Effects of Resolution,Range,and Image Contrast on Target Acquisition Performance

Abstract

Objective:

We sought to determine the joint influence of resolution, target range, and image contrast on the detection and identification of targets in simulated naturalistic scenes.

Background:

Resolution requirements for target acquisition have been developed based on threshold values obtained using imaging systems, when target range was fixed, and image characteristics were determined by the system. Subsequent work has examined the influence of factors like target range and image contrast on target acquisition.

Method:

We varied the resolution and contrast of static images in two experiments. Participants (soldiers) decided whether a human target was located in the scene (detection task) or whether a target was friendly or hostile (identification task). Target range was also varied (50–400 m). In Experiment 1, 30 participants saw color images with a single target exemplar. In Experiment 2, another 30 participants saw monochrome images containing different target exemplars.

Results:

The effects of target range and image contrast were qualitatively different above and below 6 pixels per meter of target for both tasks in both experiments.

Conclusion:

Target detection and identification performance were a joint function of image resolution, range, and contrast for both color and monochrome images.

Application:

The beneficial effects of increasing resolution for target acquisition performance are greater for closer (larger) targets.

Keywords

sensory and perceptual processes vision displays

In recent years, electronic sighting technologies have been introduced into small arms weapon systems. The ability of a soldier to acquire targets using such systems is not fully understood. Because electronic displays divide the image into numerous pixels, the question of display resolution, in particular the resolution of the target image, becomes important. The resolution of the sensor, image processing characteristics, zoom settings, and display resolution can all affect the target resolution, that is, the number of pixels used to render the target. Target resolution can be limited by any stage in the sequence from the sensor to the display surface.

In the target acquisition literature (e.g., Johnson, 1958; Maurer, Wilson, & Driggers, 2013; Ratches, Vollmerhausen, & Driggers, 2001), the terms detection, recognition, and identification have commonly used meanings: Detection is discriminating an object of interest from its background (i.e., a target is present), recognition is classifying an object into a broad category (e.g., the target is a human), and identification is the ability to specify class membership (e.g., the human is an enemy soldier). Typically, more target detail is considered necessary to perform each successive task. Thus, determining the identity of a human target class requires the perception of greater detail (e.g., uniform texture, weapon characteristics) than simply detecting the target. In this study, we focus on detection and identification only.

Johnson (1958) proposed a set of criteria that could be used to define the threshold resolution required to detect or identify targets of various types. These were based on studies in which Johnson adjusted the range of targets until they were just detectable (or on other trials, until they were just identifiable) using different imaging systems. Gratings (black and white stripes oriented horizontally or vertically) were placed at the same range, and the spatial frequency of the grating was reduced until it could no longer be resolved. From this, Johnson deduced the minimum number of cycles (where a cycle is a pair of grating stripes) on target required to detect or identify it (Donohue, 1991; Young, Driggers, & Jacobs, 2008). The values for a standing soldier were 1.5 cycles for detection and 8.0 for identification (Johnson, 1958).

Johnson’s threshold resolution values were calculated using the minimum (or limiting) dimension of the target—for the soldier, this was the horizontal dimension. If we consider a pixel pair as a cycle, this implies that we would need 2 × 1.5 cycles = 3 horizontal pixels on a display surface to detect a standing soldier. If we assume the same resolution for both dimensions and that a soldier is 0.5 m wide and 1.83 m tall, the minimum resolution for target detection would be 3 × 11 pixels. The corresponding figure for target identification would be 16 × 59 pixels. In some applications, it is useful to calculate the pixels per meter (ppm) of target size (Peterson, 2009). Thus, if the Johnson criteria value for detection of a human target is 3 pixels for 0.5 m, then that is equivalent to 6 ppm. Similarly, we double 16 pixels to obtain 32 ppm for identification.

For our studies, we define target resolution in this way (i.e., in ppm). However, in Johnson’s approach, resolution was defined as the number of resolvable elements. These elements were limited by the system, not the human eye. In Johnson’s studies, several different imaging systems were used, each with its own display parameters, and image magnification (which affected the target size) and contrast were fixed for the target and the grating. In particular, Johnson took pains to ensure that the number of resolvable elements was limited by the system rather than the eye. However, with our definition of target resolution, if a target was very small (at a large range or at low magnification) and the display resolution very high, the pixels may subtend only a small visual angle. In this case, we may see the effects of eye-limited performance. We view this as part of the general question of how the effects of resolution interact with the range or visual angle of the target.

There is little doubt that the Johnson criteria have had a strong impact on the understanding of the detection and identification of military targets (Sjaardema, Smith, & Birch, 2015). The development of threshold values for target acquisition was a fundamental contribution. However, in fixing the values of other parameters to eliminate their effects on resolution threshold estimation, Johnson ignored the potential contribution of those factors to target acquisition performance more broadly (Donohue, 1991). For example, a target of fixed size subtends a smaller visual angle as its distance from a camera or sensor is increased. For any given display resolution, there will be fewer pixels on target as target range increases. Even if target resolution is kept constant by increasing the resolution of the sensor or display as range increases, there may still be a penalty associated with a small target.

More recent studies have gone beyond Johnson’s methods and examined the effects of factors like blur (Moyer & Devitt, 2005), target size (Barten, 2004), image quality (Vollmerhausen, Jacobs, & Driggers, 2004), and magnification (Teaney & Fanning, 2008) on target acquisition. Their results showed that such factors have important effects on target acquisition in their own right. Related modeling work explicitly incorporates the eye’s contrast sensitivity function (Vollmerhausen et al., 2004). Johnson’s methods did not allow a full examination of the effects of these factors given that he held them fixed. Furthermore, the Johnson criteria do not consider the quality of performance above threshold. Does performance improve with increased resolution above threshold? Does the improvement occur at the same rate regardless of target range? We were interested in examining if the effects of degraded image resolution would show interactions with factors like range (affecting target size) and contrast if the three factors were manipulated independently.

The Johnson criteria are based on the limiting spatial frequency at the average target contrast (Vollmerhausen, Jacobs, Hixson, & Friedman, 2006). That is, they represent resolution thresholds at average contrast levels. While it is expected that performance should degrade with reduced contrast, it is not clear how it might interact with target resolution and range. For example, does increasing resolution affect low-contrast targets more than high-contrast targets?

When a target is not detectable with the naked eye, a soldier will move the weapon scope (which typically provides a magnified image) across the visual scene to detect a target. For our experimental setup, we had the observer scan an electronic image with the naked eye. Scanning with the eye will likely be faster than with a sight, and some targets may have been detected peripherally. However, given the low resolution levels (and small target sizes) we used, we think it likely that scanning was largely sequential, as it would be with a weapon scope. We did not examine a magnification capability in these studies given that it would increase the experimental complexity without helping to resolve the question of how resolution could interact with target range and contrast.

