Skip to main content

Rendering algorithms for aberrated human vision simulation


Vision-simulated imagery―the process of generating images that mimic the human visual system―is a valuable tool with a wide spectrum of possible applications, including visual acuity measurements, personalized planning of corrective lenses and surgeries, vision-correcting displays, vision-related hardware development, and extended reality discomfort reduction. A critical property of human vision is that it is imperfect because of the highly influential wavefront aberrations that vary from person to person. This study provides an overview of the existing computational image generation techniques that properly simulate human vision in the presence of wavefront aberrations. These algorithms typically apply ray tracing with a detailed description of the simulated eye or utilize the point-spread function of the eye to perform convolution on the input image. Based on the description of the vision simulation techniques, several of their characteristic features have been evaluated and some potential application areas and research directions have been outlined.


Vision is one of the main mechanisms used to perceive the world. By capturing the light reflected from the surrounding objects, the brain can gather a wide range of information and support a variety of tasks in everyday lives.

The organs that enable vision are the brain and human eye. The human eye acts as an optical system and is responsible for collecting incoming light rays. It is a slightly ovoid organ comprising the cornea and crystalline lens as its main refractive components. The incoming light rays pass through these elements and focus on the back of the eye (the retina), where millions of light-sensitive photoreceptors (the rods and cones) are activated to sense this light. The information thus collected is then forwarded to the brain, which performs various postprocessing tasks to correct the optical limitations of the eye, generating a final image as output, referred to as vision.

The performance of the human eye is heavily affected by a variety of factors despite its robust optical design [1]. Optical limitations of the eye are a typical source of visual imperfection that can be characterized by wavefront or visual aberrations. Wavefront aberrations describe the deviations that the optical system causes in light wave paths as light passes through the refractive elements of the system. Measuring these aberrations using clinical devices (wavefront aberrometers) is a common ophthalmic process in the study of visual aberrations for the characterization of the visual acuity of the eye.

An important and highly unfortunate aspect of visual aberrations is that they affect every living person; this is because the eye’s inherent complexity causes some aberrations even in otherwise healthy vision [2]. In addition, wavefront aberrations induced by external factors have considerable impact on a large portion of the population. Such effects are a natural result of aging and exposure to the outside world, with common causes that include deformation of the cornea, stiffening of the crystalline lens, and changes in the axial length of the eye. Consequently, aberrations are unique to each person and eye condition, as shown in Fig. 1. Thus, vision simulation is invaluable and has many potential applications in ophthalmology [3,4,5,6].

Fig. 1
figure 1

Comparison of three common eye conditions. The visual acuity and the overall characteristics of these eyes differ significantly, which is the main reason why vision-simulated imagery is an important tool of ophthalmology

The problem of simulating the performance of an optical system is not new in the field of computer graphics. A prominent component of plausible visual simulations is the proper use of the depth-of-field (DOF), for which researchers have developed a large set of different rendering algorithms [7]. Most existing vision simulation methods draw inspiration from these techniques and extend them with interesting new ideas to address the complexity and intricacies of the human visual system. This study provides an overview of these techniques, focusing on the optical simulation of human vision in the presence of wavefront aberrations.

Vision simulation using ray tracing

Ray tracing for image synthesis has a long history in computer graphics. It is a technique that utilizes the rays recursively bouncing around in the scene to collect color information through the intersection of these rays with objects and light sources. A large number of ray-tracing approaches have been proposed to address the computational complexity of the problem in an efficient manner. These methods are mainly categorized as forward ray tracing (the rays are traced from the light source to the objects in the scene through the optical system and then captured on the image sensor) and backward ray tracing (the rays are traced outwards from the image plane through the optical system to the objects and light sources in the scene). However, there also exist hybrid algorithms that combine the advantages of these two approaches. Covering the details of these algorithms is beyond the scope of this survey. The reader is referred to ref. [8] for a more elaborate introduction into the inner workings of a ray tracer.

Several studies have utilized ray tracing for human vision simulations. Table 1 provides an overview of the studies discussed in this section. In the following section, the main aspects of vision simulation using ray tracing is outlined.

Table 1 Summary of the methods discussed in this study that utilize ray tracing for vision simulation

Distributed ray tracing

The first method for simulating human vision via ray tracing was introduced by Mostafawy et al. [9]. Their image formation approach utilizes distributed ray tracing [18] combined with analytical surface descriptions to model the various elements of the optical system [19]. The simulated image is constructed by averaging the contributions of a bundle of rays traced through the optical system, which is performed for every output pixel. Each ray is launched toward a different point on the anterior lens surface (the back of the crystalline lens), and thus samples a different point from the physical pupil disk. This process is generally referred to as backward ray tracing and is illustrated in Fig. 2. The quality of the simulation is determined mainly by the number of per-pixel rays traced.

Fig. 2
figure 2

Schematic of distributed ray tracing. Rays starting from the retina are traced through the eye, sampling the back surface of the crystalline lens to determine their outgoing directions. Mostafawy et al. modeled the retina as a flat surface (left), whereas more recent approaches often use a spherical representation (right)

Regarding the description of the eye, the approach of Mostafawy et al. [9] is built on the Gullstrand eye model that uses spherical surfaces to model the refractive elements of the eye. To simulate focusing and low-order aberrations, the model was extended with scaling terms and modifiable curvatures and corneal refractive zones were employed to simulate the impact of photorefractive keratectomy on the subject’s vision.

Spherical retinas

In the approach proposed by Mostafawy et al. [9], the image plane is assumed to be flat, which fails to properly account for curved retinas. Consequently, the peripheral areas of the resulting simulations are significantly over blurred.

Dias et al. [14] circumvented this limitation by employing a spherical retina shape and replacing the Gullstrand model with the wide-angle aspherical Navarro eye model. Figure 2 displays a comparison of the ray-tracing process on two retina shapes.

The final simulation is then obtained by projecting the resulting spherical image onto the image plane using an orthographic projection matrix. Alternatively, Fink and Micol [10] demonstrated that an idealized eye model (i.e., one with minimal aberrations) can be used to unwarp the resulting images.

Efficient sampling

The distributed ray-tracing algorithm described above randomly samples the anterior lens during the construction of ray bundles. Such an approach can potentially lead to a large number of rays being blocked by the pupil, as shown in Fig. 3. Consequently, the resulting image quality can be severely degraded unless a large number of rays are cast for each output pixel.

Fig. 3
figure 3

Comparison of strategies for determining the outgoing ray directions. Sampling the back of the lens leads to many rays being blocked by the pupil (left), whereas every ray can trace through the whole eye unblocked by utilizing pupil sampling instead (right)

To overcome this issue, Wu et al. [20] proposed the use of a ray-tracing method, which is an extension of the bidirectional path-tracing algorithm [21]. They demonstrated the applicability of their approach using multiple schematic eye models, with each comprising spherical and aspherical elements [11].

The gist of their solution is to sample the entrance pupil directly during the ray-generation step, which guarantees that no sample is blocked by the physical pupil. This is called a camera subpath. Rays are also initiated from the light sources of the scene in the light subpath. The algorithm then attempts to connect the various hits along these two subpaths. If an unblocked connection can be made, the starting location of the ray on the entrance pupil is traced through the eye model to find the corresponding image-space projection. Such a sampling strategy can also be combined with distributed ray tracing, as shown in Fig. 3.

Chromatic aberration

Chromatic aberration results from the wavelength-dependent refraction index of the optical elements of the eye, leading to a different focus distance per wavelength of light. Cholewiak et al. [15] studied the impact of chromatic aberration on accommodation and depth perception and concluded that wavelength-related cues are important elements of both processes. They modeled the refractive elements of the eye with a simple finite-aperture lens and simulated the spectral effects using a wavelength-dependent focal length.

A different yet more physically correct approach of simulating chromatic aberration is to assign a different wavelength to each ray and compute the wavelength-dependent refraction index of the simulated optical elements during ray tracing. This approach was adopted in the algorithm of Steinert et al. [22], which was also utilized by Lian et al. [16] in their vision simulation framework called ISET3d. A simplified visualization of the algorithm is shown in Fig. 4. In this approach, spectrum sampling is typically performed uniformly for the three RGB channels or randomly, however, both approaches lead to a significant increase in the number of rays necessary to keep noise levels low.

Fig. 4
figure 4

Overview of spectral ray tracing. Each outgoing ray is assigned a different wavelength, which is used to calculate the wavelength-dependent refraction indices during ray tracing

Diffraction effects

The distributed ray-tracing approach described thus far ignores the wave-related effects. Although the effect of diffraction is generally small with large pupil diameters, the severity of its impact increases as the pupil size decreases. Therefore, not considering these effects can significantly reduce the quality of the simulation.

To simulate diffraction, the Heisenberg uncertainty ray bending (HURB) method [23] was employed by Lian et al. [16] in their ISET3d framework. The HURB method randomizes the direction of the rays as they pass through the pupil. A schematic of this approach is shown in Fig. 5. The probability distribution function is a bivariate Gaussian function, where the variance (angle of scattering) increases toward the edge of the pupil. Such a formulation leads to more blurring when the pupil size decreases because an increasing number of rays originate close to the edge of the pupil. The tradeoff of this approach is an increase in the number of rays and per-ray computation time.

