Facial morphs created between two identities resemble both of the faces used to create the morph. Consequently, humans and machines are prone to mistake morphs made from two identities for either of the faces used to create the morph. This vulnerability has been exploited in "morph attacks" in security scenarios. Here, we asked whether the "other-race effect" (ORE)-the human advantage for identifying own- vs. other-race faces-exacerbates morph attack susceptibility for humans. We also asked whether face-identification performance in a deep convolutional neural network (DCNN) is affected by the race of morphed faces. Caucasian (CA) and East-Asian (EA) participants performed a face-identity matching task on pairs of CA and EA face images in two conditions. In the morph condition, different-identity pairs consisted of an image of identity "A" and a 50/50 morph between images of identity "A" and "B". In the baseline condition, morphs of different identities never appeared. As expected, morphs were identified mistakenly more often than original face images. Of primary interest, morph identification was substantially worse for cross-race faces than for own-race faces. Similar to humans, the DCNN performed more accurately for original face images than for morphed image pairs. Notably, the deep network proved substantially more accurate than humans in both cases. The results point to the possibility that DCNNs might be useful for improving face identification accuracy when morphed faces are presented. They also indicate the significance of the race of a face in morph attack susceptibility in applied settings.

Deep convolutional neural networks (DCNNs) have achieved human-level accuracy in face identification (Phillips et al., 2018), though it is unclear how accurately they discriminate highly-similar faces. Here, humans and a DCNN performed a challenging face-identity matching task that included identical twins. Participants ( = 87) viewed pairs of face images of three types: same-identity, general imposters (different identities from similar demographic groups), and twin imposters (identical twin siblings). The task was to determine whether the pairs showed the same person or different people. Identity comparisons were tested in three viewpoint-disparity conditions: frontal to frontal, frontal to 45° profile, and frontal to 90°profile. Accuracy for discriminating matched-identity pairs from twin-imposter pairs and general-imposter pairs was assessed in each viewpoint-disparity condition. Humans were more accurate for general-imposter pairs than twin-imposter pairs, and accuracy declined with increased viewpoint disparity between the images in a pair. A DCNN trained for face identification (Ranjan et al., 2018) was tested on the same image pairs presented to humans. Machine performance mirrored the pattern of human accuracy, but with performance at or above all humans in all but one condition. Human and machine similarity scores were compared across all image-pair types. This item-level analysis showed that human and machine similarity ratings correlated significantly in six of nine image-pair types [range = 0.38 to = 0.63], suggesting general accord between the perception of face similarity by humans and the DCNN. These findings also contribute to our understanding of DCNN performance for discriminating high-resemblance faces, demonstrate that the DCNN performs at a level at or above humans, and suggest a degree of parity between the features used by humans and the DCNN.

Many tasks in image-guided surgery require a clinician to manually position an instrument in space, with respect to a patient, with five or six degrees of freedom (DOF). Displaying the current and desired pose of the object on a 2D display such as a computer monitor is straightforward. However, providing guidance to accurately and rapidly navigate the object in 5-DOF or 6-DOF is challenging. Guidance is typically accomplished by showing distinct orthogonal viewpoints of the workspace, requiring simultaneous alignment in all views. Although such methods are commonly used, they can be quite unintuitive, and it can take a long time to perform an accurate 5-DOF or 6-DOF alignment task. In this article, we describe a method of visually communicating navigation instructions using translational and rotational arrow cues (TRAC) defined in an object-centric frame, while displaying a single principal view that approximates the human's egocentric view of the physical object. The target pose of the object is provided but typically is used only for the initial gross alignment. During the accurate-alignment stage, the user follows the unambiguous arrow commands. In a series of human-subject studies, we show that the TRAC method outperforms two common orthogonal-view methods-the triplanar display, and a sight-alignment method that closely approximates the Acrobot Navigation System-in terms of time to complete 5-DOF and 6-DOF navigation tasks. We also find that subjects can achieve 1 mm and 1° accuracy using the TRAC method with a median completion time of less than 20 seconds.