At first blush, it would seem desirable to use real-world imagery or even video to determine imagery levels for target acquisition studies. However, there are inherent tradeoffs in the design space. With real-world camera images, it can be difficult to keep factors like lighting, contrast, background movement (e.g., wind in trees), target position, and so on constant from one image to the next. For example, cloud movement means that outdoor illumination varies moment to moment, affecting mean luminance and contrast, and it is difficult to control target position and target characteristics (e.g., clothes moving in wind, posture) with high consistency across images. These factors can be controlled with simulated images. Furthermore, the simulation environment Virtual Battlespace 2 (VBS2) provides reasonably high-fidelity images that have proven useful in similar research (e.g., Banko, Binsch, & Levesque, 2014; Binsch & Kistemaker, 2011; Glaholt, 2014; Ho, Hollands, Tombu, Ueno, & Lamb, 2013; Reiner, Hollands, & Jamieson, 2017; Singer, Barnett, & Taylor, 2010; Tombu, Ueno, & Lamb, 2016; Whitney, Fidock, & Ferguson, 2012). Although static images are artificial, the use of dynamic video transforms target acquisition into a motion detection task, where resolution may not play such a big role. Other studies examining resolution requirements have also used static targets (Harrison, Mullins, & Etienne-Cummings, 2011; Schmieder & Weathersby, 1983; Teaney & Fanning, 2008; Vollmerhausen, Driggers, & Wilson, 2008). For these reasons, we created realistic static images using VBS2 with the virtual camera placed at different distances (ranges) from human targets.

We also thought it important to examine the effects of color on target acquisition. Under daylight conditions, chromatic information is available and can have important effects on both detection and identification. However, some optical systems (e.g., night vision technologies, infrared cameras) provide a monochrome image, so we also examined the detection and identification of targets under monochromatic conditions.

Thus, we sought to determine how target acquisition performance is jointly affected by resolution, contrast, and target range. In two experiments, soldiers performed target detection and target identification tasks with precisely controlled simulated images, and target range, resolution, and contrast were varied. In Experiment 1, participants were shown color images with a single target exemplar. In Experiment 2, monochrome images containing different target exemplars were used.

Experiment 1

For Experiment 1, we chose several resolution levels around the Johnson criteria values for detection (between 2 and 12 ppm). We used the same resolution levels for identification because pilot testing with our stimuli suggested that near-perfect identification performance could be achieved at resolution levels below 32 ppm (the Johnson criteria value for identification). We chose target ranges that were realistic for small arms (50–400 m).

We made the following predictions for both tasks: Accuracy should improve with increased resolution and contrast and decrease with range. For detection, we expected that accuracy levels above and below the Johnson criteria value for detection (approximately 6 ppm) would behave differently. In particular, we expected that increasing range should impair accuracy above but not below the criterion value. If the Johnson criteria value represents a resolution threshold, then there is insufficient resolution to detect the target below the threshold, regardless of range (which affects target size). However, above the threshold value, we should expect to see an effect of range. The contrast sensitivity function (Campbell, Carpenter, & Levinson, 1969; Rovamo, Franssila, & Nasanen, 1992) would predict that as distance increases and the target becomes smaller, target details will occupy higher spatial frequencies and performance will degrade. Moreover, contrast should interact with range because sensitivity for contrast is lower at high spatial frequencies. As simulated distance increases and the target becomes smaller, target details will occupy higher spatial frequencies, where sensitivity for contrast is reduced. We would therefore expect that the performance decrement associated with range (reduced target size) will be greater in the low contrast condition. Similar effects are likely for identification: At higher resolutions, the increased target size should increase the likelihood of distinguishing friend from enemy. We anticipated that contrast effects would be even more likely for identification than detection given the increased need to detect target detail.

Method

Participants

Thirty male soldiers (ages 25–46 years) with shooting experience served as participants. All had normal or corrected-to-normal vision. All were qualified by the Canadian Armed Forces (CAF) to fire small arms. They were reimbursed $25.44 CAD for their participation. For both experiments, the research complied with the Canadian Tri-Council Policy Statement on Ethical Conduct for Research Involving Humans and was approved by the Defence Research and Development Canada Human Research Ethics Committee, Protocol No. 2013-074.

Design and procedure

The experiment had a 2 × 4 × 5 within-subjects design for each task (detection and identification). The independent variables were: contrast (high, low), resolution (2, 4, 8, and 12 ppm), and range (50, 100, 200, 300, and 400 m). Dependent variables were accuracy (proportion correct) and response time (RT).

Each participant read an information sheet describing the experiment and signed a consent form. Participants were seated in a dimly lit room in front of the monitor at what they considered to be a normal viewing distance (approximately 40 to 80 cm) and were asked to maintain that distance throughout the experiment. Participants were instructed not to lean toward the display screen. Participants performed two tasks. In the detection task, participants indicated whether they observed a target (a soldier) by pressing one of two designated keyboard buttons. The target was present on half the trials and absent on the rest. Participants saw an example of the target prior to the experiment. In the identification task, the target was always present, and participants classified the target as a CAF soldier or an OPFOR (opposing force). On half the trials, the target was a CAF soldier; on the other half, it was OPFOR. Examples of each type were shown prior to the experiment.

The trials for each task were run as a block. The order of the tasks was counterbalanced. Participants performed 400 trials in each block (10 trials for each of the 40 conditions). Prior to each task, there was a block of 8 practice trials using different stimuli. Feedback on performance was provided during the practice trials only. To create each block of experimental trials, all stimuli produced by the factorial combination of resolution, range, and contrast were presented in a random order (without repetition).

On each trial, the stimulus image was presented for a 10-second viewing period. The participant was free to move his eyes during the period (a “free viewing” paradigm) and responded by pressing one of two keys. After the viewing period, the image disappeared, and the participant was required to make a response. Participants were instructed to respond as quickly and accurately as possible. After every 25 trials, participants were prompted to take a break. Each trial was initiated by pressing the space bar. Each response was scored as correct or incorrect, and the elapsed time from stimulus onset to key press (RT) was recorded.

The experiment was conducted in two sessions on different days and took about 3 hours to complete (in total). At the conclusion of the experiment, the experimenter thanked and debriefed the participant and answered questions.

Stimulus images

We first describe how we created the 50 original stimulus images for each task, which portrayed targets at various ranges. Then we describe how we manipulated image resolution, our experimental apparatus and setup, and finally our manipulation of contrast to create a total set of 400 color images per task.

For each task, 50 images were generated by taking screenshots of simulated three-dimensional scenes within the VBS2 gaming environment (Bohemia Interactive Simulations, 2011). The pixel count resolution of the original images was 3,840 × 2,160 (horizontal by vertical), equivalent to the monitor’s display resolution. Default VBS2 display/video options were used, with terrain, objects, and texture detail all set to normal. The images were constructed so that the target was not occluded by any other object in the scene, regardless of range. The target height was set to 1.83 m in VBS2.

For the detection task, an urban desert scene was used that included buildings and trees viewed across a field. Five target-present images were produced at each range. The scene was divided vertically into five equally sized regions, and then each target-present image was created by placing the target on the horizon within one of the regions. To control target salience, we placed each target at a location having an approximately uniform background. We also used 5 target-absent images, which were identical to the target-present images except that the target was removed, producing a total of 10 images per range condition. The distance from the target to the viewpoint was manipulated by moving the position of the virtual camera. This produced the 5 range conditions for a total of 50 images (5 ranges × 5 positions × 2 [target present/absent]).