Fig. 5
figure 5

Diffraction simulation using the HURB method. The outgoing ray directions are perturbed using a Gaussian distribution when they pass through the pupil. The amount of scattering depends on the distance from the edge of the pupil

Reducing trace complexity

All the ray-tracing algorithms described above model the eye using a complex system of spherical or aspherical elements. Although such a description has several benefits (such as the ability to support a wide range of eye conditions and chromatic aberration), it also contributes to lengthy computation times; this makes the process of simulating simple eye aberrations unnecessarily complex.

To overcome these issues, Wang and Xiao [12] used a simple eye model comprising a convex lens, pupil, and flat retina. This stripped-down model allows for the derivation of simple geometrical formulas for eyes with myopia, hyperopia, and astigmatism. These formulas are much more efficient than full-blown ray tracing with intersection tests and refractions. The main limitation of this approach is that the supported types of aberrations are very limited and the algorithm ignores all the aspects of vision simulation (chromatic aberration, non-flat retinas, etc.) described above.

Complex surface descriptions

In contrast to the previous case, there are situations where the main concern is quality and not the cost of ray tracing. While the aspheric surfaces and multiple refractive zones in the previous techniques can be used to simulate a wide range of eye conditions, the model is not detailed enough to accurately capture every nuance of a real human eye.

To construct a model that can simulate arbitrary conditions, the use of the circle polynomials of Zernike [24] was proposed by Fink and Micol to describe the various optical elements of the human eye [10]. The Zernike polynomials are a set of orthogonal polynomials over the unit circle and are commonly used in the ophthalmic literature. A least-squares fitting procedure [25] was utilized by the authors to obtain the coefficients necessary for modeling the refractive surfaces and employed backward ray tracing to obtain the final image. To circumvent the complexity of the surfaces thus constructed, the necessary intersection points were calculated using the iterative Newton–Raphson method [26].

The main disadvantage of this approach is the cost of calculating intersection points. The iterative nature of the Newton–Raphson method and the complexity of evaluating surface points, both play an important role in this problem. The use of polygon meshes was proposed by Wei et al. [13] to increase ray-tracing speed. An example of a polygon mesh-based eye model is shown in Fig. 6. Such an approach retains the ability to model arbitrary eye conditions and facilitates the application of modern tools dedicated to ray tracing with polygon meshes. Combined with backward ray tracing, their approach yielded substantial improvements in computational performance.

Fig. 6
figure 6

An example eye model suitable for the algorithm of Wei et al., where each diffracting surface is modeled using triangles. Such a representation facilitates the calculation of intersection points using triangle-based tools instead of the more expensive analytical formulas

Reducing noise

The stochastic nature of distributed ray tracing introduces noise into the resulting simulations. The impact of noise can be reduced by increasing the number of rays traced per pixel; however, this approach causes an inordinate increase in computation times. Denoising [27] is another common form used to drastically improve the quality of renderings with low per-pixel sample counts but requires an extra nontrivial step on top of the existing rendering pipeline.

To overcome this issue, Vu et al. [17] proposed a different method that combines forward ray tracing with triangulation, similar to the lens flare-rendering algorithm of Hullin et al. [28]. An overview of this method is presented in Fig. 7. The algorithm of Vu et al. uses a uniform grid of rays originating from each light source, which was traced toward the entrance pupil of the eye. After the rays successfully trace through the eye and hit the retina, the neighboring rays are used to construct a set of triangles. These triangles are then manually rasterized onto the resulting image to generate vision-simulated imagery. This approach guarantees that the resulting images have no gaps between the ray-traced samples, leading to smooth and noise-free outputs.

Fig. 7
figure 7

Schematic of the forward ray-tracing approach of Vu et al. a An incoming ray bundle is traced through the optical system; b The resulting retinal points are used to form a triangular grid; c The triangles are rasterized onto the output image

Image-space approaches

In addition to ray tracing, another main method for visual aberration simulations is via postprocessing filters that operate on prerendered images. Performance often being a key factor, these approaches are usually employed in rasterization-based systems. Rasterization works by taking a set of input primitives (typically in the form of triangles in 3D space) and determining the list of image pixels that each primitive covers. Per-object transformations and camera projections are handled by transforming the vertices of the input primitives using 4 × 4 transformation matrices. A pinhole camera model is often assumed, with physical camera-based effects simulated via postprocessing. Correct depth ordering is usually resolved using a per-pixel depth buffer, which outputs a depth map, and is often utilized by image-space filters. For a more elaborate explanation of rasterization-based image synthesis, the reader is referred to ref. [29].

Many algorithms operate in the image space to produce vision-simulated images. Table 2 provides an overview of the studies discussed in this section. The following section covers several important aspects of these methods.

Table 2 Summary of the image-space human vision simulation methods discussed in this study

PSF-based convolution

One of the earliest image-space approaches simulating visual aberrations was based on convolution with the PSF of the eye, which is the diffraction pattern of an ideal point source. Given the PSF, the eye can be treated as a black box, and vision can be simulated as the superposition of every object point modulated by the PSF of the optical system. This process is illustrated in Fig. 8.

Fig. 8
figure 8

Vision simulation using PSF-based convolution

Therefore, devising a method for the computation of the PSF is necessary. Even with the goal of simulating vision using convolution, ray tracing can still be used to obtain the PSF. Paraxial ray tracing with a measured corneal topography map was used by Camp et al. [30] to obtain the PSF. In this approach, a single ray originating from a fixed distance along the optical axis is traced against each measured point on the corneal surface. The rays are then refracted according to the local optical power and extended to the image plane. The resulting PSF was then used to perform convolution with a simplified Snellen eye chart.

Although the approach by Camp et al. was an important milestone for convolution-based vision simulation, it has several limitations. Specifically, the simplified eye model removes the inherent aberrations of the eye, which is further exacerbated by the inability of the paraxial approach to correctly simulate wavefront aberrations and diffractions. Greivenkamp et al. [31] demonstrated that the aforementioned problems can be solved using exact ray tracing and proper eye models that comprise aspherical elements. Moreover, the Stiles-Crawford effect [46] was considered in this model by utilizing a Gaussian falloff function centered on the entrance pupil.

Multiple object distances

The two convolution-based approaches described above focus on 2D images. Although the ability to simulate a single object plane holds tremendous value, extending the approach to cover the entire focal region could provide a more holistic image of the individual’s vision. Furthermore, such an extension could also facilitate the simulation of 3D scenes and open the door for creating fully immersive simulations.

Motivated by these reasons, Barsky [33] introduced the concept of object-space point-spread function (OSPSF), which is a depth-dependent version of the traditional PSF. An example of a set of depth-dependent PSFs is shown in Fig. 9.

Fig. 9
figure 9

Through-focus PSFs for six different object distances

Similar to previous works, Barsky used ray tracing to compute the OSPSFs. However, instead of utilizing the eye model directly, Barsky proposed using wavefront aberrations of the eye to calculate perturbed ray directions. In their work, aberration functions are obtained using a Shack-Hartmann wavefront sensor, allowing vision to be simulated without measuring the physical properties of the eye.

Given a set of OSPSFs, vision is simulated by partitioning the input rendering into distinct depth regions and then convolving each slice with the OSPSF corresponding to its central depth value. This approach has a disadvantage—if an object spans several different depth regions, the convolution process may include discarded pixel values along the borders of the depth regions. Barsky referred to this phenomenon as a discretization artifact. To eliminate these artifacts, edge detection can be employed for object identification [37]. By including every object in the scene in all relevant depth layers, it is ensured that no incorrect values are used during convolution.

Faster PSF computation

If the wavefront aberrations of an optical system are known, the PSF can also be computed using diffraction theory. To this end, Watson [38] used the Fraunhofer diffraction formula [47] as a more computationally efficient approach than ray tracing because it facilitates the computation of the PSF via the fast Fourier transform of the complex-valued generalized pupil function.

More recently, Csoba and Kunkli [44] proposed the use of the extended Nijboer-Zernike (ENZ) diffraction theory [48] as an alternative means of computing human PSFs for vision simulation purposes. The ENZ approach is based on the more accurate Debye-Wolf diffraction integral [49] and defines the PSF as a linear combination of a set of independent functions. Such a representation of the PSF not only enables computation with arbitrary precision but also facilitates the reuse of the linearized terms, yielding substantial performance benefits. They also proposed an efficient graphical processing unit (GPU)-based PSF computation method using the ENZ approach [50], which further improves the performance of the PSF computation step and facilitates the interactive computation of the necessary PSF kernels.

Efficient convolution of 3D scenes

All the image-space methods described above used the convolution theorem (convolution in the frequency domain) to produce the final images. Although such an approach works well with 2D inputs or depth-based partitioning, it fails to accurately represent the spatially varying nature of the PSF. Furthermore, processing multiple slices is prohibitively expensive in interactive environments.

