Several ophthalmic diseases could cause visual field defects (VFD), including glaucoma, stroke, and retinitis pigmentosa (10). Visual field defects would bring difficulties in patients' daily life such as driving, crossing the road, reading, and visual searching (11–13). Due to the restricted vision field, the patients with VFD are less sensitive to surrounding dangers, threatening their health severely. In addition, some VFD are caused by brain injury, which cannot be reversed by traditional medicine or surgical treatment (14). Fortunately, the appearance of AR gives a considerable solution that provides a visual aid and improves searching capability.

Image remapping, overlaid window, visual multiplexing, and danger indicator has been developed to provide a visual aid (Figure 2). Sayed et al. proposed customized digital spectacles with the image remapping method (10). The users were asked to measure their visual field at first. Then the images captured by the camera were remapped using resizing and shifting algorithms to adapt to the measured visual field (Figure 2B). A series of prospective case studies have been held to verify the functional field's efficiency (10, 18). However, digital spectacles cannot preserve the user's original vision. Another method implemented an overlaid window to display the overview scene captured by the camera (Figure 2C). The window is overlaid on the actual visual field of the user (15). Unfortunately, this method had an inherent contradiction between the augmented contextual information and local unscreened information since the overlaid window would inevitably block the natural view. Peli proposed visual multiplexing, defined as two or more signals transmitted on the same channel. In this way, the complete information can be used at the receiving end (19, 20). Spatial multiplexing, one of the visual multiplexing methods, has been implemented to aid tunnel vision (16). The minified edge images of the environmental scene were overlaid onto the user's natural vision (Figure 2D). The edge pixels did not block the realistic view since they only occupied a tiny part of a field of view. Another AR system is developed to notice patients with the tunnel vision of surrounding danger (17). The system would track the moving objects in real-time and extract their characteristics to determine their dangerous degree. The circles are superimposed on the edge of the visual field, and the perilous degree determines their color while moving objects assess their position (Figure 2E). Similarly, Ichinose et al. used edge indicator substitute circles to supplement the information in the lost vision of the patients (21).

Despite providing a visual aid for the VFD patient, improving searching capability is another direction. In Zhao's study, CueSee offered five visual cues for users who cannot finish visual search tasks (13). Among them, guidelines and sun rays are the cues designed for users with peripheral vision loss. The former connected the center of the display and the object with a red guideline, and the latter converged the center of the target with eight red guidelines. The location of the object could be indicated in these ways.

Color vision deficiency (CVD), also known as color blindness, is a group of ophthalmic diseases that affect 8% of males globally. Patients have difficulties perceiving and distinguishing specific colors (22, 23). CVD brings obstacles to patients' daily life and restricts their occupations (24). Although CVD cannot be cured by medical treatment right now, the AR system could help users improve their ability to distinguishing colors and even perform close to healthy people (24).

Several commercial AR devices provide prototypes for CVD aiding research (24). Omnicolor and Chroma are applications based on Google Glass, one of the most popular intelligent glasses (25, 26). Popleteev's applications and Chroma glasses are based on Epson's Moverio, another AR device cheaper than Google Glass (27, 28). Besides, Schmitt et al. developed applications on the smartphone to help patients with CVD (29). Other researchers assemble homemade AR systems or modify existing devices (25, 26, 30).

The processing technologies can be classified into two categories: substituting colors and augmenting visual information (Figure 3). Substituting colors is a group of strategies to transfer the target color to another one, including daltonization, highlight, and contrast. The most representative and popular algorithm is daltonization (Figure 3B), which attempts to shift colors to achieve less confusing color combinations for patients with CVD (25–27, 31, 32). Besides, the highlight method refers to highlight the target colors by replacing them with colors that are easier to see [(26, 28, 29); Figure 3C]. In addition, color contrast is a special algorithm for distinguishing a pair of colors (26, 28, 29). The mechanism of the algorithm is to transfer the target pair of colors to another particular pair of colors that are easier to distinguish (Figure 3D). Augmented visual information includes outline the shape of the area of target colors and indicates the target color with texts or icons [(26, 28–30); Figures 3E,F].