For the identification task, the image showed either an OPFOR target (hostile) or a CAF soldier (friendly). The same hostile or friendly target was used for every trial. A neutral desert scene was chosen for the identification task. This was done to minimize the difficulty of target detection and also reduce possible crowding effects, which are more likely to affect identification than detection (Whitney & Levi, 2011). Both target types carried a rifle and were identical in size, posture, and orientation. The CAF soldier wore the desert CADPAT (CAnadian Disruptive PATtern) uniform, and the OPFOR wore traditional Afghan dress, including a blue-gray tunic and a light-gray head scarf. (This represented a typical appearance for OPFOR insurgents during Canada’s missions in Afghanistan.) As in the detection task, the camera was placed at five different distances from the target and the target placed within each of five regions along the horizon. This produced a total of 50 images (5 ranges × 5 positions × 2 target types).

Table 1 shows the vertical and horizontal target sizes at each range. For the manipulation of range, the visual angle subtended by a target was approximately equal to that subtended at the eye (given a 1.83 m target at the specified range) for a viewing distance of 71 cm (within the approximate range of viewing distances).

Table 1:

Experiment 1: Horizontal and Vertical Target Size as a Function of Target Range

Target Range (m)	Target Width (cm)	Target Height (cm)
50	1.60	2.60
100	0.75	1.27
200	0.42	0.67
300	0.24	0.44
400	0.20	0.35

Resolution manipulation

For each task, Adobe Photoshop CS6 (64-bit version) was used to modify the resolution of the original 50 VBS2 images. The resolution of each entire image was manipulated such that the number of pixels used to render the target height was approximately equal to one of the four resolution levels. The Photoshop mosaic filter was used to create composite pixels by grouping pixels together in square clusters. The actual ppm values deviated slightly from the nominal ppm values. Table 2 shows the actual and nominal ppm values along with the effective pixel count resolution (in horizontal and vertical dimensions) for each condition. There were 200 images per task (the original 50 images × the 4 resolution levels).

Table 2:

Experiment 1: Pixels Per Meter (ppm), Nominal and Actual, Pixels Per 1.83 m Target (ppt), and Horizontal and Vertical Pixel Count Resolution as a Function of Range

Range (m)	Nominal ppm	Actual ppm	ppt	Horizontal Pixels	Vertical Pixels
50	2	2.2	4	102	57
	4	3.8	7	167	94
	8	9.3	17	384	216
	12	12.6	23	549	309
100	2	2.2	4	226	128
	4	3.8	7	350	197
	8	8.7	16	480	270
	12	10.4	16	960	540
200	2	2.2	4	427	240
	4	3.3	6	640	360
	8	6.0	11	1,280	720
	12	9.3	17	1,920	1,080
300	2	2.2	4	640	360
	4	4.9	9	1,280	720
	8	6.6	12	1,920	1,080
	12	13.1	14	3,840	2,160
400	2	2.2	4	960	540
	4	3.3	6	1,280	720
	8	4.9	9	1,920	1,080
	12	10.4	19	3,840	2,160

Apparatus

Image presentation and data collection (including the timing of responses) were controlled by a Visual Basic 6.0 program running on a Hewlett Packard Z420 workstation with Xeon processor (Microsoft Windows 7 operating system). A Python 2.7 script collated and randomized the image sets. Images were rendered on an 80 cm (diagonal), 70 cm × 39.5 cm (horizontal × vertical) 4K ASUS monitor, model PQ321Q. The monitor had ultra-high definition 3,840 × 2,160 resolution with 5.5 dots per mm and a response time of 8 milliseconds. The monitor was set to standard color mode, calibrated to the gamma standard set by the manufacturer’s instructions.

Room illumination and screen reflectance

The experiment took place in a dimly lit room (40 lux). A wall sconce to the participant’s right provided the only room illumination other than the monitor; the sconce was outside the participant’s field of view. Using a Minolta CS-100 photometer, with the monitor off, we measured the light reflected from the monitor surface at five different locations (upper left, upper right, middle, lower left, lower right) under these illumination conditions. The mean luminance was 0.15 cd/m² (s = 0.09).

Contrast manipulation

High- and low-contrast images for each task were produced by adjusting the output levels of each image using Adobe Photoshop. This doubled the number of images to 400 per task. For low-contrast images, the minimum and maximum pixel output levels were 191 and 255, respectively; the values for high-contrast images were 0 and 127. To compute the associated luminance levels, we created nine uniform grayscale images with the following pixel output levels: 0, 31, 63, 95, 127, 159, 191, 223, and 255. Using the photometer, we measured display luminance at the same five locations as previously described for each grayscale image. The mean luminances (averaged across the locations) corresponding to the nine grayscale images were: 0.53, 1.64, 6.17, 19.80, 38.18, 65.14, 99.04, 131.40, and 163.60 cd/m². Based on these values, we computed Michelson contrast values of 0.24 and 0.97 for the low- and high-contrast images, respectively. Our manipulation of pixel output levels meant that luminance was inversely correlated with contrast: The mean image luminance levels were 13.6 and 138.6 cd/m² for high- and low-contrast levels, respectively.

Examples of the detection task stimuli are shown in Figure 1. Figure 2 shows images systematically varying in resolution at a 50 m range with targets marked. Figure 3 shows cropped images of the Figure 2 targets, again varying in resolution. Figures 4, 5, and 6 show corresponding images from the identification task.

Figure 1.

Experiment 1 detection task: Examples of stimulus images. (a) High contrast, 50 m range, at 12 pixels per meter (ppm). (b) Low contrast, 50 m range, at 12 ppm. (c) High contrast, 50 m range, at 4 ppm. (d) High contrast, 400 m range, at 12 ppm. Original images were much larger than shown here.

Figure 2.

Experiment 1 detection task: Examples of target-present stimuli with targets marked. All images are from the high-contrast 50 m condition. ppm = pixels per meter.

Figure 3.

Experiment 1 detection task: Zoomed-in targets from the high-contrast 50 m condition (as shown in Figure 2). (a) 2 ppm. (b) 4 ppm. (c) 8 ppm. (d) 12 ppm. ppm = pixels per meter.

Figure 4.

Experiment 1 identification task: Examples of stimulus images. (a) High contrast, 50 m range, at 12 ppm. (b) Low contrast, 50 m range, at 12 ppm. (c) High contrast, 50 m range, at 4 ppm. (d) High contrast, 400 m range, at 12 ppm. Original images were much larger than shown here. ppm = pixels per meter.

Figure 5.

Experiment 1 identification task: Examples of target-present stimuli with targets marked. All images are from the high-contrast, 50 m condition. (a) 2 ppm. (b) 4 ppm. (c) 8 ppm. (d) 12 ppm. ppm = pixels per meter.

Figure 6.

Experiment 1 identification task: Zoomed-in targets from high-contrast 50 m condition (as shown in Figure 5). Top row, OPFOR target; bottom row, CAF soldier. (a) 2 ppm. (b) 4 ppm. (c) 8 ppm. (d) 12 ppm. OPFOR = opposing forace; CAF = Canadian Armed Forces; ppm = pixels per meter.