To overcome these limitations, Csoba and Kunkli [44] utilized tiled splatting [51] with a GPU-based PSF interpolation technique to simulate visual aberrations at approximately real-time frame rates. Instead of relying on the convolution theorem, their approach performs a convolution directly in the spatial domain. The input image is split into independent tiles, and a set of per-tile buffers is constructed out of all pixels with overlapping PSFs, as shown in Fig. 10. The tile buffers are then sorted by depth for proper occlusion handling. The final image is produced by traversing the sorted per-tile buffers for each output pixel and accumulating the pixel contributions that are weighted by approximations of the dense per-pixel PSF.

Fig. 10
figure 10

Collecting overlapping pixels for the middle tile by the tiled convolution algorithm and a set of example samples. The dashed circles represent the PSFs of the pixels and the pixel colors indicate the inclusion of the pixel in the tile buffer

To produce a dense PSF, their approach builds on Barsky’s concept [37]. The PSFs are computed for a small set of distinct object distances and stored in a GPU-accelerated texture. This texture is then used to approximate the PSF at any object distance via linear interpolation of the neighboring precomputed PSFs. This approach was validated by comparing their results to those obtained using convolution with the true dense PSF, and it was demonstrated that the outputs of their interpolation-based solutions are nearly identical to the ground-truth images.

Chromatic aberration

As described in the subsection on chromatic aberration (Vision simulation using ray tracing), chromatic aberration stems from the wavelength-dependent refractive index of the eye. As a result, the PSF of the eye also depends on the wavelength of light. An example of a set of chromatic PSFs is shown in Fig. 11. Consequently, chromatic aberration can be modeled by employing a wavelength-dependent PSF, which is usually performed in practice using one kernel for each channel of the input RGB images.

Fig. 11
figure 11

Chromatic PSFs for five different wavelengths

The chromatic PSF can be computed in multiple ways. In the case of ray-traced PSFs, the same strategies described in the Vision simulation using ray tracing section, can be employed. The situation is somewhat more complicated with diffraction theory, in which the wavelength-dependent aberration coefficients need to be obtained to compute the corresponding PSFs. Watson [38] and Cholewiak et al. [15, 41] demonstrated that this problem can be solved using empirical formulas to derive the amount of defocus introduced by chromatic aberration.

Alternatively, if the eye structure is available, aberration coefficients can also be obtained using ray tracing and least-squares coefficient fitting, the applicability of which was demonstrated by Csoba and Kunkli [44]. Their approach estimates the physical properties of the simulated eye by fitting the parameters of a customized eye model to the input monochromatic aberration coefficients.

Off-axis aberrations

A common limitation of the image-space convolution methods described above is that they ignore the angle of incidence during the PSF calculation and consider the aberrations (and the corresponding PSFs) to be uniform across the entire visual region. This approach is often referred to as the isoplanatic assumption [52, 53]. While such a simplification has several advantages (e.g., input data reduction, faster PSF computation, faster convolution), it fails to provide a complete view of the eye’s actual visual performance.

To overcome this issue, Rodríguez Celaya et al. [34] used the ray-tracing approach of Barsky to compute a 3D grid of PSFs, with different axes corresponding to the horizontal angle, vertical angle, and object distance. An example of an off-axis PSF grid for a single object distance is shown in Fig. 12. During the simulation, the per-pixel PSF from the grid was approximated via trilinear interpolation, using the neighboring eight PSFs.

Fig. 12
figure 12

Off-axis PSFs computed for the same object distance

The algorithm of Rodríguez Celaya et al. was later extended by Gonzalez Utrera [42] to support chromatic aberration. In addition, a more sophisticated interpolation algorithm was proposed in this study to split the visual field into several 2 × 2 interpolation regions. The corresponding four PSFs of each region were then used to construct a set of basis functions for a more efficient computation of the PSF at an arbitrary point inside the region.

Simple kernel approximations

Similar to ray tracing, the computation time for convolution-based approaches can be reduced by confining the simulation to certain types of aberrations. Such approaches typically employ a simplified kernel that only approximates the true PSF and displays some favorable properties that enable a more efficient convolution.

Repeated filtering

Rokita [32] proposed the use of a simple 3 × 3 kernel that focuses most of the energy on the central pixel. This kernel is then used to repeatedly filter the input image with the number of iterations determined using the per-pixel radius of the blur circle, which is typically referred to as the circle of confusion (CoC).

Uniform elliptical disks

One of the main issues with the repeated-filtering approach is the poor mapping to modern GPUs. This is because current GPUs are severely affected by the speed of memory transactions, and repeated filtering requires a significant number of passes to produce a result for highly defocused images. Consequently, most modern algorithms employ larger kernels that involve only a small number of render passes.

Barbero and Portilla [40] approximated the PSF using a uniform elliptical disk. For each output pixel, the result is produced by taking a fixed set of samples from the neighborhood, based on the corresponding blur ellipse. The parameters of the ellipse (major and minor radii and angle between the major axis and abscissa) are calculated on a per-pixel basis. This process is also performed for each channel of the RGB input, allowing the simulation of chromatic effects.

Gaussian kernels

Although a single-pass filtering approach has significant benefits over several considerably small passes, the requirement for a large number of samples still poses some limitations on the extent of possible blurring at a reasonable speed. In computer graphics, this problem is typically solved by using separable kernels [54,55,56], which are functions that can be computed as the product of lower-rank kernels, to reduce the complexity of the convolution from quadratic to linear.

One of the most well-known separable kernels is the Gaussian kernel, which was used by Tang and Xiao [39] as a substitute for the true PSF of the eye. A comparison of an example PSF and its Gaussian approximation is shown in Fig. 13.

Fig. 13
figure 13

Separable PSF approximation strategies. \(c\) denotes the number of complex phasors used to construct the kernels

The approach of Tang and Xiao utilizes a small neural network to calculate the radii of the blur ellipse for each input pixel, which they refer to as the blur distribution function (BDF). They employ a schematic eye model to build a custom training dataset via ray tracing. The neural network uses the focal length, pupil size, object distance, and angles of incidence as input, and thus, their algorithm naturally supports off-axis aberrations.

Complex phasors

One of the biggest drawbacks of Gaussian kernels is their significant deviation from the true PSF. To overcome this issue, the use of complex phasors has been proposed as a separable filtering approach [57] for DOF simulations. The PSF is approximated as a linear combination of a set of complex-valued basis functions with the number of terms included being application-dependent; 1–3 components are fairly typical, as shown in Fig. 13 for a representative PSF. Although the computational cost of complex phasors is higher than that of the simple Gaussian-filtering approach, complex phasors can produce outputs that much better approximate convolution with the true PSF.

Complex phasors can also be used for vision simulation, as demonstrated by Csoba and Kunkli [43]. Their approach computes the PSF of the simulated eye using the Fraunhofer diffraction formula and fits an ellipse to the resulting PSF. Vision is then simulated using a stretched and rotated version of the original complex kernel, which enables the simulation of low-order visual aberrations and achieves a higher accuracy than Gaussian kernels.

Partial occlusion

Image-based vision simulation algorithms are typically implemented as a postprocessing filter for input images that were originally rendered using pinhole camera models. This approach results in a one-to-one mapping between object space and image space, which fails to properly account for the partially covered object points that are invisible on the pinhole image. The problem is visualized in Fig. 14.

Fig. 14
figure 14

Overview of partial occlusion. Rays are unable to reach the image plane when a pinhole camera model is used (green). Parts of the occluded object are reachable by considering the physical extent of the lens (orange)

This issue is typically solved by using layered inputs [51, 58] to provide the missing background information. Lima et al. [45] used a similar layered method to simulate low-order aberrations of the eye. This approach places a regularly spaced grid of samples on the pupil for each output pixel. They then utilize a tree data structure to precompute the list of relative pixel locations across the layers that contribute to each unique grid location on the pupil. The final result is produced by traversing the tree for each pupil sample, accumulating individual samples.

Multiview synthesis

A different approach for simulating distributed effects is to place virtual viewpoints on the pupil and render the scene from each viewpoint using rasterization. This technique, often referred to as the accumulation-buffering method, was introduced by Haeberli and Akeley [59]. A schematic of this technique is shown in Fig. 15. A common limitation of this approach is that smooth high-resolution outputs require a prohibitively large number of views to be rendered.

Fig. 15
figure 15

Multiview image synthesis. The physical lens is sampled to construct several different viewpoints. The scene is then rendered from each view and accumulated into a single image

To simulate vision using accumulation buffering, Kakimoto et al. [35] proposed the use of a 3D blur field that describes the amount of blurring produced by the eye at a specific location in the scene for a given sample on the pupil. They demonstrated the applicability of wavefront tracing [35] and conoid tracing [36] for precomputing the blur field. Both approaches attempt to improve the computation speed of traditional ray tracing by considering only the higher-level properties of the induced wavefronts. During the simulations, a unique view is constructed for each pupil sample. These views are rendered by applying an offset to each vertex of the triangular scene, which was determined by sampling the blur field at the world-space location of the vertex. The final vision simulation is then produced by accumulating the different views into a single output image.


