyph array increases. Designing a glyph array as a notification system necessitates balancing between the detectability and size of a glyph array required on a display. According to this requirement and the information shown in Fig. 5(b), we selected a glyph array with a size of 16
2 (960
120 pixels) as the design for the notification system used in Experiment II. The width of 120 pixels, corresponding to 2.0° visual angle, occupies only a small space on the screen and an array larger than 32 glyphs contributes a limited improvement on the JND of the phase shift. For a glyph array of fixed size and shape (i.e., 16
2 in this case), the frequency or phase shift can be used as a coding dimension to create a notification system. For example, if the frequency is maintained at a certain level (e.g., 1 Hz) and if four phase-shift levels are used (e.g., 5, 35, 75, and 105 time units), a notification system comprising a set of four different stimuli can be created. The number of such different stimuli that can be accurately recognized by participants represents the discriminability of this design. In the described case, the phase shift was used as a single coding dimension in the glyph array. Similarly, a notification system in which the frequency is used as a single coding dimension can also be created. Furthermore, both the frequency and phase shift can be manipulated simultaneously as two coding dimensions to create a notification system. In Experiment II, two single-dimensional designs and one two-dimensional design were examined. The settings, apparatus, and participants of this experiment were the same as those in Experiment I. 4.1. Single-dimensional stimuli: phase shift 4.1.1. Design and procedures for single-dimensional stimuli: phase shift In the first single-dimensional design, the phase shift was used as an experimental stimulus and was tested at three frequency levels (i.e., 0.25, 0.5, and 1 Hz) and viewed at two angles (i.e., 25° and 45°). The amount of notification information was tested at four levels: two stimuli (phase shift = 5 and 105 time units), four stimuli (phase shift = 5, 35, 75, and 105 time units), six stimuli (phase shift = 5, 15, 35, 55, 75, and 105 time units), and eight stimuli

(phase shift = 5, 15, 25, 35, 55, 75, 85, and 105 time units), where one time unit equals t 360/512 = 0.703°. Thus, this stage of the experiment comprised three factors: the amount of notification information, viewing angle, and frequency. The three factors were considered within-subject factors. The six (2
3 = 6) combinations of the viewing angle and frequency were considered blocks and the participants performed six combinations in random order. Within each block, the participants underwent four levels of the amount of notification information from two to eight stimuli in a fixed order. In other words, the participants were tested with two stimuli first, followed by four, six, and finally eight stimuli in a fixed order. Because the levelsetting rule requires that the items in the smaller amount of notification information for the phase shift be the subset of the items in the larger amount of notification information for the phase shift, this rule and the fixed testing order provided optimal memory for the participants. Before the experiment, the participants were instructed by the experiment administrator to familiarize themselves with the process of the experiment. Each of the four sessions comprised two subsessions: training and testing. Training: In the training subsession, the experiment administrator presented the stimuli of the specific session to the participants. The participants were requested to perceive information by using their peripheral vision and to memorize the dynamic patterns generated by the corresponding glyph array and associate that pattern to a specific number, which was considered the correct answer at the subsequent testing stage. There was no time limit for the training subsession. When the participants were ready, the administrator proceeded to the testing subsession. Testing: There were 20 trials in the testing subsession. The stimuli of this specific session were randomly presented to the participants, who were requested to answer the question ‘‘What is the corresponding number of this pattern?” The testing session was completed after 20 questions. The accuracy rate of these 20 trials (i.e., number of correct answers/20) was considered a data point of the dependent variable corresponding to the specific session within one block. The participants were allowed to pause the experiment at any time when a testing subsession ended to prevent mental or visual fatigue. On completion of a testing subsession, the training subsession of another session was evaluated until all the four sessions of one block (one of six combinations of the viewing angle and frequency) were completed. When a participant finished one block, he or she rested for at least 1 h before proceeding to the next block. 4.1.2. Result from single-dimensional stimuli: phase shift Table 3 lists the means and standard deviations for accuracy rates of the three main effects among all 10 participants. The results are consistent with our expectation that the accuracy rate decreases as the number of stimuli increases. The results also indicate that the participants demonstrated favorable performance when the gray levels of the glyph arrays were changed at lower frequencies and viewed at smaller angles. The three-way repeated measures ANOVA test results indicate three significant main effects: the amount of notification information, F(3, 27) = 60.654, p < 0.001; frequency, F(2, 18) = 6.449, p < 0.01; and viewing angle, F(1, 9) = 5.444, p < 0.05. The post hoc test indicated that the amount of notification information comprised four groups, meaning that the participants’ accuracy rates at four levels are significantly different from each other (p < 0.05), as illustrated in Fig. 6(a). Regarding the factor of frequency, when the frequency was equal to 0.25 Hz, the participants’ accuracy rates were more favorable than those obtained when the frequency was equal to 0.5 or 1 Hz (p < 0.01); however, there was no significant difference between the accuracy rates when the fre-