Results

Our primary dependent measure was accuracy (proportion of trials in which targets were detected or identified), but we also measured RTs. For each task, each trial was scored as correct or incorrect. A mean accuracy was computed for each participant in each condition, and mean RTs were computed for accurate trials. In general, factors leading to reduced accuracy also increased RTs or had no effect on RTs. For brevity, we only report the exceptions; that is, we report RTs when a speed-accuracy tradeoff occurred.

Mean accuracy data were submitted to a 2 × 4 × 5 within-subjects analysis of variance (ANOVA). Contrast, resolution, and range served as independent variables. All reported effects remained significant after degrees of freedom adjustment under both Greenhouse-Geisser and Huynh-Feldt corrections. For simplicity, we report the uncorrected degrees of freedom.

Detection task: accuracy

Detection accuracy was greater for high contrast images, F(1, 29) = 212.52, MSE = 0.083, η_p² = 0.89, p < .0001. Accuracy increased with target resolution, F(3, 87) = 478.73, MSE = 0.0178, η_p² = 0.94, p < .0001, and generally decreased with increasing range, F(4, 116) = 105.52, MSE = 0.0145, η_p² = 0.78, p < .0001. However, these main effects were subsumed under higher-order interactions. The effect of contrast on accuracy was greater for higher target resolutions, F(3, 87) = 35.42, MSE = 0.0071, η_p² = 0.55, p < .0001, and at longer ranges, F(4, 116) = 13.79, MSE = 0.0094, η_p² = 0.32, p < .0001. The means for these interactions are listed in Tables 3 and 4.

Table 3:

Experiment 1: Mean Detection Accuracy as a Function of Contrast and Resolution

	Resolution
Contrast	2	4	8	12
High	0.50	0.63	0.87	0.90
Low	0.51	0.54	0.77	0.78

Table 4:

Experiment 1: Mean Detection Accuracy as a Function of Contrast and Range

	Range
Contrast	50	100	200	300	400
High	0.76	0.76	0.77	0.73	0.61
Low	0.73	0.73	0.69	0.59	0.51

An interaction between resolution and range showed that accuracy increased with resolution, but the increase was steepest for the shorter ranges, F(12, 348) = 73.76, MSE = 0.0075, η_p² = 0.72, p < .0001. As Figure 7 shows, at 2 ppm, range had little effect, but as resolution increased, the effect of range becomes more evident so that there is a good deal of spread (a difference of over 30%) in mean accuracy as a function of range when the resolution was 8 or 12 ppm.

Figure 7.

Experiment 1: Mean detection accuracy as a function of resolution and range. Error bars indicate 95% within-subject confidence intervals in all graphs (Jarmasz & Hollands, 2009). ppm = pixels per meter.

The resolution by range interaction described previously was moderated by contrast, as shown in Figure 8. For the higher contrast images, accuracy increased with target resolution, but the increase was steepest for the shorter ranges; for the lower contrast images, accuracy did not increase with target resolution at the longest range but did for the other ranges, and again, this increase was steepest for the shorter ranges, F(12, 348) = 5.50, MSE = 0.0075, η_p² = 0.16, p < .0001. The effect of range on accuracy for the higher resolutions was greater for low-contrast images (producing differences in accuracy more than 40%) than high-contrast images. Figure 8 also shows that accuracy in the high-contrast, 4 ppm, 300 m condition was anomalously high.

Figure 8.

Experiment 1: Mean detection accuracy as a function of contrast, resolution, and range.

Detection task: RT

RTs increased with target resolution (a tradeoff with accuracy), F(3, 87) = 24.09, MSE = 4.06 × 10⁶, η_p² = 0.45, p < .0001. An interaction between resolution and range showed that the increase in RT with resolution was limited to the greater target ranges, F(12, 348) = 12.28, MSE = 8.96 × 10⁵, η_p² = 0.30, p < .0001, as shown in Figure 9.

Figure 9.

Experiment 1: Response times in milliseconds (ms) as a function of resolution and range for the detection task.

Identification task: Accuracy

Identification accuracy was greater for high-contrast images, F(1, 29) = 179.54, MSE = 0.019, η_p² = 0.86, p < .0001. Accuracy increased with target resolution, F(3, 87) = 219.17, MSE = 0.027, η_p² = 0.88, p < .0001, and generally decreased with range, F(4, 116) = 34.75, MSE = 0.021, η_p² = 0.55, p < .0001. However, the main effects were subsumed under higher-order interactions. An interaction between contrast and resolution showed that contrast had a greater effect on accuracy with lower target resolutions, F(3, 87) = 3.29, MSE = 0.019, η_p² = 0.10, p < .05 (opposite to the effect observed for detection). An interaction between contrast and range showed that the effect of contrast on accuracy was greater at longer ranges (as seen with detection), F(4, 116) = 18.72, MSE = 0.0129, η_p² = 0.39, p < .0001. The means for these effects are listed in Tables 5 and 6, respectively.

Table 5:

Experiment 1: Mean Identification Accuracy as a Function of Contrast and Resolution

	Resolution
Contrast	2	4	8	12
High	0.67	0.82	0.95	0.96
Low	0.57	0.68	0.84	0.88

Table 6:

Experiment 1: Mean Identification Accuracy as a Function of Contrast and Range

	Range
Contrast	50	100	200	300	400
High	0.86	0.88	0.85	0.88	0.80
Low	0.79	0.84	0.78	0.69	0.63

There was an interaction between resolution and range, shown in Figure 10. Accuracy increased with resolution, but the increase was less steep for the longer ranges, F(12, 348) = 12.53, MSE = 0.0111, η_p² = 0.30, p < .0001. Accuracy increased markedly with range for the higher resolutions (note the spread among the various range conditions on the right side of Figure 10: 8 and 12 ppm), but the effect of range at lower resolutions was reduced and less systematic.

Figure 10.

Experiment 1: Identification accuracy as a function of resolution and range.

There was also a three-way interaction, shown in Figure 11, F(12, 348) = 3.50, MSE = 0.0095, η_p² = 0.11, p < .0001. For the high-contrast images, the increase in accuracy with target resolution was about the same regardless of range, but for the low-contrast images, the increase was steepest (and then reached a ceiling) for the shortest ranges. Figure 11 also shows that for high-resolution targets (on the right of each graph), the mean accuracies were more spread out for low-contrast images; that is, the effect of range was more pronounced with low contrast.

Figure 11.

Experiment 1: Identification accuracy as a function of contrast, resolution, and range.

Discussion

For both tasks, we predicted that accuracy should increase with resolution and contrast and decrease with range. These results were generally supported. Furthermore, we expected that accuracy levels above and below the Johnson criteria value for detection (approximately 6 ppm) would behave differently. In particular, we expected that increasing range should impair accuracy above but not below 6 ppm. These predictions were generally supported for both tasks. Below 6 ppm, there was some variation in accuracy with range for both tasks, but it was much reduced relative to above 6 ppm and not consistent across resolution levels.