In previous sections, the working principles of the most common approaches for simulating aberrant human vision were described. In this section, several of the most important properties, which can guide the selection of the correct approach for a given problem, are discussed and compared. An overview of these properties is provided in Table 3. In the remainder of this section, a detailed discussion on the properties of the algorithms described in this survey is presented.

Table 3 Comparison of several important properties of the selected works discussed in this study

Input data and personalization

One of the key differentiators between ray-tracing and image-space approaches is the required input data. Most image-space approaches rely on PSFs, which are comparatively simple to acquire using a wavefront aberrometer, as demonstrated in several recent studies. Furthermore, wavefront aberrations are typically described using the well-understood and standardized approach of Zernike coefficients [60], which can be obtained from aberration databases or even computed from spectacle lens prescriptions using conversion formulas [61].

Ray-traced vision simulation algorithms, however, require a complete eye model to support arbitrary visual aberrations. Such a description can be obtained by utilizing an existing schematic eye model, but the uses of such models are limited because they are not representative of the individual’s eye structure. It is possible to extend the schematic eye model using additional optical elements to simulate glasses; however, this approach is still not ideal if the true physical characteristics of the simulated eye are important. Taking measurements for all required eye parameters is a viable alternative and is often performed in population studies (such as in ref. [62]). However, this process requires several devices and measurements, reducing the reliability of the approach.

Recently, the use of an optimization-based eye-structure estimation process was proposed by Csoba and Kunkli [44] as an effective means of overcoming these limitations. A custom parametric eye model was used with an optimization procedure to find a set of eye parameters that would approximate the input aberration coefficients suitably. Such an approach facilitates the use of wavefront aberrations as input to the simulation but also provides a usable description of the entire eye structure as the output of the process. Consequently, eye structure estimation preserves the best characteristics of the two approaches, with its main disadvantage being the computation time.

Dynamic eye states

The behavior of the eye with varying pupil sizes and focus distances is also of interest. While exploring the performance of the eye in one state is tremendously useful in itself, aberrations vary greatly as the pupil diameter or state of accommodation changes. We also demonstrate the effects of changing the pupil diameter and focus distance in Fig. 16 on an example scene. Accounting for these dynamic parameters can provide a more holistic view of the overall performance of the eye and is crucial for studying the different scenarios that occur in the everyday life of a patient.

Fig. 16
figure 16

Showcasing the effects of varying pupil sizes (\(D\)) and focus distances (\({d}_{f}\)) on vision

Handling a variable pupil size is trivial for all the approaches presented in this study. First, the pupil diameter is already a part of the model when the physical eye structure is used. Second, methods that rely on the CoC include the diameter in the definition of the CoC. Finally, image-space approaches that use Zernike coefficients to calculate the PSF can use scaling formulas to obtain the corresponding Zernike coefficients [38].

Refocusing, however, is a significantly more difficult problem. During accommodation, the shape of the crystalline lens changes drastically. As a result, even if the full physical structure of the relaxed eye is available, only the exact physical properties of the focused lens can be estimated. This issue is even more severe for algorithms that rely on wavefront aberrations because these algorithms do not possess any information about the shape of the eye’s optical elements. Consequently, the only viable approaches that we are aware of are extra aberration measurements and eye structure estimation (as described in input data and personalization subsection).


HOA have a significant impact on the overall quality of vision [63], as shown in Fig. 17. Several factors affect the HOA of the eye, with corneal deformation and internal optical elements being common causes [64].

Fig. 17
figure 17

Comparing the impact of simulating HOA. Low-order aberrations are the same in both simulations. Omitting HOA produces a substantially different output and thus gives an incorrect representation of the individual’s vision

The complex origin of HOA hinders ray-tracing methods from being able to capture HOA when the full eye structure is not used. Furthermore, the deformation of the cornea requires special treatment even if a complete eye model is used, as a single aspherical element lacks the flexibility to represent the arbitrary deformations of a real cornea.

The issue is less severe for image-space approaches because including HOA in the PSF is easily accomplished using Zernike coefficients. However, the same cannot be said about the various approximation approaches (the Gaussian kernel and complex phasor algorithms), as the kernels that they employ are specifically chosen to approximate low-order aberrations (mostly, defocus and astigmatism).

Chromatic aberration

Another crucial aspect of aberrations is their wavelength-dependent nature. As demonstrated by Cholewiak et al. [15], chromatic aberration is an important factor in the accommodation process. Therefore, simulating chromatic aberration is essential in reducing the vergence-accommodation conflict that typically occurs in virtual reality (VR) approaches. The effect of chromatic aberration is demonstrated in Fig. 18, where vision simulations are compared using monochromatic and trichromatic PSF models.

Fig. 18
figure 18

Demonstrating the impact of chromatic aberration. Although a monochromatic simulation is a good representation of the overall visual acuity, convolution with the wavelength-dependent PSF more accurately simulates the true eye performance

As described in chromatic aberration subsection (in vision simulation section using ray tracing), simulating chromatic aberration is trivial with ray tracing if a full eye model is employed. The ray-tracing process can be extended by introducing a wavelength dimension and sampling, to correctly model the full spectrum. However, even without the full-eye model, it is possible to acquire correct simulations using the simplified ray-tracing approach of Wang and Xiao [12], which accounts for the wavelength-dependent focal length of the eye—as demonstrated by Cholewiak et al. [15].

Regarding image-space approaches, simulating chromatic aberration is equivalent to using a wavelength-dependent PSF kernel. The main problem is the calculation of the necessary Zernike coefficients, which can be solved by using a conversion formula—as demonstrated by Watson [38]. However, kernel-approximation approaches typically ignore chromatic aberration. The understanding of the authors is that this simplification is a result of their performance-focused nature and not an inherent limitation of the approaches because when the different channels of the input image are treated separately it is trivial for these algorithms to handle chromatic aberration.

Peripheral vision

Many existing aberration simulation techniques focus on the central area of the visual field and only attempt to simulate on-axis aberrations of the eye. Such an approach is often referred to as isoplanatic approximation, and the way it works is described by Barsky [33]—when the eye moves toward the object to be focused, it automatically aligns the focused object with the visual axis. However, the information acquired through the off-axis area (which can be simulated using anisoplanatic techniques) still possesses important cues for the human brain, as demonstrated by several studies that focus on the loss of peripheral vision [65, 66]. We also demonstrate the importance of peripheral vision by comparing examples of isoplanatic and anisoplanatic rendering in Fig. 19.

Fig. 19
figure 19

Demonstrating the impact of peripheral vision. The blurriness is the same for the entire visual field when only the on-axis PSF is considered. Correctly modeling the off-axis PSFs provides a more holistic view of the simulated vision

When using physically accurate ray tracing and a complete eye model, the simulation of off-axis vision is a natural result of the rendering approach. Furthermore, the outputs are accurate as well, provided the correct retina shape is used. However, it is impossible to correctly simulate peripheral vision using a simplified ray-tracing model.

To simulate off-axis aberrations using PSF-based convolution, first the off-axis PSFs must be obtained and subsequently, the corresponding aberration descriptions. Similar to refocusing, these aberrations vary heavily from person to person and depend greatly on the physical properties of the eye. Thus, the only way of obtaining these coefficients is through measurements and eye structure estimation. However, even if the off-axis aberrations are known, very few PSF-based algorithms are constructed with these considerations. Among the algorithms described in this paper, per-pixel interpolation of the PSF grid is the only approach that is capable of handling the spatially varying off-axis PSF.

Finally, vision simulation algorithms that are based on PSF approximation can more easily support off-axis vision because such approaches only need to obtain the per-pixel CoC. To this end, the BDF-based approach of Tang and Xiao [39] can be used as an efficient means of computing the CoC values during rendering. However, the BDF is built using ray-traced data, and the physical structure of the eye is required for actual patient data, the acquisition of which can be performed using parameter measurements of the eye structure that are estimated from wavefront aberrations of the eye. Furthermore, as described earlier, these methods are unable to properly capture the exact shape of the PSF, and thus, they are mostly useful only for simulating the degree of peripheral blurring.


Another critical aspect of these algorithms is their overall performance. Vision simulation is typically employed as part of a larger system with a specific end goal, and throughput is often a key aspect of these applications. In particular, interactive eye disease simulation, vergence-accommodation conflict reduction, vision testing, aberration-correcting displays, and machine-learning data generators are all applications in which the total computation time of the vision simulation step is a serious limiting factor.

The cost of producing a single image depends significantly on the complexity of the scene owing to the nature of ray tracing. Although more recent algorithms are empowered with massive computational capabilities of the GPUs [13, 17], their authors still report computation times that are inappropriate for interactive applications, even for simple input scenes and comparatively small output resolutions. Currently, consumer-grade hardware is not sufficiently powerful to offset the inherent cost of ray tracing and the large number of rays necessary for a single output pixel. Nonetheless, ray tracing is typically used only for vision simulation when the importance of output quality outweighs the cost of producing a single image.