57

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

Table 3 Using the phase shift as the single coding dimension. Means and standard deviations of the accuracy rates for the three main effects: amount of notification information, frequency, and viewing angle were calculated for all participants. Amount of Notification Information

Mean SD

Frequency

4

6

8

0.25

0.5

1

25

45

1.000 0.000

0.877 0.153

0.665 0.200

0.583 0.170

0.825 0.202

0.766 0.231

0.753 0.234

0.808 0.212

0.754 0.234

quency was equal to 0.5 or 1 Hz (p > 0.05). There was no significant interaction term in this analysis. Notably, the obtained accuracy rates may not be normally distributed. Nonetheless, an ANOVA is not sensitive to moderate deviations from normality. Simulation studies [43,44], using various nonnormal distributions, have shown that the false-positive rate is not affected substantially. Because the significant main effects observed in Experiment II were mostly highly significant with P values of less than 0.001, the deviations from normality should not affect the interpretation of these terms.

Single Dimension-PhaseShift

(a) 100% Accuracy rate (%)

Viewing Angle

2

80% 60% 40%

4.2. Single-dimensional stimuli: frequency 45°

20%

25°

0%

2

4

6

8

Number of Stimuli

Single Dimension-Frequency

Accuracy rate (%)

(b) 100% 80% 60% 40% 45°

20%

25°

0%

2

4

8

6

Number of Stimuli

Multi Dimension -Phase Shift * Frequency

Accuracy rate (%)

(c) 100% 80% 60% 40% 45°

20% 0%

25°

2

4

6

8

10

12

14

16

Number of Stimuli Fig. 6. In Experiment II, the performance of the discriminability of human eyes on glyph arrays using peripheral vision was examined. Fig. 6(a) shows the accuracy rates of the four levels of stimuli for two viewing angles when the phase shift was used as the single coding dimension. The numbers of stimuli were from 2 to 8. Fig. 6 (b) shows the accuracy rates of the four levels of stimuli for two viewing angles when the frequency was used as the single coding dimension. The numbers of stimuli were from 2 to 8. Fig. 6(c) shows the accuracy rates for two viewing angles when the phase shift and frequency were used simultaneously to create the eight levels of stimuli. The numbers of stimuli were from 2 to 16.

4.2.1. Design and procedures for single-dimensional stimuli: frequency The second single-dimensional design was similar to the first design, except that the frequency (instead of the phase shift) was controlled as the experimental stimuli. The six experimental blocks comprised two peripheral viewing angle levels (i.e., 25° and 45°) and three phase-shift levels (i.e., 5, 50, and 105 time units). The amount of notification information was also tested at four levels: two stimuli (frequency = 0.25 and 2 Hz), four stimuli (frequency = 0.25, 0.50, 1.0, and 2 Hz), six stimuli (frequency = 0.25, 0.33, 0.5, 0.67, 1, and 2 Hz), and eight stimuli (frequency = 0.25, 0.29, 0.33, 0.4, 0.5, 0.67, 1, and 2 Hz). Thus, the second singledimensional experiment also comprised three factors: the amount of notification information, viewing angle, and phase shift. The three factors were also considered within-subject factors. The six (2
3 = 6) combinations of the viewing angle and phase shift were considered blocks and the participants also performed six combinations in random order. Within each block, the participants underwent four levels of the amount of notification information for frequency, from two to eight (four sessions), in a fixed order on the basis of the same rationale for Experiment I. The training and testing procedures were the same as those in the previous experiment, except that the stimuli in Experiment II comprised different frequencies instead of phase shifts. 4.2.2. Results from single-dimensional stimuli: frequency Table 4 lists the means and standard deviations for the accuracy rates of the three main effects. The results, again, indicate that when the amount of stimulus increased, the accuracy rate decreased. Although the results presented in Table 4 imply that lower phase shifts and narrower viewing angles generate optimal performance, the three-way repeated measures ANOVA test results indicate only one significant main effect: the amount of notification information, F(3, 27) = 91.910, p < 0.001. The post hoc test results reveal that the participants’ accuracy rates were significantly different from each other at the four levels (p < 0.05), as shown in Fig. 6(b). There was no significant interaction term. 4.3. Multidimensional stimuli: frequency
phase shift 4.3.1. Design and procedures for multidimensional stimuli According to human information processing theory, multidimensional design is a more effective method for improving discriminability compared with single-dimensional design [45].