For both tasks, accuracy increased with resolution, and the increase was steepest at shorter ranges. That is, increasing resolution had a greater effect on the detection and identification of closer than more distant targets. The steepest part of these increases with resolution occurred across 6 ppm, most noticeably at the shorter ranges. Indeed, performance in both tasks was qualitatively different above and below 6 ppm.

We anticipated that contrast effects would be more likely for identification than detection, given the need to identify target detail. However, there were main effects for contrast in both tasks. Moreover, contrast contributed to three-way interactions in both tasks such that the interactions between resolution and range were more pronounced under low contrast. That is, with high-resolution images at long ranges, the effect of contrast is large (high better than low); at short ranges (50-100 m), the effect of contrast becomes negligible.

Detection task

When the target resolution exceeded 6 ppm, target range (i.e., target size) affected detection accuracy. At 2 ppm, target range had a negligible effect. The low-resolution target could not be reliably detected regardless of its visual angle. Put another way, for resolutions greater than 6 ppm, a more distant target was less detectable than a closer target even though the resolution on target in ppm was roughly equivalent.

The three-way interaction also showed that for the high-contrast images, detection accuracy generally increased with target resolution, with the increase steepest for the shortest ranges. For the low-contrast images, accuracy did not increase with resolution at the longest range but did for the other ranges (and again this increase was steepest for the shortest ranges). Thus, contrast moderated the effect of resolution at different ranges. Increased contrast improved accuracy most in those cases where the target was otherwise hard to detect or identify—being small (due to range) or at low resolution.

The effect of contrast on detection accuracy was greater with higher target resolutions. At the lowest resolution (2 ppm), participants were performing at near-chance levels with high contrast and could not perform worse with low contrast, so it appears that this interaction was produced by a floor effect. The effect of contrast on accuracy was greater at longer ranges (>100 m). Reduced contrast may have made it more difficult to detect target elements given the reduced target size. This result is consistent with the fact that the contrast sensitivity function drops off at high spatial frequencies. Thus, the effect of contrast appears to depend on the absolute size of the target, holding resolution constant.

The increased target detection accuracy at 300 m for 4 ppm appears to have resulted from a display artifact. The actual mean ppm for the stimuli in this condition (4.9 ppm) was higher than the nominal 4 ppm value (see Table 2). The same was true for the corresponding low-contrast condition, but accuracy was much lower in that case. It appears that given sufficient contrast, the near 5 ppm resolution allowed the gap between the legs of the target to become salient (see Figure 12).

Figure 12.

Experiment 1 detection task: Example of stimulus used in the 300 m, 4 ppm (actually 4.9 ppm) condition. High contrast on left, low contrast on right. The combination of slightly higher resolution and high contrast made the gap between the target’s legs more salient on the left. ppm = pixels per meter.

Target detection RTs increased with increasing image resolution. This was a speed-accuracy tradeoff since participants were more accurate with greater resolution. It would appear that participants took more time to maximize detection accuracy. The rate of the RT increase with accuracy was greater for the larger target ranges. A maximization of accuracy strategy will increase RTs more severely at greater ranges given the smaller targets and larger search area. For the distant targets, as resolution improved, so did target detection, but at the cost of longer RTs. For the closer targets, target detection improved with resolution but with less RT penalty. In general, these results are consistent with a model in which participants sought to maximize accuracy by sampling the display over a longer time period.

Identification task

For identification, performance was above chance at low resolutions (accuracies about .6 to .8). (For detection, performance at low resolution levels was near chance, except in one or two anomalous conditions.) Furthermore, higher contrast improved identification performance in these cases (which it generally failed to do for detection). This may have resulted from the use of a single target-distractor pair: The blue-gray color patch in the OPFOR clothing may have been identifiable even at low resolutions. Regardless, range had little effect when identifying targets at low resolutions. Participants were able to identify targets above chance levels below 6 ppm, but range had differential effects above and below 6 ppm.

Identification performance is likely to depend on the specific visual differences between target and distractor (Wolfe, 1997). In our case, the presence or absence of blue-gray color in the OPFOR clothing (which may have persisted even at low resolutions) was noted by several participants. In Experiment 2, we used different target types to examine this issue.

Experiment 2

For both tasks, the Experiment 1 results showed a large improvement in accuracy between 4 and 8 ppm, and that performance was affected by resolution and target range (and therefore target size) in combination. Accuracy decreased with range for resolutions above but not below the Johnson criteria value for detection (approximately 6 ppm). These performance results are consistent with a threshold value between 4 and 8 ppm. The value of 6 ppm is well below the Johnson criteria value for identification. This may have been because the Experiment 1 identification task used only one friend and one enemy target and a unique blue-gray fabric color for the enemy target.

Experiment 2 was a replication of Experiment 1 with the following changes. First, monochrome images were used to eliminate any possible target color artifacts. Second, the specific resolution levels were adjusted. We added a 6 ppm condition given its apparent importance in Experiment 1, and we dropped the 2 ppm condition since it produced performance at near-chance levels. Third, multiple targets were used (5 targets for the detection task; 5 friendly and 5 hostile targets for identification). This was done to be more realistic and reduce the likelihood of a target artifact (e.g., a specific item of clothing) in identification. Fourth, the viewing distance (distance of participant’s eyes from the display surface) was fixed in Experiment 2.

Method

In general, the methods were identical to Experiment 1. We note exceptions in the following.

Participants

Thirty CAF soldiers (29 male; 1 female; ages 21–55 years) served as participants. All had normal or corrected-to-normal vision. None had served in Experiment 1. They were reimbursed $25.44 CAD for their participation. All were qualified by the CAF to fire small arms.

Stimuli and apparatus

For each task, 2,000 monochrome images were generated. There were four resolution levels (4, 6, 8, and 12 ppm). The same five simulated ranges were used as in Experiment 1, and there was a low- and high-contrast condition, producing 5 × 4 × 2 = 40 conditions.

Table 7 shows the vertical and horizontal target sizes at each range. For the manipulation of range, we set the target distances to the particular range values in the simulation environment. Then we computed the visual angle of the 1.83 m target at each distance. The viewing distance that produced the equivalent visual angles on the screen was 62 cm. That is, the visual angle subtended by a target on the display at a 62 cm viewing distance was equal to that subtended at the eye given a 1.83 m target at the specified range.

Table 7:

Experiment 2: Horizontal and Vertical Target Size and Visual Angle as a Function of Target Range

Target Range (m)	Target w (cm)	Visual Angle of w (°)	Target h (cm)	Visual Angle of h (°)
50	1.22	1.13	2.30	2.15
100	0.66	0.61	1.20	1.13
200	0.46	0.43	0.60	0.59
300	0.26	0.24	0.40	0.39
400	0.20	0.19	0.30	0.30

Note. The viewing distance from eye to display was 62 cm. w = width; h = height.

The images were manipulated in Adobe Photoshop to approximate the four resolution levels. As in Experiment 1, the actual ppm values deviated slightly from the nominal ppm values. Table 8 shows the actual and nominal ppm values for each condition. Table 9 reports the visual angle of the target for each range condition, and the pixels per degree (ppd) of target for each condition. Spatial frequency therefore varied from 1.63 to 29.51 cycles per degree (cycles per degree = ppd/2).