Blindness is defined as the best-corrected visual acuity of a patient's better-seeing eye is <20/400 (33). The dominant causes of blindness are age-related diseases, such as age-related macular degeneration (34). Up to now, there is no therapy that can reverse blindness since the diseases disturb the transmission of visual data from the eye to the brain. Blindness brings severe obstacles to the patient's life, especially restricting their mobility. Perceiving the surrounding environment and avoiding obstacles are essential for improving mobility. Therefore, several distance-based vision aid AR systems have been proposed (35–38). These AR systems make use of color perception, light perception, and hearing to convey information. For patients with blindness who still can perceive color vision, two studies use colors to indicate distance (Figures 4A,B). Their system is applied on Microsoft HoloLens, calculating the distances between objects and users based on the video stream (35, 37). Similarly, for patients considered blind but retain light perception, Hick et al. developed an HMD using the brightness of the light-emitting diodes to inform the patients about the distance (36). While the distance between object and user was shortened, the light-emitting diodes would be bright (Figure 4C). For total blindness, the spatialized sound was used to express the distances in Liu's study (38). The intensity of the sound increased as the distance shortened.

Low vision is defined as the best-corrected visual acuity of a patient's better-seeing eye is better than 20/400 and <20/60 (33). Patients with low vision have difficulties in recognizing things such as reading. The applications of AR in low vision aid focus on strengthening the recognizing capability in different strategies, including magnification, edge enhancement, contrast enhancement, object recognition, and text extraction. Magnification is the single most common strategy (39–43). The images after magnification are showed in a window or covered the user's sight (Figures 4D,E). However, both of them are inevitable to reduce the field of view. Therefore, some research adjusts the transparency of the magnified images for the patients to see the real environment and magnified images together (40, 43). Besides, edge enhancement can avoid this trouble, enhancing the edge of the objects while remaining the view of the users [(13, 39, 42, 44); Figure 4F]. Hwang and Peli used the positive and negative Laplacian filters to enhance the edges (44). The former one would highlight the edge with clear surroundings, while the latter one is the opposite. In their AR system, the users can choose one of three levels to enhance the edge according to their situation. In addition, contrast enhancement helps users recognize things by increasing the contrast of the images [(13, 39, 42); Figure 4G]. In Zhao's study, the contrast enhancement methods include maintaining the hues while increasing the contrast and image binarization processing (39). In recent years, with the rapid development of artificial intelligence, especially convolutional neural networks, breakthroughs have been made in image recognition (45). Among them, object or facial recognition technology [(39, 42, 45); Figure 4H] and text extraction [(39, 40, 46); Figure 4I] using optical character recognition have been combined with AR as strategies to improve recognizing capability of the user with low vision. After object or facial recognition and text extraction, the AR systems return with audio feedback and text information.

In contrast to color blindness, nyctalopia has problems with rods rather than cones (47). Therefore, patients with nyctalopia cannot recognize things clearly in a dark environment (Figures 5A,B). AR has offered a new way in the therapy of nyctalopia by brightening the vision in real-time. Hu et al. proposed a night vision enhancement based on see-through glasses (48). The glasses first inverted the dark image, then used the de-hazing algorithms to process the inverted image. After that, the processed image will be resized and calibrated to the real environment (Figure 5C). Another research developed Troyoculus, which used the self-illumination filter to brighten the video streaming and implemented a bright excess filter to prevent bright excess (50). Troyoculus was developed on two prototypes, HMD and smartphone.