58

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

Table 4 Using the frequency as the single coding dimension. Means and standard deviations of the accuracy rates for the three main effects: amount of notification information, phase shift, and viewing angle were calculated for all participants. Amount of notification information

Mean SD

Phase shift

Viewing Angle

2

4

6

8

5

25

45

25

45

1.000 0.000

0.783 0.164

0.593 0.176

0.548 0.173

0.751 0.207

0.723 0.238

0.720 0.249

0.740 0.230

0.723 0.234

itive effect of the viewing angle was not obvious when the number of stimuli was low (64) or high (P14). The two-way repeated measures ANOVA test for the accuracy rate on the multidimensional stimuli (phase shift
frequency) indicated two significant main effects, which are the amount of notification information, F(7, 63) = 80.758, p < 0.001 and viewing angle, F(1, 9) = 65.162, p < 0.001. The two-way interaction term between the amount of notification information and viewing angle shown in Fig. 6(c) was also significant, F(7, 63) = 4.591, p < 0.001. The contrast test indicated that the accuracy pairs between 25° and 45° were statistically significant only when the number of stimuli was equal to 6, 8 10, and 12. Other accuracy pairs were not significant (p > 0.05).

Therefore, in the third part of Experiment II, we simultaneously manipulated two dimensions—the phase shift and frequency—to create stimuli for notification information and to measure the participants’ performance on such patterns. The design of this experiment was similar to those of the previous two experiments. However, because the phase shift and frequency were controlled simultaneously to create the stimuli as notification information in this experiment, this experiment comprised only two factors: the amount of notification information and viewing angle. The amount of notification information was tested at eight levels, from 2 to 16 stimuli (i.e., 2, 4, 6, 8, 10, 12, 14, and 16 stimuli). The maximal number of the stimuli tested in this experiment was 16, and these 16 stimuli were combined orthogonally from four phase-shift levels (5, 35, 75, and 105 time units) and four frequency levels (0.25, 0.50, 1, and 2 Hz). The rules for setting the levels were similar to those used in the previous experiments in that the items in the smaller amount of notification information were the subset of the items in the larger amount of notification information. For example, if the participants underwent the experiment when the amount of notification information was equal to 2, then the first stimulus was frequency = 2, phase shift = 5, whereas the second stimulus was frequency = 0.25, phase shift = 105. If the participants underwent the experiment when the amount of notification information was equal to 4, in addition to the two mentioned stimuli, then the third stimulus was frequency = 1, phase shift = 35, and the final stimulus was frequency = 0.5, phase shift = 75. The two viewing angle levels were considered random blocks, and within each block, the participants underwent eight levels of the amount of notification information, from 2 to 16 stimuli (eight sessions), in fixed order on the basis of the same rationale for the previous two experiments. The training and testing procedures in this experiment were the same as those in the previous two single-dimensional design experiments. Because some sessions were associated with a high number of stimuli in this experiment, the training subsession could last a relatively long time. In some cases, participants required more than 30 min for a single training subsession, even with the help of the accumulative memorizing scheme adapted by the level-setting rule.

4.4. Discussion In general, the results from the second part of this study show that when two single-dimensional designs that include different coding dimensions are compared, the accuracy rates of the design that involves using the phase shift as the coding dimension are always higher than those of the design that entails using the frequency, except for the cases in which the number of stimuli is equal to 2 (the accuracy rates are 100% for both cases). Further observation of two single-dimensional designs indicated that despite the viewing angle influencing the accuracy rates of the phase shift (p < 0.05), it does not affect those of the frequency (p > 0.05). In addition, as shown in Fig. 6(a) and 6(b), the accuracy rates drop faster as the number of stimuli increases for the case of frequency compared with the case of the phase shift. Thus, compared with the phase shift, frequency, when used as a coding dimension, is less sensitive to viewing angles, but is more sensitive to the number of stimuli. The comparison among the three experiments that involve measuring discriminability suggests that the multidimensional design consistently outperformed the other two singledimensional designs across all levels of notification information comprising up to eight stimuli at both viewing angles. Comparing the cases in which the amount of notification information increased by up to eight stimuli show that the accuracy rates of the multidimensional design are significantly higher than those of both single-dimensional deigns (p < 0.01). The benefits are particularly notable when the viewing angle is 25°. In this study, the amount of notification information was considered an experimental factor and used to verify the effectiveness of the proposed design of the glyph array. However, when the number of stimuli increases, the difference between the stimuli to be discriminated decreases. For example, in the two-stimulus