Table 8:

Experiment 2: Pixels Per Meter (ppm), Nominal and Actual, Pixels Per 1.83 m Target (ppt), and Horizontal and Vertical Pixel Count Resolution as a Function of Range

Range (m)	Nominal ppm	Actual ppm	ppt	Horizontal Pixels	Vertical Pixels
50	4	3.8	7	167	94
	6	6.0	11	406	229
	8	7.7	14	542	305
	12	11.5	21	677	381
100	4	3.8	7	350	197
	6	5.5	10	480	270
	8	7.7	14	768	432
	12	12.0	22	1,084	610
200	4	4.4	8	640	360
	6	6.6	12	1,219	686
	8	8.2	15	1,626	914
	12	12.0	22	2,438	1,372
300	4	3.8	7	1,084	610
	6	6.0	11	1,761	991
	8	7.7	14	2,167	1,219
	12	11.5	21	3,387	1,905
400	4	3.8	7	1,490	838
	6	6.0	11	2,303	1,295
	8	7.7	14	2,980	1,676
	12	9.8	18	3,840	2,160

Table 9:

Experiment 2: Pixels Per Target (ppt), Target Visual Angle (°), Pixels Per Degree (ppd), and Degrees Per Pixel for Each Range and Resolution Condition

Range (m)	Resolution in ppm (Nominal)	ppt	Target Visual Angle (°)	ppd	Degrees Per Pixel (1/ppd)
50	4	7	2.153	3.25	0.308
	6	11	2.153	5.11	0.196
	8	14	2.153	6.50	0.154
	12	21	2.153	9.75	0.103
100	4	7	1.127	6.21	0.161
	6	10	1.127	8.87	0.113
	8	14	1.127	12.42	0.081
	12	22	1.127	19.51	0.051
200	4	8	0.591	13.53	0.074
	6	12	0.591	20.29	0.049
	8	15	0.591	25.36	0.039
	12	22	0.591	37.20	0.027
300	4	7	0.388	18.04	0.055
	6	11	0.388	28.34	0.035
	8	14	0.388	36.07	0.028
	12	21	0.388	54.11	0.018
400	4	7	0.305	22.95	0.044
	6	11	0.305	36.07	0.028
	8	14	0.305	45.91	0.022
	12	18	0.305	59.02	0.017

Note. The viewing distance was 62 cm.

High- and low-contrast images were produced by adjusting the output levels of each original image using Adobe Photoshop. The minimum and maximum pixel output levels were set to the same values as for Experiment 1, and the Michelson contrast values were equivalent. As in Experiment 1, luminance was inversely correlated with contrast: The mean image luminance levels were 15.3 and 134.3 cd/m² for high- and low-contrast levels, respectively.

For the detection task, there were 5 friendly target exemplars (shown in Figure 13; top), each shown in 5 different locations, for 25 target-present images per condition, or 25 × 40 = 1,000 target-present images. An equal number of target-absent images produced a total of 2,000 images. For the identification task, the same 5 target exemplars were used as friendly targets, and another set of 5 was used as the OPFOR targets (Figure 13; bottom). Each of these 10 exemplars was shown in 5 different locations, producing 50 images for each of the 40 conditions, or 2,000 total images.

Figure 13.

Experiment 2 target exemplars. The exemplars in the top row were used in the detection task. All exemplars were used in the identification task.

Images were rendered on the same display as Experiment 1. The participant’s viewing distance was fixed at 62 cm using a chin rest. The visual angle of the 70 × 39.5 cm display was therefore approximately 58.9° × 34.9° (horizontal × vertical). This meant that 1 display pixel on the monitor subtended approximately 0.015° (0.92 arcminutes). Table 9 shows the degrees per pixel for each of the Experiment 2 resolution and range conditions. The greatest range (400 m) and highest target resolution (12 ppm) required 0.017° per pixel, very close to the resolution limits of the monitor (the difference is rounding error).

Design and procedure

The design was identical to Experiment 1 (2 × 4 × 5 within-subjects for each task, or 40 conditions) except for the changes in resolution levels noted previously. There were also five target locations for each task, as in Experiment 1. Given the five target exemplars, we randomly assigned particular exemplars to the conditions (and target locations) to create five image sets, each containing 400 of the 2,000 total images. To keep the experimental session about the same length as Experiment 1, each participant was shown one of the image sets. That is, each participant saw a subset of the 2,000 images. For a given participant, some combinations of exemplars with conditions and locations were more frequent than others given the random assignment. However, over the five image sets, the total number of each exemplar-location and exemplar-condition pairings was equal. Thus, over a set of five participants, exemplars were assigned to each location and condition equally often.

Results

The responses were scored as in Experiment 1, and the same statistical procedures were used.

Detection task: Accuracy

There was a main effect for resolution: Accuracy increased with resolution from 4 to 6 ppm but did not increase further at greater resolutions (i.e., detection accuracy at 6, 8, and 12 ppm was approximately the same), F(3, 87) = 155.87, MSE = 0.0167, η_p² = 0.84, p < .0001. The mean accuracy values were 0.66, 0.82, 0.85, and 0.86 for 4, 6, 8, and 12 ppm, respectively. There was also a main effect for range, F(4, 116) = 58.42, MSE = 0.0342, η_p² = 0.67, p < .0001, such that accuracy decreased with range from 100 to 300 m. More importantly, there was an interaction between resolution and range, F(12, 348) = 7.93, MSE = 0.0101, η_p² = 0.22, p < .0001, shown in Figure 14. The increase in accuracy with increased resolution was more pronounced at shorter than longer ranges. The interaction also showed that range generally had a greater effect on detection accuracy at higher resolutions.

Figure 14.

Experiment 2: Detection accuracy as a function of resolution and range.

Identification task: Accuracy

There were main effects for resolution, F(3, 87) = 126.64, MSE = 0.0160, η_p² = 0.81, p < .0001, and range, F(4, 116) = 32.68, MSE = 0.0184, η_p² = 0.53, p < .0001, but more importantly, there was an interaction between resolution and range, F(12, 348) = 13.63, MSE = 0.0154, η_p² = 0.32, p < .0001. As Figure 15 shows, accuracy increased with resolution at the shorter ranges, but the increase was reduced at longer ranges, and at the longest range (400 m), the increase was negligible. An examination of the vertical spacing of the means on the right of the graph shows that accuracy decreased markedly with range at the higher resolutions (8 and 12 ppm), but at 6 ppm and below (means on the left of the graph), the effect of range was much reduced. Finally, there was a main effect for contrast, F(1, 29) = 5.56, MSE = 0.0273, η_p² = 0.16, p < .025, such that accuracy was greater at higher contrast (.67 vs. .65, respectively).

Figure 15.

Experiment 2: Identification accuracy as a function of resolution and range.

Discussion

Detection task

In Experiment 2, there was a large increase in detection accuracy as target resolution increased from 4 to 6 ppm, but accuracy did not increase further with greater resolution. The increase was greater at shorter ranges. The results also showed that detection accuracy at 6 ppm and above was a function of range. That is, rather than 6 ppm producing a constant level of accuracy, detection accuracy varied with range, and therefore the visual angle of the target (target size) appears to play a role. This was also true at higher resolutions.