Motivated by the need to support those self-managing chronic pain, we report on the development and evaluation of a novel pressure-based tangible user interface (TUI) for the self-report of scalar values representing pain intensity. Our TUI consists of a conductive foam-based, force-sensitive resistor (FSR) covered in a soft rubber with embedded signal conditioning, an ARM Cortex-M0 microprocessor, and Bluetooth Low Energy (BLE). In-lab usability and feasibility studies with 28 participants found that individuals were able to use the device to make reliable reports with four degrees of freedom as well map squeeze pressure to pain level and visual feedback. Building on insights from these studies, we further redesigned the FSR into a wearable device with multiple form factors, including a necklace, bracelet, and keychain. A usability study with an additional 7 participants from our target population, elderly individuals with chronic pain, found high receptivity to the wearable design, which offered a number of participant-valued characteristics (e.g., discreetness) along with other design implications that serve to inform the continued refinement of tangible devices that support pain self-assessment.

In order to better understand how scene motion is perceived in immersive virtual environments, we measured scene-motion thresholds under different conditions across three experiments. Thresholds were measured during quasi-sinusoidal head yaw, single left-to-right or right-to-left head yaw, different phases of head yaw, slow to fast head yaw, scene motion relative to head yaw, and two scene illumination levels. We found that across various conditions 1) thresholds are greater when the scene moves with head yaw (corresponding to gain < 1:0) than when the scene moves against head yaw (corresponding to gain > 1:0), and 2) thresholds increase as head motion increases.

Virtual reality locomotion simulators are a promising tool for evaluating the effectiveness of vision aids to mobility for people with low vision. This study examined two factors to gain insight into the verisimilitude requirements of the test environment: the effects of treadmill walking and the suitability of using controls as surrogate patients. Ten "tunnel vision" patients with retinitis pigmentosa (RP) were tasked with identifying which side of a clearly visible obstacle their heading through the virtual environment would lead them, and were scored both on accuracy and on their distance from the obstacle when they responded. They were tested both while walking on a treadmill and while standing, as they viewed a scene representing progress through a shopping mall. Control subjects, each wearing a head-mounted field restriction to simulate the vision of a paired patient, were also tested. At wide angles of approach, controls and patients performed with a comparably high degree of accuracy, and made their choices at comparable distances from the obstacle. At narrow angles of approach, patients' accuracy increased when walking, while controls' accuracy decreased. When walking, both patients and controls delayed their decisions until closer to the obstacle. We conclude that a head-mounted field restriction is not sufficient for simulating tunnel vision, but that the improved performance observed for walking compared to standing suggests that a walking interface (such as a treadmill) may be essential for eliciting natural perceptually-guided behavior in virtual reality locomotion simulators.

Virtual environments (VEs) allow safe, repeatable, and controlled evaluations of obstacle avoidance and navigation performance of people with visual impairments using visual aids. Proper simulation of mobility in a VE requires an interface, which allows subjects to set their walking pace. Using conventional treadmills, the subject can change their walking speed by pushing the tread with their feet, while leveraging handrails or ropes (self-propelled mode). We developed a feedback-controlled locomotion interface that allows the VE workstation to control the speed of the treadmill, based on the position of the user. The position and speed information is also used to implement automated safety measures, so that the treadmill can be halted in case of erratic behavior. We compared the feedback-controlled mode to the self-propelled mode by using speed-matching tasks (follow a moving object or match the speed of an independently moving scene) to measure the efficacy of each mode in maintaining constant subject position, subject control of the treadmill, and subject pulse rates. Additionally, we measured the perception of speed in the VE on each mode. The feedback-controlled mode required less physical exertion than self-propelled. The average position of subjects on the feedback-controlled treadmill was always within a centimeter of the desired position. There was a smaller standard deviation in subject position when using the self-propelled mode than when using the feedback-controlled mode, but the difference averaged less than six centimeters across all subjects walking at a constant speed. Although all subjects underestimated the speed of an independently moving scene at higher speeds, their estimates were more accurate when using the feedback-controlled treadmill than the self-propelled.