4.3.2. Results from multidimensional stimuli: phase shift
frequency The experimental results shown in Table 5 are consistent with our expectation. When the amount of notification information increased, the participants’ discriminability decreased. These trends were consistent throughout the three experiments. Furthermore, Fig. 6(c) indicates that although the accuracy rates at the viewing angle of 25° are more favorable than those at 45°, the pos-

Table 5 Using the phase shift and frequency as two coding dimensions simultaneously. Means and standard deviations of the accuracy rates for the two main effects: amount of notification information, and viewing angle were calculated for all participants. Viewing angle

Mean SD

Amount of notification information

25

45

2

4

6

8

10

12

14

16

0.711 0.182

0.621 0.194

1.000 0.000

0.920 0.046

0.830 0.146

0.745 0.129

0.555 0.140

0.460 0.149

0.425 0.136

0.395 0.129

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

session, it is relatively easy to detect the stimulus of 0.25 Hz because the other one is 2 Hz. However, in the four-stimulus session, it may be not as easy to detect the same stimulus of 0.25 Hz because it is more likely to be confused with a closer stimulus of 0.5 Hz. The situation becomes even more complicated in the multidimensional session. Thus, the effect of the different number of stimuli may need to be addressed in addition to the difference between neighboring stimuli in the current experimental design. When the number of stimuli increases, the observed deterioration in discriminability performance in the current study may be partially due to the limit imposed by human working memory span (memory constraint), and partially caused by a smaller difference between neighboring stimuli (sensory constraint). 5. Conclusion This study designed and validated a novel notification system by applying the concept of calm technology. The main design principle involves ensuring that the dynamic patterns generated by the glyph array are easily detected using peripheral vision without requiring users to shift their visual attention from the focal area. Also, the onset of the dynamic patterns is designed elegantly and smoothly without using any auditory stimuli or abrupt visual cue. 5.1. Flexibility in geometric shapes for aesthetic requirements A crucial finding of this study is that when the glyph array comprises eight or more glyphs, the shape of a glyph array does not affect the JND in detecting the phase shift. When applying this peripheral notification system in real designs, a designer may have more flexibility to explore different geometric shapes to fulfill aesthetic requirements, because the shape of a glyph array with such a design minimally affects the JND. 5.2. Choosing between single-dimensional and multidimensional designs According to the results, one may conclude that the multidimensional design is more favorable. However, from the observation during training sessions, it is apparent that the participants required more time to learn the dynamic patterns generated from glyph arrays, and some stimulus combinations were more easily confused. This situation was not observed in the cases of the single-dimensional design. Thus, if the number of stimuli is low for a notification system, we recommend a single-dimensional design that entails using the phase shift as a coding dimension. If the number of stimuli for the notification system is high and a multidimensional design is required, the stimulus combinations must be carefully selected to ensure that each stimulus can be distinguished with equal effort. 5.3. Long-term effect and rich context The accuracy rates reported in this study associated with a lower number of stimuli may be underestimated because considering the accumulative memorizing scheme adapted in the level setting rule indicates that the levels with a lower number of stimuli appear earlier during a testing session and thus have a lower number of training trials. Using the glyph arrays in the experiment for discriminability reflects only a short-term result. Because the notification system is designed to support daily work for users, it should be tested further under the condition of long-term usage; hence, in addition to accuracy, the mental workload of users should also be considered for adopting a notification system in the real world.

59

The testing trials in this study were conducted under a contextfree condition under which each pattern was mapped only to a number. In a real-world application, it is possible to associate each pattern to a specific event (e.g., a friend logs on to Facebook) and enable the connection to be relevant (e.g., a closer friend associated with a higher frequency). Therefore, the results obtained in this study can be considered worst-case lower bounds of performance. Lower mental workload and richer context can be easily adapted in real applications. Nonetheless, this study provides useful guidelines for designers to understand how particular factors fundamentally affect user performance in this novel notification system design.

Acknowledgment This work is partially supported by NSC, Taiwan, R.O.C. under grants NSC 97-2221-E-035-067-MY3.