Metamorphopsia is a group of macular disorders that cause the patient's sight to twist (51). The common causes of metamorphopsia include age-related macular degeneration, diabetic macular edema, and vitreoretinal interface disorders. Bozzelli developed an AR system that compensated or reduced visual geometric distortions caused by metamorphopsia in real-time, according to the precise measurement (52). The system tested and adjusted its algorithms constantly until the user's vision was corrected. The correction algorithm mapped the video streams onto a polygonal mesh and offset its vertices.

Amblyopia is a disorder that affects the spatial vision of monocular or binocular (53). It is a developmental disease caused by strabismus or anisometropia. Several VR and AR systems are developed to correct amblyopia by providing games that force users to use their lazy eyes more important than healthy eyes (Figures 5D–F). While using the VR systems, users are completely immersed in the virtual environment, which increases the occurrence of accidents. On the contrary, users can see the real environment while using the AR system. Therefore, AR is considered as a better substitution than VR in amblyopia therapy (49).

Ophthalmic surgery and diagnosis are difficult to get started due to their meticulousness. As a result, medical students need to practice a lot to be qualified. However, many students cannot get enough chances to practice because of the lack of cases. The surgical or diagnostic training simulators based on VR or AR systems have been presented to solve this problem. Students can acquire adequate practice through simulators to improve surgical or diagnostic quality. An inevitable defect of the VR simulators is that VR blocks the connection with the real environment. On the contrary, the AR simulators can reserve the real environment while simulating the surgical or diagnostic information. The AR simulators have been implemented in direct or indirect ophthalmoscopy and microsurgery. Schuppe et al. presented EYESI, a training simulator for indirect ophthalmoscopy, in 2009 for the first time (54). EYESI could show the hands of the examiner from the real world while simulating the patient and fundus. Besides, the AR training simulator of indirect ophthalmoscopy has been developed on low-cost mobile platforms (55). The simulator has two prototypes. One is based on the smartphone used as a direct ophthalmoscope, and the other is based on HMD. Although it was helpful for users to learn ophthalmoscopy, most participants thought they could not master this technology. Moreover, Ropelato et al. developed an AR system for training micromanipulation skills, equipped with a simulative training environment and assessing system (56). In this system, Microsoft HoloLens presented a surgical environment that allowed users to stay in touch with the real world, such as the real instruments or the assistants involved in the surgical procedure. Similarly, CatAR was developed for cataract surgical training (57). What had improved was that they updated the display's resolution, which improved the reality of the simulated surgery.

Since ophthalmic diseases might cause severe consequences and significantly impact life quality, prevention is more important than treatment (58). Besides, although the public might know a little about ophthalmology, a healthy person might not understand how badly the patients suffer from ocular diseases (59). As a result, it is necessary to educate and inform the public. The following research has developed several devices to simulate ophthalmic diseases. Ates et al. presented a low-cost simulation named SIMVIZ (60). Two wide-angle cameras catch the real world, and then the filters deal with the video stream in different ways according to different disease modes and finally send it to the user's sight. The filters can simulate macular degeneration, diabetic retinopathy, glaucoma, cataracts, color blindness, and diplopia using different algorithms. However, SIMVIZ provides an immersive simulative environment for users. It has problems of inconvenience, low resolution, and accuracy. Similarly, Jones et al. proposed methods to simulate six vision, including disability glare, blur, metamorphopsia, perceptual filling-in, and color vision deficits (61). These methods are implemented in smartphones and HMD. In subsequent research, Jones proved the effectiveness of these methods (62). Moreover, an eye-tracked AR system was developed to improve the accuracy of simulated cataracts (63). The processing system included the following parts: reducing visual acuity, reducing contrast, applying color shift, simulating dark shadows, and simulating sensitivity to light. In order to find the best parameters, they conducted interactive experiments with cataract patients. The cataract patients had undergone surgery on one eye while the other had not. They were asked to compare the simulation to their cataract view and return the result to adjust the parameters. In the end, several parameters were constantly adjusted to attach the best simulation. Furthermore, simulating the vision of patients with ocular diseases can enlarge the sample size for research (64).