Image-space algorithms, however, typically display much better performance characteristics. The main factors that contribute to the ray-tracing cost are the large number of per-pixel rays and the complexity of tracing a single ray through the optical system, which are hidden costs in the pre-computation part of convolution. Therefore, convolution-based approaches tend to be substantially faster than ray tracing. Furthermore, they also have the potential to achieve interactive, or even real-time performance provided GPU-based acceleration is available, as demonstrated by multiple recent approaches [39, 44, 45], making these approaches an ideal choice for low-latency, high-throughput applications.

Scene configurations

Finally, the possible limitations of vision simulation algorithms are discussed with regard to the types of supported input scenes. Owing to the continued efforts of computer graphics and ophthalmology researchers, ray-tracing and convolution-based vision simulation algorithms are now readily available for both 2D and 3D input scenes. The main factor to consider is the performance impact of the input type, as described in the previous section.

The handling of partial occlusions, however, is of particular interest. As the name of the phenomenon suggests, partial occlusion is only relevant to 3D scenes and materializes when a blurry foreground object partially covers the background, as described in partial occlusion subsection. Handling partial occlusion is trivial in using ray tracing because the resulting image is constructed by sampling the full pupil; therefore, with a sufficiently large number of samples, all the relevant parts of the scene contribute to the final value of each pixel.

However, image-space approaches typically struggle with partial occlusion because the pinhole camera model that is traditionally used to render the input only considers rays that pass through the center of the pupil. As described in the partial occlusion subsection, a commonly employed solution to this issue is the use of layered input images to supply the necessary information about the missing background pixels. However, the disadvantage of this approach is the increased cost of rendering and processing extra layers of the input.


As mentioned in the introduction, vision simulation algorithms have numerous potential applications [3,4,5,6]. In this section, a short introduction on the areas that would benefit from the proper simulation of human visual aberrations is provided.

Visual acuity metrics and vision testing

One of the earliest uses of visual aberration simulations was the calculation of visual acuity metrics. Such methods utilize computer simulations to generate images corresponding to the vision of an individual. The simulation of visual aberrations is key in these works [31], which is often extended by various neural-processing functions to obtain a more holistic view of the overall acuity of the observer [67, 68]. The resulting simulations are then analyzed using image metrics, such as modulation and Strehl ratio, to evaluate acuity.

Despite the existence of physical devices that carry out such tasks, as noted by Kordek et al. [69], computer-based acuity simulations provide some important advantages over clinical approaches. Simulations are not hampered by optical setup limitations, requiring only a comparatively inexpensive desktop computer, reducing the impact of eyelid squinting.

Study of the human eye

Computer simulations of human visual aberrations can also be beneficial for studying the mechanisms of the human eye. A typical approach in these types of experiments is to construct computer models and derive metrics based on specific image characteristics, visual acuity, or aberrations, using a digital model. In addition, the results can be compared to data obtained from real-world measurements performed on the human eye, which are tremendously useful for testing the validity of hypotheses about the human eye. For example, Tabernero et al. [70, 71] used a similar approach to study the mechanisms of aberration compensation in the eye. The aberrations of digital eye models were computed using data acquired through aberrometer measurements. The use of computer-based ray-tracing tools is essential for obtaining these results.

Most of the existing studies are limited to analyzing only the aberration structure of the eye, whereas utilizing vision-simulated imagery can lead to important new theoretical results on the mechanisms of human vision, providing a deeper understanding of the human visual system and its deficiencies. For example, Cheng et al. [72] used a similar approach to study the impact of HOA on low-order refractive errors. Through-focus vision simulations were used in the presence of visual aberrations to calculate several acuity metrics. The computation of vision-simulated imagery was essential in facilitating this study.

Another area where vision-simulated imagery can be highly useful is the visualization of eye diseases in virtual environments. These experiments typically utilize vision simulations to determine an individual’s ability to perform certain tasks in the presence of reduced vision. Using this approach, experts can gain a better understanding of the impact of such conditions on the everyday lives of affected individuals, and thereby facilitate the development of proper tools and devices that better support the lives of these people. To this end, synthetic [61, 73,74,75,76,77], VR [78,79,80], and augmented reality (AR) [80,81,82] environments have already been successfully used.

Surgery and lens planning

Simulating visual aberrations of the eye can be tremendously important for invasive processes, such as intraocular lens implants [83] and laser surgery [84]. These procedures have a substantial impact on eye performance and often introduce undesired visual aberrations. Vision simulations can reduce the likelihood of post-surgery patient dissatisfaction by facilitating the inclusion of patient-specific information during the planning process. Instead of relying only on numerical metrics, the use of vision-simulated imagery of an individual’s pre- and post-surgical vision can be greatly beneficial.

Another common form of vision correction is the use of progressive addition lenses (PALs). Corrective spectacle lenses are widely used to treat presbyopia (the gradual loss of focusing ability on near objects), which is a common age-related condition. The main differentiating property of these lenses from traditional spectacle lenses is the continuous change in optical power over their surfaces, which significantly increases the complexity of designing a lens that properly treats the vision of the individual. Vision simulation methods are a quick and inexpensive way to evaluate the performance of various PAL designs in virtual environments [34, 42, 85,86,87]. The success of the procedure can be further improved if the knowledge of the user’s visual aberrations can be included in the simulation. Utilizing vision-rendering techniques can significantly improve the fit of the resulting progressive lenses and provide better user satisfaction.

Vision-correcting devices

Non-invasive vision-correcting devices have recently made significant progress, and vision simulation is also a crucial element for their development. Among the earliest devices to be developed, vision-correcting displays typically apply inverse blurring to the displayed imagery to compensate for the aberrations of the observer of the display [88,89,90].

Image-prefiltering approaches often suffer from contrast issues and ringing artifacts. One way of addressing these problems is by utilizing multilayer displays [91, 92] such that multiple semi-transparent layers are stacked on top of one another, with each displaying a different prefiltered image. The key aspect of this approach is selecting the filtering to maximize the contrast of the combined resulting image.

Light-field displays have also been proposed as an alternative to overcome the limitations of single-image solutions, whereby a microlens array is placed on top of the display to provide a higher degree of control over the directionality of the emitted light. Vision correction can be achieved using such displays [93,94,95] by utilizing vision simulations to prefilter the displayed images.

With recent advancements and increased accessibility of optical see-through head-mounted displays (OST-HMDs), vision correction using these devices has become a potential application area [96]. A camera attached to the OST-HMD is used to capture scene information. This image is then filtered using vision simulation methods to obtain an ideal image that an emmetropic eye would see. In the last step, the resulting image is further processed using aberrations of the wearer’s eye to display an image on the virtual screen that is aberration-free.

Recently, holographic near-eye display systems [97] have been used for vision correction. These devices operate by placing in front of the viewer a holographic display that modulates the amplitude and phase of the incoming light. Vision simulation methods are crucial for computing the modulations necessary to correct an individual’s aberrations. However, they can also be used for device calibration [98].

Reducing discomfort from wearing extended reality headsets

XR) displays are also a topic of active research and are heavily influenced by the human visual system. A typical problem with such systems is the inconvenience and fatigue experienced by the user, often because of the disparity between the displayed synthetic imagery and the eye’s natural vision. Enhancing simulated renderings with cues, such as chromatic aberration [15, 41] and depth-dependent blur [32, 99,100,101], can help alleviate this disparity. These techniques attempt to enhance the presented imagery with aberrations that are natural to the wearer’s vision, using vision simulation and considering the inherent visual aberrations of the user to achieve the best possible results.

Fatigue caused by visual disparity is not the only source of discomfort in these devices. Individuals with aberrant vision are often required to wear corrective glasses while using XR headsets, which severely degrades the overall experience of the user. Through the application of vision simulation, the presented imagery can be preprocessed with the inverse aberrations of the wearer, similar to certain vision-correcting displays. Such an enhancement of head-mounted devices eliminates the need to wear glasses, facilitating their use and further improving the convenience of users [102].

Another common form of inconvenience associated with AR displays is user fatigue. Users are often required to rapidly switch their cognitive attention and eye accommodation between virtual and real information, which can lead to significant fatigue. Recently, Arefin [103] used vision simulation to design a special-purpose font to reduce the fatigue originating from these sources when using AR displays.

Training virtual humans

Another area where vision simulation is highly useful is the training of autonomous virtual humans. To ensure that the behavior of a virtual human closely matches that of a real person, it is essential to accurately mimic all aspects of the human visual system. Because visual aberrations play significant role in vision and affect each individual, incorporating them into the simulation pipeline is necessary to achieve the desired results. Therefore, Nakada et al. [104, 105] utilized a ray-tracing approach for the optical simulation of vision. With the advancements and availability of a wide range of vision simulation approaches, we believe that this area could greatly benefit from future experiments with other types of algorithms.