Identification task

For identification, accuracy increased with resolution the most for the shorter ranges. The improvement in accuracy as target resolution increased from 4 to 12 ppm was about 36% at 50 m; at 400 m, the improvement was only 5%, within the margin of error. These results show that at short ranges, high levels of identification accuracy can be produced at resolutions much lower than the Johnson criteria value for identification (32 ppm). However, at longer ranges, when the target is very small, it seems doubtful that one could reach high identification accuracy levels. Even at 12 ppm, our results showed an accuracy of about .62 at 400 m. Put another way, at high resolutions (e.g., 12 ppm), a reduction in range improves accuracy, but at low resolutions (e.g., 4 ppm), changing the range has little effect.

General Discussion

The Johnson criteria are based on resolution thresholds with fixed display parameters. We were interested in how target acquisition performance is jointly influenced by the independent manipulation of resolution, target range (which affects target size), and image contrast. We conducted two experiments in which military participants (soldiers) detected and identified human targets in simulated natural scenes. The resolution and contrast of the images were varied, as was the simulated range of the target. We predicted that accuracy should increase with resolution and contrast and decrease with range. For detection, we predicted that accuracy levels above and below the Johnson criteria value for detection (approximately 6 ppm) would behave differently. Increasing range should impair accuracy above but not below 6 ppm. In general, these predictions were supported. We predicted that the effects of contrast would be more likely for identification than detection given the increased need to perceive target detail. The Experiment 1 results showed effects of contrast for both tasks such that the effects of degraded resolution and increased target range were greater under low contrast conditions. Experiment 2 showed greater identification accuracy with high than low contrast but no effect of contrast for detection.

Experiment 2 differed in several important ways from Experiment 1. These included the use of monochrome images, different resolution levels, varying the targets trial to trial, and fixing the viewing distance. Despite these differences, the results were similar.

Although our methods differed in many respects from those of Johnson (1958), when the Johnson criteria value for detection was expressed in terms of the number of pixels used to depict a meter of target, the resulting number played an important role in our data for both tasks. When resolution was below 6 ppm, performance was not only worse, but the pattern of responding was qualitatively different than when resolution was above the value. Target range matters little below 6 ppm—target detection is near chance and target identification much poorer in these conditions. However, above 6 ppm, factors like range and contrast have a marked effect on performance. In this sense then, the detection threshold value serves to distinguish how range and contrast affect target acquisition performance. With a more distant (and therefore smaller) target, necessary information resides in higher spatial frequencies (cycles per degree). That is, while an observer can use a large range of spatial frequencies to detect or identify a close target, with a more distant target, only the higher spatial frequencies are available. The contrast sensitivity function shows that the visual system is less sensitive to contrast at these higher spatial frequencies.

Through pilot testing, we found that our identification task could be performed at resolutions much lower than 32 ppm (Johnson’s identification threshold value expressed in pixels per meter of target). Performance varied between chance levels and perfect performance across conditions in our target identification task using the same resolution values we used in the detection task (between 2 and 12 ppm). The use of 32 ppm would likely have led to near-perfect performance in all conditions. Identification of target objects can be impeded by clutter and crowding effects (Pelli, 2008; Schmieder & Weathersby, 1983; Whitney & Levi, 2011). In our identification task, the background was uncluttered, but a more cluttered background might have impaired identification performance at lower resolution levels.

Caveats

Experimenters must choose to hold certain factors constant while varying others. In this study, we focused on the number of pixels on the target (target resolution). As noted at the outset, target resolution can be affected by many factors, including those of the sensor. However, we did not consider the characteristics of sensors and imaging systems per se. In follow-on work, we are examining detection and identification performance with real sensor imagery. We did not consider the effects of target motion in this study, identities other than friendly and hostile, or our participants’ confidence in their decisions. There is a need to address such factors in future work. The question of the resolution necessary to detect and identify targets is made more complex when one considers clutter or crowding effects, less than ideal lighting conditions, or the potential for target obscuration given atmospheric conditions.

The luminance of our high-contrast images was lower than our low-contrast images, which means that our manipulation of contrast was confounded with luminance variation. It is possible that changes in luminance led to the observed effects for contrast. While we cannot rule out this possibility, there are several reasons to suggest that the observed results occurred due to contrast rather than luminance change. First, the phenomenon of lightness constancy (Land & McCann, 1971) means that dark and light regions of a display retain those subjective properties even over large luminance shifts, with the net effect that contrast effects are relative rather than absolute (in fact, contrast measures are scaled to mean or total luminance). Second, the visual system adapts quickly to changes in luminance during explorations of a visual scene (Rieke & Rudd, 2009). Third, we would expect to see changes in performance with contrast based on the contrast sensitivity function.

Conclusions

Our objective was to determine how target acquisition is jointly affected by resolution, target range (which affects target size), and contrast. The results showed that the effects of resolution are modulated by target range and contrast. The range effects show that target size affects both detection and identification performance. The fact that similar results were obtained across the experiments suggests that these effects are independent of color and the number of target exemplars. In summary, we showed that target size is an important factor in determining whether an observer can detect and identify targets and that the number of cycles on target as predicted by the Johnson criteria is not the sole determinant of performance. The beneficial effects of increasing target resolution are greater for larger targets.

The fact that increasing target resolution improves target acquisition should come as no surprise. However, our results suggest that increasing target resolution will have greater or lesser effect depending on the target’s visual angle, which can be controlled using a sensor’s optical zoom capability. For instance, the data in Figure 15 would suggest that increasing resolution from 8 to 12 ppm would not improve identification much for a 400 m target (which subtends 0.30° of arc vertically), but if the visual angle of the target was increased to 2.15° using optical zoom (the same as a 50 m target), a substantial improvement in target identification performance would be expected. Statements of requirements for military systems procurement need to consider both zoom and target resolution factors in combination to maximize the probability of target acquisition in use.

Beyond the military context, we believe that the results would likely generalize to other target acquisition situations in which a target is relatively distant and small. This might occur in a variety of public security applications, such as with surveillance imagery collected from an unmanned aerial vehicle or from security cameras used by police forces.

Key Points

Image resolution affects the detection and identification of human targets, but its interaction with target range and contrast is less well understood.

We varied the resolution and contrast of simulated, naturalistic scenes containing targets at different ranges (50-400 m).

The effects of target range and image contrast were qualitatively different above and below 6 pixels per meter of target.

Target detection and identification performance were joint functions of image resolution, range, and contrast for both color and monochrome images.

Footnotes

Acknowledgements

This work was conducted under the Defence Research and Development Canada (DRDC) Future Small Arms Research Project. We thank Dorothy Wojtarowicz for help in coordinating and scheduling participants, Victor Reyes Osorio for generating stimuli, and Michael Tombu, Mackenzie Glaholt, and anonymous reviewers for helpful comments.

Justin G. Hollands is a defense scientist in the Human Effectiveness Section at Defence Research and Development Canada (Toronto Research Centre). He obtained a PhD in psychology from the University of Toronto in 1993.