References [1] C. Andrews, A. Endert, B. Yost, C. North, Information visualization on large, high-resolution displays: issues, challenges, and opportunities, Inf. Vis. 10 (2007) 341–355. [2] C. Plaue, J. Stasko, Animation in a peripheral display: distraction, appeal, and information conveyance in varying display configurations, in: Proceedings of GI’07. [3] D. Angelucci, A. Cardinali, L. Tarantino, A customizable glanceable peripheral display for monitoring and accessing information from multiple channels, Information Technology and Innovation Trends in Organizations, 2011, pp. 199–207. [4] L. Ventä, M. Isomursu, A. Ahtinen, S. Ramiah, My phone is a part of my soul – how people bond with their mobile phones, in: Proceedings of UBICOMM ’08, 2008, pp. 311–317. [5] H. Müller, M. Pielot, R. Oliveira, Towards Ambient Notifications. Peripheral Interaction Workshop, INTERACT 2013. [6] D.S. McCrickard, M. Czerwinski, L. Bartram, Introduction: design and evaluation of notification user interfaces, Int. J. Hum. Comp. Stud. 58 (2003) 509–514. [7] D.S. McCrickard, Maintaining information awareness with Irwin, in: Proceedings of the World Conference on Educational Multimedia/ Hypermedia and Educational Telecommunications (EDMEDIA ’99), 1999, pp. 552–557. [8] Q.A. Zhao, J.T. Stasko, What’s happening? Promoting community awareness through opportunistic, peripheral interfaces, in: Proceedings of the working conference on AVI’02, 2002, pp. 69–74. [9] S.T. Iqbal, E. Horvitz, Notifications and awareness: a field study of alert usage and preferences, in: Proceedings of CSCW ’10, ACM, 2010. [10] S.D. Paolo, L. Tarantino, A peripheral notification display for multiple alerts: design rationale, Inf. Syst.: People Organ. Inst. Technol. 9 (2010) 521–528. [11] P. Tarasewich, C.S. Campbell, T. Xia, M. Dideles, Evaluation of visual notification cues for ubiquitous computing, Lecture Notes of Computer Sciences, vol. 2864, 2003, pp. 349–366. [12] J.S. Weber, M. Newman, M.E. Pollack, Multi-format notifications for multitasking, in: Proceedings of INTERACT, 2009, pp. 247–260. [13] S.T. Iqbal, B.P. Bailey, Oasis: a framework for linking notification delivery to the perceptual structure of goal-directed tasks, ACM Trans. Comput. – Hum. Interact. (TOCHI) 17 (4) (2010) 15. 1–28. [14] D. Warnock, M.R. McGee-Lennon, S.A. Brewster, The role of modality in notification performance, Interact 2 (6947) (2011) 572–588. [15] E. Arroyo, T. Selker, A. Stouffs, Interruptions as multimodal outputs: which are the less disruptive? in: Proceedings of 4th IEEE International Conference on Multimodal Interfaces, 2002, pp. 479–482. [16] B. Deatherage, Auditory and other sensory forms of information presentation, Human Engineering Guide to Equipment Design, Government Printing Office, Washington, DC, 1972. [17] E.D. Mynatt, J. Rowan, A. Jacob, S. Craighill, Digital family portraits: supporting peace of mind fir extended family members, in: Proceedings of CHI’01, 2001, pp. 333–340. [18] T. Skog, S. Ljungblad, L.E. Holquist, Between aesthetics and utility: designing ambient information, in: Proceedings of InfoVis, 2003, pp. 233–240. [19] J. Stasko, T. Miller, Z. Pousman, C. Plaue, O. Ullah, Personalized peripheral information awareness through information art, in: Proceedings of Ubicomp, 2004, pp. 18–35. [20] N. Streitz, C. Magerkurth, T. Prante, C. Roker, From information design to experience design: smart artifacts and the disappearing computer, Interactions 12 (4) (2005) 21–25. [21] X. Shen, An evaluation methodology for ambient displays, J. Eng. Comput. Archit. 1 (2) (2007) 38–46.

60

K.H. Tang, Y.H. Lee / Displays 41 (2016) 50–60

[22] N. Jafarinaimi, J. Forlizzi, A. Hurst, J. Zimmerman, Breakaway: an ambient display designed to change human behavior, in: Proceedings of CHI’05, Late Breaking Results, 2005, pp. 1945–1948. [23] J. Mankoff, A.K. Dey, G. Hsieh, J. Kientz, S. Lederer, M. Ames, Heuristic evaluation of ambient displays, in: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, ACM, New York, NY, 2003, pp. 169– 176. [24] Y.H. Hung, M. Ostovari, Using cursor hints to supplement a less interruptive desktop notification, in: Proceedings of the 58th Human Factors and Ergonomics Society, 2014, pp. 1521–1525. [25] D.S. McCrickard, R. Catrambone, J.T. Stasko, Evaluating animation in the periphery as a mechanism for maintaining awareness, in: Proceedings of the IFIP Conference on Human-Computer Interaction (INTERACT ‘01), Georgia Tech Institute, 2001, pp. 148–156. [26] X. Bi, R. Balakrishnan, Comparing usage of a large high-resolution display to single or dual desktop displays for daily work, in: Proceedings of CHI 2009, pp. 1005–1014. [27] S. Parsowith, G. Fitzpatrick, B. Segall, S. Kaplan, Tickertape: notification and communication in a single line, in: Proceedings of 3rd Asia Pacific Computer Human Interaction (APCHI ’98), Kanagawa, Japan, 1998, pp. 139–144. [28] S.J. Anderson, K.T. Mullen, R.F. Hess, Human peripheral spatial resolution for achromatic and chromatic stimuli: limits imposed by optical and retinal factors, J. Physiol. 442 (1991) 47–64. [29] G.E. Legge, J.S. Mansfield, S.T.L. Chung, Psychophysics of reading XX. Linking letter recognition to reading speed in central and peripheral vision, Vision Res. 41 (6) (2001) 725–743. [30] S.T.L. Chung, G.E. Legge, B.S. Tjan, Spatial-frequency characteristics of letter identification in central and peripheral vision, Vision. Res. 42 (18) (2002) 2137–2152. [31] K.M. Joffe, C.T. Scialfa, Texture segmentation as a function of eccentricity, spatial frequency and target size, Spat. Vis. 9 (1995) 325–342. [32] R.F. Hess, S.C. Dakin, Absence of contour linking in peripheral vision, Nature 190 (1997) 602–604. [33] A.K. Nugent, R.N. Keswani, R.L. Woods, E. Peli, Contour integration in peripheral vision reduces gradually with eccentricity, Vision. Res. 43 (2003) 2427–2437.

[34] S.G. Kuai, C. Yu, Constant contour integration in peripheral vision for stimuli with good Gestalt properties, J. Vision 6 (12) (2006) 1412–1420. [35] G. Osterberg, Topography of the layer of rods and cones in the human retina, Acta Ophthamol. Suppl. 6 (1935) 1–103. [36] J. Gordon, I. Abramov, Colour vision in the periphery. II. Hue and saturation, JOSA 67 (2) (1977) 202–207. [37] T. Matthews, Designing and evaluating glanceable peripheral displays, in: Proceedings of MM2004, ACM International Conference on Multimedia. [38] L. Bartram, C. Ware, T. Calvert, Moving icons: detection and distraction, in: Proceedings of the Eight International Conference on Human-Computer Interactions (INTERACT 2001), Tokyo, Japan, 2001, pp. 157–165. [39] J.Y. Park, T.J. Nam, Understanding dynamic design elements for information representation of ambient media, in: Proceedings of IASDR 2007, Hong Kong. [40] S. Yamada, K. Kobayashi, N. Mori, Peripheral cognition technology: approach and implementation, in: Proceedings of 2012 International Workshop on Human-Agent Interaction (iHAI 2012), TW8_0001, 2012. [41] S. Bakker, D. Hausen, T. Selker, E. van den Hoven, A. Butz, B. Eggen, Peripheral interaction: shaping the research and design space, in: Proceedings of CHI’14 Extended Abstracts on Human Factors in Computing Systems, 2014, pp. 99– 102. [42] X. Bi, R. Balakrishnan, Comparing usage of a large high-resolution display to single or dual desktop displays for daily work, in: proceedings of the 27th International Conference on Human Factors in Computing Systems, Boston, USA, 2009. [43] M.R. Harwell, E.N. Rubinstein, W.S. Hayes, C.C. Olds, Summarizing Monte Carlo results in methodological research: the one- and two-factor fixed effects ANOVA cases, J. Educ. Stat. 17 (1992) 315–339. [44] L.M. Lix, J.C. Keselman, H.J. Keselman, Consequences of assumption violations revisited: a quantitative review of alternatives to the one-way analysis of variance F test, Rev. Educ. Res. 66 (1996) 579–619. [45] C.D. Wickens, Engineering Psychology and Human Performance, 2nd ed., Harper Collins, New York, 1992.