In this study, several approaches for simulating human vision that are affected by wavefront aberrations have been described. These methods were categorized as either based on ray tracing or image-space convolution, depending on the principal tools employed. Ray-tracing methods utilize a wide range of different physical eye models and distributed ray tracing to produce the final results; whereas, convolution-based techniques employ ray tracing or wavefront aberrations to compute the PSF of the eye, which is then used as the kernel for convolution. Based on a description of the various approaches, we also compared several important characteristics, which can serve as guidelines in selecting the correct algorithm when solving a problem that involves the simulation of human vision aberrations. Finally, we outlined several important application areas where simulation of human vision and visual aberrations play an essential role.

Despite the large number of available techniques, there is still room for further research in this field. In particular, performance is a critical area where improvements are always welcome. With increased throughput, ray-tracing algorithms can become a viable alternative to convolutional approaches in real-time environments, particularly in case of more complex eye models. The significant recent advancements made in the field of dedicated ray-tracing hardware accelerators and neural network-based approaches, such as reconstruction from low sample-count renderings [106], could be useful in achieving these goals. However, the length of precomputation and the cost of including peripheral vision and partial occlusion are areas where further research would be beneficial for image-space methods. Efficient means of computing human PSFs and carrying out convolution, such as in the Laplacian domain [107], are essential in alleviating these issues. Finally, patient-specific information is seldom utilized in existing studies because data acquisition is a severely limiting factor. Additional methods that specifically tailor the simulation to the individual (such as the eye structure estimation approach proposed by Csoba and Kunkli [44]) could considerably improve the personalized usability of vision simulation algorithms.

Visual aberrations play an important role in our everyday lives, and with recent advancements in fields such as aberration-correcting devices, head-mounted displays, and XR devices, perfected vision simulation tools are essential to pave the way for future progress in these areas.

Availability of data and materials

Not applicable.





Heisenberg uncertainty ray bending


Point-spread function


Object-space point-spread function


Extended Nijboer-Zernike


Graphical processing unit


Circle of confusion


Blur distribution function


Higher-order aberrations


Virtual reality


Augmented reality


Progressive addition lens


Optical see-through head-mounted display


Extended reality


  1. Artal P, Benito A, Tabernero J (2006) The human eye is an example of robust optical design. J Vis 6(1):1–7.

    Article  Google Scholar 

  2. Thibos LN, Hong X, Bradley A, Cheng X (2002) Statistical variation of aberration structure and image quality in a normal population of healthy eyes. J Opt Soc Am A 19(12):2329–2348.

    Article  Google Scholar 

  3. Ong CW, Tan MCJ, Lam M, Koh VTC (2021) Applications of extended reality in ophthalmology: systematic review. J Med Internet Res 23(8):e24152.

    Article  Google Scholar 

  4. Li TK, Li CH, Zhang XY, Liang WT, Chen YX, Ye YP et al (2021) Augmented reality in ophthalmology: applications and challenges. Front Med 8:733241.

    Article  Google Scholar 

  5. Aydındoğan G, Kavaklı K, Şahin A, Artal P, Ürey H (2021) Applications of augmented reality in ophthalmology [Invited]. Biomed Opt Express 12(1):511–538.

    Article  Google Scholar 

  6. Iskander M, Ogunsola T, Ramachandran R, McGowan R, Al-Aswad LA (2021) Virtual reality and augmented reality in ophthalmology: a contemporary prospective. Asia Pac J Ophthalmol 10(3):244–252.

    Article  Google Scholar 

  7. Barsky BA, Kosloff TJ (2008) Algorithms for rendering depth of field effects in computer graphics. Paper presented at the 12th WSEAS international conference on computers, World Scientific and Engineering Academy and Society, Heraklion, 23–25 July 2008

  8. Glassner AS (1989) An introduction to ray tracing. Morgan Kaufmann Pub., San Francisco

    MATH  Google Scholar 

  9. Mostafawy S, Kermani O, Lubatschowski H (1997) Virtual eye: retinal image visualization of the human eye. IEEE Comput Graph Appl 17(1):8–12.

    Article  Google Scholar 

  10. Fink W, Micol D (2006) simEye: computer-based simulation of visual perception under various eye defects using Zernike polynomials. J Biomed Opt 11(5):054011.

    Article  Google Scholar 

  11. Wu JZ, Zheng CW, Hu XH, Xu FJ (2011) Realistic simulation of peripheral vision using an aspherical eye model. Paper presented at the 32nd Annual Conference of the European Association for Computer Graphics, The Eurographics Association, Llandudno, 11–15 April 2011.

  12. Wang ZL, Xiao SJ (2013) Simulation of human eye optical system properties and depth of field variation. Int J Mach Learn Comput 3(5):413–418.

    Article  Google Scholar 

  13. Wei Q, Patkar S, Pai DK (2014) Fast ray-tracing of human eye optics on graphics processing units. Comput Methods Programs Biomed 114(3):302–314.

    Article  Google Scholar 

  14. Dias C, Wick M, Rifai K, Wahl S (2016) Peripheral retinal image simulation based on retina shapes. Paper presented at the 37th Annual Conference of the European Association for Computer Graphics, The Eurographics Association, Lisbon, 9–13 May 2016.

  15. Cholewiak SA, Love GD, Srinivasan PP, Ng R, Banks MS (2017) Chromablur: rendering chromatic eye aberration improves accommodation and realism. ACM Trans Graph 36(6):210.

    Article  Google Scholar 

  16. Lian T, MacKenzie KJ, Brainard DH, Cottaris NP, Wandell BA (2019) Ray tracing 3D spectral scenes through human optics models. J Vis 19(12):23.

    Article  Google Scholar 

  17. Vu CT, Stock S, Fan LT, Stork W (2020) Highly parallelized rendering of the retinal image through a computer-simulated human eye for the design of virtual reality head-mounted displays. Paper presented at SPIE Photonics Europe, SPIE, Online only, France, 6–10 April 2020.

  18. Cook RL, Porter T, Carpenter L (1984) Distributed ray tracing. Paper presented at the 11th annual conference on computer graphics and interactive techniques, Association for Computing Machinery, Minneapolis, 23–27 July 1984.

  19. Kolb C, Mitchell D, Hanrahan P (1995) A realistic camera model for computer graphics. Paper presented at the 22nd annual conference on computer graphics and interactive techniques, Association for Computing Machinery, Los Angeles, 6–11 August 1995.

  20. Wu JZ, Zheng CW, Hu XH, Wang Y, Zhang LQ (2010) Realistic rendering of bokeh effect based on optical aberrations. Vis Comput 26(6):555–563.

    Article  Google Scholar 

  21. Lafortune EP, Willems YD (1993) Bi-directional path tracing. In: Abstracts of the third international conference on computational graphics and visualization techniques, Association for Computing Machinery, Alvor, 6–10 December 1993.

  22. Steinert B, Dammertz H, Hanika J, Lensch HPA (2011) General spectral camera lens simulation. Comput Graph Forum 30(6):1643–1654.

    Article  Google Scholar 

  23. Freniere ER, Gregory GG, Hassler RA (1999) Edge diffraction in Monte Carlo ray tracing. Paper presented at SPIE’s International Symposium on Optical Science, Engineering, and Instrumentation, SPIE, Denver, 18–23 July 1999.

  24. Mahajan VN (1994) Zernike circle polynomials and optical aberrations of systems with circular pupils. Appl Opt 33(34):8121–8124.

    Article  Google Scholar 

  25. Malacara-Hernandez D, Carpio-Valadez M, Sanchez-Mondragon JJ (1990) Wavefront fitting with discrete orthogonal polynomials in a unit radius circle. Opt Eng 29(6):672–675.

    Article  Google Scholar 

  26. Ypma TJ (1995) Historical development of the Newton-Raphson method. SIAM Rev 37(4):531–551.

    Article  MathSciNet  MATH  Google Scholar 

  27. Huo YC, Yoon SE (2021) A survey on deep learning-based Monte Carlo denoising. Comput Vis Media 7(2):169–185.

    Article  Google Scholar 

  28. Hullin M, Eisemann E, Seidel HP, Lee S (2011) Physically-based real-time lens flare rendering. ACM Trans Graph 30(4):108.

    Article  Google Scholar 

  29. Marschner S, Shirley P (2015) Fundamentals of computer graphics, 4th edn. CRC Press, Boca Raton, pp 162–173

    MATH  Google Scholar 

  30. Camp JJ, Maguire LJ, Robb RA (1990) An efficient ray tracing algorithm for modeling visual performance from corneal topography. Paper presented at the first conference on visualization in biomedical computing, IEEE, Atlanta, 22–25 May 1990.

  31. Greivenkamp JE, Schwiegerling J, Miller JM, Mellinger MD (1995) Visual acuity modeling using optical raytracing of schematic eyes. Am J Ophthalmol 120(2):227-240.

    Article  Google Scholar 

  32. Rokita P (1996) Generating depth-of-field effects in virtual reality applications. IEEE Comput Graph Appl 16(2):18-21.

    Article  Google Scholar 

  33. Barsky BA (2004) Vision-realistic rendering: simulation of the scanned foveal image from wavefront data of human subjects. Paper presented at the 1st symposium on applied perception in graphics and visualization, Association for Computing Machinery, Los Angeles, 7–8 August 2004.