Phil Terhaar is a lead research technologist at Defence Research and Development Canada (Toronto Research Centre). He received a master’s degree in cognitive science from the University of Guelph in 2008.

Nada J. Pavlovic is a senior research technologist in the Human Effectiveness Section at Defence Research and Development Canada (Toronto Research Centre). She obtained a MASc in human factors and ergonomics from the University of Toronto in 2006.

References

Banko

Binsch

Levesque

(2014). Assessment of a rifle-mounted target hand-off system by professional soldiers with a human-in-the-loop simulation (Joint TNO Report 2014 R11266). Soesterberg, Netherlands: TNO.

Barten

P. G. J.

(2004). Formula for the contrast sensitivity of the human eye. In Miyake

Rasmussen

D. R.

(Eds.), Image quality and system performance, proceedings of SPIE IS&T electronic imaging (Vol. 5294, pp. 231–238). Bellingham, WA: International Society for Optics and Photonics.

Binsch

Kistemaker

J. A.

(2011). “Het is zwaar om commandant te zijn”: Het effect van een gesimuleerde patrouilletaak op de stressrespons van militairen (TNO Report, TNO-DV 2010 IN478). Soesterberg, Netherlands:TNO.

Bohemia Interactive Simulations. (2011). Virtual Battlespace 2 (Version 1.60.92909) [Computer software]. Orlando, FL: Author.

Campbell

F. W.

Carpenter

R. H. S.

Levinson

J. Z.

(1969). Visibility of aperiodic patterns compared with that of sinusoidal gratings. Journal of Physiology, 204, 283–298.

Donohue

(1991). Introductory review of target discrimination criteria (Phillips Laboratory Technical Report, PL-TR-92-2129). Hanscom Air Force Base, MA: Air Force Systems Command.

Glaholt

M. G.

(2014). Automated target cueing during visual search of natural scenes: Performance benefits and costs (DRDC Scientific Report R14-0703-1032). Toronto, Canada: Defence Research and Development Canada.

Harrison

Mullins

Etienne-Cummings

(2011). Sensor and display human factors based design constraints for head mounted and tele-operation systems. Sensors, 11, 1589–1606.

Hollands

J. G.

Tombu

Ueno

Lamb

(2013). Blue force tracking: Effects of spatial error on soldier performance. Proceedings of the Human Factors and Ergonomics Society 57th Annual Meeting (pp. 182–186). Santa Monica, CA: Human Factors and Ergonomics Society.

10.

Jarmasz

Hollands

J. G.

(2009). Using confidence intervals in repeated-measures designs: The number of observations principle. Canadian Journal of Experimental Psychology, 63, 124–138.

11.

Johnson

(1958). Analysis of image forming systems. In Proceedings of the Image Intensifier Symposium (ASTIA document AD220160). Bellingham, WA: International Society for Optics and Photonics.

12.

Land

E. H.

McCann

J. J.

(1971). Lightness and retinex theory. Journal of the Optical Society of America, 61, 1–11.

13.

Maurer

Wilson

D. L.

Driggers

R. G.

(2013). Search and detection modeling of military imaging systems. Optical Engineering, 52(4), 041108-1–041108-11.

14.

Moyer

Devitt

(2005). Resolvable cycle criteria for identifying personnel based on clothing and armament variations. In Holst

G. C.

(Ed.), Infrared imaging systems: Design, analysis, modeling, and testing XVI (Vol. 5784, pp. 60–71). Bellingham, WA: International Society for Optics and Photonics.

15.

Pelli

D. G.

(2008). Crowding: A cortical constraint on object recognition. Current Opinion in Neurobiology, 18, 445–451.

16.

Peterson

(2009). How to calculate image resolution (Technical report). Wilsonville, OR: Theia Technologies.

17.

Ratches

J. A.

Vollmerhausen

R. H.

Driggers

R. G.

(2001). Target acquisition performance modeling of infrared imaging systems: Past, present, and future. IEEE Sensors Journal, 1(1), 31–40.

18.

Reiner

Hollands

J. G.

Jamieson

G. A.

(2017). Target detection and identification performance using an automatic target detection system. Human Factors, 59, 242–258.

19.

Rieke

Rudd

M. E.

(2009). The challenges natural images pose for visual adaptation. Neuron, 64, 605–616.

20.

Rovamo

Franssila

Nasanen

(1992). Contrast sensitivity as a function of spatial frequency, viewing distance, and eccentricity with and without spatial noise. Vision Research, 32, 631–637.

21.

Schmieder

D. E.

Weathersby

M. R.

(1983). Detection performance in clutter with variable resolution. IEEE Transactions on Aerospace and Electronic Systems, AES-19, 622–630.

22.

Singer

M. J.

Barnett

Taylor

(2010). Military training in virtual environments: User interface evaluations. Proceedings of the Human Factors and Ergonomics Society 54th Annual Meeting (pp. 2403–2407). Santa Monica, CA: Human Factors and Ergonomics Society.

23.

Sjaardema

T. A.

Smith

C. S.

Birch

G. C.

(2015). History and evolution of the Johnson criteria (Sandia Report SAND2015-6368). Albuquerque, NM: Sandia National Laboratories.

24.

Teaney

B. P.

Fanning

(2008). Effect of image magnification on target acquisition performance. In Holst

G. C.

(Ed.), Infrared imaging systems: Design, analysis modeling and testing XIX (Vol. 6941). Bellingham, WA: International Society for Optics and Photonics.

25.

Tombu

Ueno

Lamb

(2016). The effects of automatic target cueing reliability on shooting performance in a simulated military environment (DRDC Scientific Report DRDC-RDDC-2016-R036). Toronto, Canada: Defence Research and Development Canada.

26.

Vollmerhausen

R. H.

Driggers

R. G.

Wilson

D. L.

(2008). Predicting range performance of sampled imagers by treating aliased signal as target-dependent noise. Journal of the Optical Society of America, 25(8), 2055–2065.

27.

Vollmerhausen

R. H.

Jacobs

Driggers

R. G.

(2004). New metric for predicting target acquisition performance. Optical Engineering, 43(11), 2806–2818.

28.

Vollmerhausen

R. H.

Jacobs

Hixson

Friedman

(2006). The targeting task performance (TTP) metric: A new model for predicting target acquisition performance (US Army CERDEC Technical Report AMSEL-NV-TR-230). Fort Belvoir, VA: US Army.

29.

Whitney

Levi

D. M.

(2011). Visual crowding: A fundamental limit on conscious perception and object recognition. Trends in Cognitive Sciences, 15, 160–168.

30.

Whitney

S. J.

Fidock

J. T.

Ferguson

(2012). Assessing the effectiveness of simulation-based counter-IED training. Journal of Battlefield Technology, 15(1), 57–64.

31.

Wolfe

J. M.

(1997). Visual search. In Pashler

(Ed.), Attention (pp. 13–73). Hove, UK: Psychology Press/Erlbaum.

32.

Young

S. S.

Driggers

R. G.

Jacobs

E. L.

(2008). Signal processing and performance analysis for imaging systems. Norwood, MA: Artech House.