  34. Rodríguez Celaya JA, Brunet Crosa P, Ezquerra N, Palomar JE (2005) A virtual reality approach to progressive lenses simulation. In: Abstracts of the XV Spanish Computer Graphics Conference, The Eurographics Association, Granada, 13–16 September 2005.

  35. Kakimoto M, Tatsukawa T, Mukai Y, Nishita T (2007) Interactive simulation of the human eye depth of field and its correction by spectacle lenses. Comput Graph Forum 26(3):627–636.

    Article  Google Scholar 

  36. Kakimoto M, Tatsukawa T, Nishita T (2010) An eyeglass simulator using conoid tracing. Comput Graph Forum 29(8):2427–2437.

    Article  Google Scholar 

  37. Barsky BA (2011) Vision-realistic rendering: simulation of the scanned foveal image with elimination of artifacts due to occlusion and discretization. In: Richard P, Braz J (eds) Computer vision, imaging and computer graphics. Theory and applications. International joint conference, VISIGRAPP, Angers, 2010. Communications in computer and information science, vol 229. Springer, Berlin, Heidelberg, pp 3–27.

  38. Watson AB (2015) Computing human optical point spread functions. J Vis 15(2):26.

    Article  Google Scholar 

  39. Tang N, Xiao SJ (2015) Real-time human vision rendering using blur distribution function. Paper presented at the 14th ACM SIGGRAPH international conference on virtual reality continuum and its applications in industry, ACM, Kobe, 30 October–1 November 2015.

  40. Barbero S, Portilla J (2017) Simulating real-world scenes viewed through ophthalmic lenses. J Opt Soc Am A 34(8):1301–1308.

    Article  Google Scholar 

  41. Cholewiak SA, Love GD, Banks MS (2018) Creating correct blur and its effect on accommodation. J Vis 18(9):1.

    Article  Google Scholar 

  42. Gonzalez Utrera D (2018) Metrology and simulation with progressive addition lenses. Dissertation, The University of Arizona

  43. Csoba I, Kunkli R (2018) Real-time rendering of sphero-cylindrical refractive errors of the human eye using separable complex convolution. In: Abstracts of the ninth Hungarian conference on computer graphics and geometry, NJSZT, Budapest, 21–22 March 2018.

  44. Csoba I, Kunkli R (2021) Efficient rendering of ocular wavefront aberrations using tiled point-spread function splatting. Comput Graph Forum 40(6):182–199.

    Article  Google Scholar 

  45. Lima ARC, Medeiros AM, Marques VG, Oliveira MM (2021) Real-time simulation of accommodation and low-order aberrations of the human eye using light-gathering trees. Vis Comput 37(9):2581–2593.

    Article  Google Scholar 

  46. Moon P, Spencer DE (1944) On the Stiles-Crawford effect. J Opt Soc Am 34(6):319–329.

    Article  Google Scholar 

  47. Hecht E (2016) Optics, Global Edition, 5th edn. Pearson Education Limited, pp 465–505

  48. Van Haver S (2010) The extended Nijboer-Zernike diffraction theory and its applications. Dissertation, Delft University of Technology

  49. Wolf E (1959) Electromagnetic diffraction in optical systems-I. An integral representation of the image field. Proc Roy Soc A: Math Phys Sci 253(1274):349–357.

    Article  MathSciNet  MATH  Google Scholar 

  50. Csoba I, Kunkli R (2022) Fast, GPU-based computation of large point-spread function sets for the human eye using the extended Nijboer-Zernike approach. Paper presented at the 2nd conference on information technology and data science, IEEE, Debrecen, 16–18 May 2022.

  51. Franke L, Hofmann N, Stamminger M, Selgrad K (2018) Multi-layer depth of field rendering with tiled splatting. Proc ACM Comput Graph Interact Tech 1(1):6.

    Article  Google Scholar 

  52. Fried DL (1982) Anisoplanatism in adaptive optics. J Opt Soc Am 72(1):52–61.

    Article  Google Scholar 

  53. Bedggood P, Daaboul M, Ashman RA, Smith GG, Metha A (2008) Characteristics of the human isoplanatic patch and implications for adaptive optics retinal imaging. J Biomed Opt 13(2):024008.

    Article  Google Scholar 

  54. Zhou TS, Chen JX, Pullen M (2007) Accurate depth of field simulation in real time. Comput Graph Forum 26(1):15–23.

    Article  Google Scholar 

  55. McGraw T (2015) Fast bokeh effects using low-rank linear filters. Vis Comput 31(5):601–611.

    Article  Google Scholar 

  56. Schuster K, Trettner P, Kobbelt L (2020) High-performance image filters via sparse approximations. Proc ACM Comput Graph Interact Tech 3(2):14.

    Article  Google Scholar 

  57. Garcia K (2017) Circular separable convolution depth of field. Paper presented at the 44th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH 2017, Association for Computing Machinery, Los Angeles, 30 July–3 August 2017.

  58. Lee S, Eisemann E, Seidel HP (2010) Real-time lens blur effects and focus control. ACM Trans Graph 29(4):65.

    Article  Google Scholar 

  59. Haeberli P, Akeley K (1990) The accumulation buffer: hardware support for high-quality rendering. ACM SIGGRAPH Comput Graph 24(4):309–318.

    Article  Google Scholar 

  60. Thibos LN, Applegate RA, Schwiegerling JT, Webb R (2002) Standards for reporting the optical aberrations of eyes. J Refract Surg 18(5):S652–S660.

    Article  Google Scholar 

  61. Krueger ML, Oliveira MM, Kronbauer AL (2016) Personalized visual simulation and objective validation of low-order aberrations of the human eye. Paper presented at the 29th SIBGRAPI conference on graphics, patterns and images, IEEE, Sao Paulo, 4–7 October 2016.

  62. Rozema JJ, Rodriguez P, Navarro R, Tassignon MJ (2016) SyntEyes: a higher-order statistical eye model for healthy eyes. Invest Ophthalmol Vis Sci 57(2):683–691.

    Article  Google Scholar 

  63. Hashemi H, Khabazkhoob M, Jafarzadehpur E, Yekta A, Emamian MH, Shariati M et al (2015) Higher order aberrations in a normal adult population. J Curr Ophthalmol 27(3-4):115–124.

    Article  Google Scholar 

  64. Wang L, Santaella RM, Booth M, Koch DD (2005) Higher-order aberrations from the internal optics of the eye. J Cataract Refract Surg 31(8):1512–1519.

    Article  Google Scholar 

  65. Rosenholtz R (2016) Capabilities and limitations of peripheral vision. Annu Rev Vis Sci 2(1):437–457.

    Article  Google Scholar 

  66. Odden JL, Mihailovic A, Boland MV, Friedman DS, West SK, Ramulu PY (2020) Assessing functional disability in glaucoma: the relative importance of central versus far peripheral visual fields. Invest Ophthalmol Vis Sci 61(13):23.

    Article  Google Scholar 

  67. Watson AB, Ahumada AJ Jr (2008) Predicting visual acuity from wavefront aberrations. J Vis 8(4):17.

    Article  Google Scholar 

  68. Fülep C, Kovács I, Kránitz K, Erdei G (2019) Simulation of visual acuity by personalizable neuro-physiological model of the human eye. Sci Rep 9(1):7805.

    Article  Google Scholar 

  69. Kordek D, Young LK, Kremláček J (2021) Comparison between optical and digital blur using near visual acuity. Sci Rep 11(1):3437.

    Article  Google Scholar 

  70. Tabernero J, Benito A, Alcón E, Artal P (2007) Mechanism of compensation of aberrations in the human eye. J Opt Soc Am A 24(10):3274–3283.

    Article  Google Scholar 

  71. Tabernero J, Berrio E, Artal P (2011) Modeling the mechanism of compensation of aberrations in the human eye for accommodation and aging. J Opt Soc Am A 28(9):1889–1895.

    Article  Google Scholar 

  72. Cheng X, Bradley A, Thibos LN (2004) Predicting subjective judgment of best focus with objective image quality metrics. J Vis 4(4):310–321.

    Article  Google Scholar 

  73. Nießner M, Kuhnert N, Selgrad K, Stamminger M, Michelson G (2013) Real-time simulation of human vision using temporal compositing with CUDA on the GPU. PARS Parallel-Algorithmen -Rechnerstrukturen und -Systemsoftware 30(1):102–110

    Article  Google Scholar 

  74. Kanazawa K, Nakano Y, Moriya T, Takahashi T (2011) Visual appearance simulation method for exhibited objects considering viewers’ eyesight and lateral inhibition. J Inst Image Electron Eng Japan 40(1):151–158.

    Article  Google Scholar 

  75. Xiong YZ, Lei Q, Calabrèse A, Legge GE (2021) Simulating visibility and reading performance in low vision. Front Neurosci 15:671121.

    Article  Google Scholar 

  76. Bennett M, Quigley A (2011) Creating personalized digital human models of perception for visual analytics. In: Konstan JA, Conejo R, Marzo J, Oliver N (eds) User modeling, adaptation and personalization. 19th International Conference, UMAP, Girona, 2011. Lecture notes in computer science, vol 6787. Springer, Berlin, Heidelberg, pp 25–37.

  77. Tural E, Tural M (2014) Luminance contrast analyses for low vision in a senior living facility: a proposal for an HDR image-based analysis tool. Build Environ 81:20–28.

    Article  Google Scholar 

  78. Jin B, Ai ZM, Rasmussen M (2006) Simulation of eye disease in virtual reality. Paper presented at 2005 IEEE engineering in medicine and biology 27th annual conference, IEEE, Shanghai, 17–18 January 2006.

  79. Krösl K, Elvezio C, Hürbe M, Karst S, Wimmer M, Feiner S (2019) ICthroughVR: illuminating cataracts through virtual reality. Paper presented at 2019 IEEE conference on virtual reality and 3D user interfaces (VR), IEEE, Osaka, 23–27 March 2019.

  80. Krösl K, Elvezio C, Hürbe M, Karst S, Feiner S, Wimmer M (2020) XREye: simulating visual impairments in eye-tracked XR. Paper presented at 2020 IEEE conference on virtual reality and 3D user interfaces abstracts and workshops, IEEE, Atlanta, 22–26 March 2020.

  81. Ates HC, Fiannaca A, Folmer E (2015) Immersive simulation of visual impairments using a wearable see-through display. Paper presented at the ninth international conference on tangible, embedded, and embodied interaction, Association for Computing Machinery, Stanford, 15–19 January 2015.

  82. Krösl K, Elvezio C, Luidolt LR, Hürbe M, Karst S, Feiner S et al (2020) CatARact: simulating cataracts in augmented reality. Paper presented at 2020 IEEE international symposium on mixed and augmented reality, IEEE, Porto de Galinhas, 9–13 November 2020.

  83. Tabernero J, Piers P, Benito A, Redondo M, Artal P (2006) Predicting the optical performance of eyes implanted with IOLs to correct spherical aberration. Invest Ophthalmol Vis Sci 47(10):4651–4658.

    Article  Google Scholar 

  84. Wang W (2020) Intelligent planning for refractive surgeries: a modelling and visualisation-based approach. Dissertation, University of Liverpool.

  85. Loos J, Slusallek P, Seidel HP (1998) Using wavefront tracing for the visualization and optimization of progressive lenses. Comput Graph Forum 17(3):255–265.

    Article  Google Scholar 

  86. Nießner M, Sturm R, Greiner G (2012) Real-time simulation and visualization of human vision through eyeglasses on the GPU. Paper presented at the 11th ACM SIGGRAPH international conference on virtual-reality continuum and its applications in industry, Association for Computing Machinery, Singapore, 2–4 December 2012.

  87. Leube A, Lang L, Kelch G, Wahl S (2021) Prediction of progressive lens performance from neural network simulations. arXiv preprint arXiv: 2103.10842.

  88. Alonso Jr M, Barreto A, Cremades JG, Jacko JA, Adjouadi M (2005) Image pre-compensation to facilitate computer access for users with refractive errors. Behav Inf Technol 24(3):161–173.

    Article  Google Scholar 

  89. Keleş O, Anarim E (2019) Adjustment of digital screens to compensate the eye refractive errors via deconvolution. Paper presented at 2019 ninth international conference on image processing theory, tools and applications, IEEE, Istanbul, 6–9 November 2019.

  90. Zhao JX, Liu L, Zhang J, Wang TH (2021) Contrast enhancement of images on retina by adjusting deconvolved images. Paper presented at the 6th international conference on image, vision and computing, IEEE, Qingdao, 23–25 July 2021.

  91. Huang FC, Lanman D, Barsky BA, Raskar R (2012) Correcting for optical aberrations using multilayer displays. ACM Trans Graph 31(6):185.

    Article  Google Scholar 

  92. Barsky BA, Huang FC, Lanman D, Wetzstein G, Raskar R (2015) Vision correcting displays based on inverse blurring and aberration compensation. In: Agapito L, Bronstein MM, Rother C (eds) Computer vision - ECCV 2014 workshops. ECCV, Zurich, 2014. Lecture notes in computer science, vol 8927. Springer, Berlin, Heidelberg, pp 524–538.

  93. Pamplona VF, Oliveira MM, Aliaga DG, Raskar R (2012) Tailored displays to compensate for visual aberrations. ACM Trans Graph 31(4):81.

    Article  Google Scholar 

  94. Huang FC, Wetzstein G, Barsky BA, Raskar R (2014) Eyeglasses-free display: towards correcting visual aberrations with computational light field displays. ACM Trans Graph 33(4):59.

    Article  Google Scholar 

  95. Holesinger J (2020) Adapting vision correcting displays to 3D. Dissertation, University of California

  96. Itoh Y, Klinker G (2015) Vision enhancement: defocus correction via optical see-through head-mounted displays. Paper presented at the 6th augmented human international conference, Association for Computing Machinery, Singapore, 9–11 March 2015.

  97. Maimone A, Georgiou A, Kollin JS (2017) Holographic near-eye displays for virtual and augmented reality. ACM Trans Graph 36(4):85.

    Article  Google Scholar 

  98. Yamamoto K, Suzuki I, Namikawa K, Sato K, Ochiai Y (2021) Interactive eye aberration correction for holographic near-eye display. Paper presented at the augmented humans conference 2021, Association for Computing Machinery, Rovaniemi, 22–24 February 2021.

  99. Xiao L, Kaplanyan A, Fix A, Chapman M, Lanman D (2018) DeepFocus: learned image synthesis for computational displays. ACM Trans Graph 37(6):200.

    Article  Google Scholar 

  100. Duchowski AT, House DH, Gestring J, Wang RI, Krejtz K, Krejtz I et al (2014) Reducing visual discomfort of 3D stereoscopic displays with gaze-contingent depth-of-field. Paper presented at the ACM symposium on applied perception, Association for Computing Machinery, Vancouver British, 8–9 August 2014.

  101. Mantiuk R, Bazyluk B, Tomaszewska A (2011) Gaze-dependent depth-of-field effect rendering in virtual environments. In: Ma M, Oliveira MF, Pereira JM (eds) Serious games development and applications. Second international conference, SGDA, Lisbon, 2011. Lecture notes in computer science, vol 6944. Springer, Berlin, Heidelberg, pp 1–12.

  102. Xu F, Li DY (2018) Software based visual aberration correction for HMDs. Paper presented at 2018 IEEE conference on virtual reality and 3D user interfaces (VR), IEEE, Tuebingen/Reutlingen, 18–22 March 2018.

  103. Arefin MS (2021) [DC] SharpView AR: enhanced visual acuity for out-of-focus virtual content. Paper presented at 2021 IEEE conference on virtual reality and 3D user interfaces abstracts and workshops, IEEE, Lisbon, 27 March–1 April 2021.

  104. Nakada M, Chen HL, Terzopoulos D (2018) Deep learning of biomimetic visual perception for virtual humans. Paper presented at the 15th ACM symposium on applied perception, Association for Computing Machinery, Vancouver British, 10–11 August 2018.

  105. Nakada M, Chen HL, Lakshmipathy A, Terzopoulos D (2021) Locally-connected, irregular deep neural networks for biomimetic active vision in a simulated human. Paper presented at the 25th international conference on pattern recognition, IEEE, Milan, 10–15 January 2021.

  106. Hou QQ, Li Z, Marshall CS, Panneer S, Liu F (2021) Fast Monte Carlo rendering via multi-resolution sampling. Paper presented at the graphics interface 2021, 28–29 May 2021.

  107. Leimkühler T, Seidel HP, Ritschel T (2018) Laplacian kernel splatting for efficient depth-of-field and motion blur synthesis or reconstruction. ACM Trans Graph 37(4):55.

    Article  Google Scholar 

Download references


We gratefully acknowledge the support of NVIDIA Corporation with the donation of the TITAN Xp GPU, which we used to render all the vision-simulated images shown in this paper.


Not applicable.

Author information

Authors and Affiliations



IC and RK provided conceptualization; IC performed the literature review and prepared the original draft of the manuscript and its figures; IC and RK modified and refined the drafts of the manuscript and figures; RK provided supervision. Both authors read and approved the final manuscript.

Corresponding author

Correspondence to István Csoba.

Ethics declarations

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.



Additional vision-simulated renderings

In this study, additional simulations were performed for several different test scenes and eye conditions. The resulting renderings, displayed in Fig. 20, compare the vision of a healthy eye with those suffering from myopia, astigmatism, and keratoconus, in a variety of scenes. All images were rendered by focusing on the far plane with a 7 mm pupil size. Similar to all vision-simulated renderings presented in this study, these images were generated using PSF-based convolution, utilizing the ENZ diffraction theory to generate the necessary PSFs.

Fig. 20
figure 20

Vision-simulated renderings for four different eye conditions and three different test scene configurations

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Csoba, I., Kunkli, R. Rendering algorithms for aberrated human vision simulation. Vis. Comput. Ind. Biomed. Art 6, 5 (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: