Skip to navigation – Site map

HomeNuméros9DossierThe Lens Flare. Photorealistic Co...

Dossier

The Lens Flare. Photorealistic Computer Graphics between Physics and Economic Viability

Les images de synthèse photoréalistes, entre exigence de réalisme et viabilité économique. Le cas du facteur de flare
Jens Schröter
Translation(s):
Les images de synthèse photoréalistes, entre exigence de réalisme et viabilité économique. Le cas du facteur de flare [fr]

Abstracts

Since roughly the 1970s, the quest for photorealism has been the driving force behind computer graphics. The goal is not only to render scenes in such a way that they look “realistic,” but specifically to render scenes in a way that they look as if they were photographed. The initial motivation for this was to integrate scenes produced with computer graphics into films that were otherwise recorded photographically. Producing images that look like photos also means reproducing some effects typical of photographic image-making processes like different kinds of blur and specific disturbances like lens flares. The simulation of lens flares is the topic of this essay, which, through close analysis of one computer science paper, will show that the processes of simulating photography are always caught between two different and competing goals: on the one hand, there is the idea of simulating a given photographic property—lens flare in the case of this paper— in a way that is physically correct, a process that may need considerable computational resources and time. On the other, there is the economic goal of performing the rendering efficiently and at comparatively low cost.

Top of page

Full text

  • 1 Bruce Gooch and Amy Gooch, Non-Photorealistic Rendering (Natick, MA: AK Peters, 2001), 1. To avoi (...)

1The recent hype around images that look like photos, created with so-called artificial intelligence, obscures the fact that there is another, much older, and (at least until now) much more important tradition of computational techniques for producing images that look like photos: photorealistic computer graphics (hereafter “PRCG”). “The driving force behind computer graphics for the past 35 years has been photorealism. The quality of images created using a computer is judged by how closely they resemble a photograph.”1

2This field has a long history, which dates back to the 1960s and was first driven by military needs—e.g., flight simulation—and later (and now predominantly) by the entertainment industry, especially the film industry, where PRCG is used to produce special effects that fuse seamlessly with otherwise photographic (cinematographic, videographic) recorded film. Today, the computer games industry is also a major actor in the field. PRCG is still the most important field of discourse and practice in which the relationship between “photography” and “algorithms” is negotiated and materialized.

  • 2 Konrad F. Karner, Assessing the Realism of Local and Global Illumination Models, PhD diss. (Vienn (...)
  • 3 See Alan Watt, “Rendering Techniques: Past, Present and Future,” ACM Computing Surveys 28, no. 1 (...)

3PRCG is a very complex and heterogeneous field in many different respects. Sometimes, it is about making “realistic” images, meaning that the pictures are supposed to look like the real world as seen by human eyes (whatever that means exactly), but sometimes (and more often) it is about images that look like photographic (filmed etc.) images of the real world. As a researcher in computer graphics puts it, “The goal of realistic image synthesis is to produce a visual experience in our mind which is ‘identical’ to that of a real scene. […] There is the so-called ‘photorealism’ produced by simulating the illumination using a computer, the physical realism (the actual world), the photographic realism obtained by taking a photograph with a camera, and the perceptual realism stored and produced in our brain.2 Sometimes, creative empirical approximations, workarounds, and simplifications are used that look “good enough,” resembling the real world and/or photographic images of the real world, but sometimes meticulously precise simulations of the physics of light etc. are preferred.3 The decision as to which algorithms are used depends not only on explicit or implicit ideals of “realism” but very often on economic questions relating to computing resources and available time too. Sometimes, accuracy is discarded in light of the assumed expectations and perceptual abilities of a potential audience. Sometimes, PRCG is combined with sampled images or textures. Most often, all these aspects—and others too—interfere in multiple different ways. There is not one way in PRCG to make images that look like photos.

  • 4 Friedrich A. Kittler, “Computer Graphics: A Semi-Technical Introduction,” Grey Room 2 (2001): 30. (...)
  • 5 Bui Tuong Phong, “Illumination for Computer Generated Pictures,” Communications of the ACM 18, no (...)

4Creating PRCG is a piecemeal process, accompanied by aesthetic and economic decisions. In this sense, it is closer to the stepwise process of constructing a painting than to the instant recording of a photographic image (although photographic images are, of course, often processed after the moment of recording). Kittler starts his famous paper on two important algorithmic strategies for global illumination, ray tracing and radiosity, not with a comparison of computer graphics with photography but with the “necessary comparison between paintings and computer images.”4 Or as a famous early paper on an illumination model puts it, “the computer system can be compared to an artist who paints an object from its description and not from direct observation of the object.”5 In that sense, PRCG is comparable to the painterly movement of photorealism (e.g., the work of Chuck Close, Richard Estes, and Ralph Goings), in which images that look like photos are painstakingly constructed.

5At this point, a question emerges: What does it mean to say that an image “looks like a photo”? What are the properties of an image that make it photo-like? This is the question that will be tackled in the first part of this paper. There are many different phenomenological properties that are photo-like or photo-specific, some of which have to be present in a rendered (or painted) image to make the image look like a photo. Each of these properties is the subject of research in computer graphics.

  • 6 Kittler, “Computer Graphics,” 37. A more detailed technical account of ray-tracing and radiosity (...)

6The piecemeal and heterogeneous character of computer graphics is evident in that firstly there is no necessity to use all the photo-specific visualities together—sometimes only some of them are used and others not (so that the resulting images may only be photo-like in some respects and not in others)—and secondly that for each of these properties in PRCG, different algorithmic procedures are available that have different advantages and problems. This last aspect has been discussed by Kittler in relation to ray-tracing and radiosity algorithms: “Two algorithms present themselves as options, but these practically contradict each other and, consequently, govern mutually exclusive aesthetics.”6 Kittler, however, deals only with algorithms for the global illumination of a scene and not with algorithms for producing effects that are photo-specific like lens flares etc. A (more or less) consistent global illumination is also needed for an image to look like a photo, but global illumination alone does not give a photographic look. In the second part of my paper, one of these photo-specific properties is selected—namely, lens flares—and, drawing on a key paper on the subject, an indication is given of how varied the strategies are that are used to construct this photo-like effect. The process invariably oscillates between “realism” and the need for “efficient” solutions.

What does it mean for an image to “look like a photo”?

  • 7 “Diskurse der Fotografie,” Suhrkamp | Insel, accessed January 6, 2024, https://www.suhrkamp.de/bu (...)
  • 8 John Tagg, The Burden of Representation: Essays on Photographies and Histories (Houndmills: Palgr (...)

7It is necessary to define what is meant by “photography” to discuss the question of a specific photographic aesthetic. On the publisher’s website advertising a well-known and important anthology on the theory of photography, it is said that “photography does not exist, but only a multitude of uses of this medium.”7 Or as John Tagg put it, “Photography as such has no identity.”8 Of course, there are many different practices with photography—but does this really mean that photography “does not exist”? If that were the case, then the question “How does a photo look?” would be meaningless too. A statement of the type “There is no such thing as photography, only a multitude of photographic practices” is already logically problematic, because it presupposes a criterion according to which we can select the various “photographic” practices from a wider field of practices. To decide what photographic practices are, we need at least a minimal definition of photography. I defend the view that “photography” does indeed exist and has an identity. Photo-graphy means “writing of/with light.” I define it as a technology that uses a sensor to record the wavelength and amplitude of electromagnetic radiation over a limited period of time and makes this recording available on a display.

8This definition is kept as general as possible. It specifies what is indispensable for photography. It does not say that visible light has to be used: there is also the photographic recording of radiation that is invisible to humans—for example, in X-ray photography. It does not say that there has to be a lens: it also includes the photogram, in which objects are placed directly on the photographic paper, and a shadow instead of an image is recorded. By focusing on wavelength and amplitude, this definition excludes holography; in emphasizing the limited time-space, a boundary line is drawn between the photographic still image and photographic film—whereby drawing this boundary already shows the difficulties of clearly defining different media and distinguishing them one from one another (film can also be photography at a basic level but does not have to be; the time-space of a film or video is also limited; and so on). The general term “sensor” is only intended to describe a technology that records the effects of electromagnetic radiation—regardless of whether this recording is chemical (as in classic photography), analog-electronic (as in video), or digital-quantum-electronic, as is widely used today. The display referred to in the definition can be a screen, a printout of varying materiality and quality, etc.

9This definition emphasizes the recording of certain (not all) information transmitted by given electromagnetic radiation. In other words, a photograph stores information that the light contains owing to its reflection, refraction, etc. of world occurrences (which is why, for example, X-ray images are taken or photos in radar traps). These images are intended to provide information about what is being photographed. Thus, the photographic image is essentially in an indexical relationship with what is depicted. There has been a lot of discussion about this, but in my opinion this aspect cannot be argued away. A photograph is always in some way also a trace of what was in front of the lens (or, in the case of the photogram, on the photosensitive paper). This does not mean that the picture shows what was in front of the camera in any exact or intelligible way: it could be blurred, incorrectly exposed, out of focus, or showing only an unintelligibly small part of an object (these are all, as we will see in a moment, photo-specific forms of imaging). The index does not automatically reveal anything about the meaning of the image, which is mostly determined by the context—e.g., other photos, accompanying text, or a specific situation of use. Moreover, photographic images—be they analog or digital (digital is easier)—can be reworked, falsified, and manipulated in many ways until the trace disappears and the image comes close to being a painting rather than being a photograph. There can, of course, be images—as per the topic of this essay—that look like photos but are, so to speak, synthetically constructed without indexically referring to anything. But in calling a painting or a digitally generated image “photorealistic,” it is already implied that there is a certain photo-like, photo-specific look. Of course, this does not mean that there are no intermedial forms that can be used in photography: there are staged photos, reworked photos, photos that look like paintings, etc. And it also does not mean that the photo-specific elements of a given photo are in themselves more valuable and to be preferred to elements or forms borrowed from other media (as Greenbergian modernism would perhaps have argued). It just means that with the technology of photography, images can be produced that are (mostly) recognizable as photos, although the borders are always blurry. The experiments carried out by painters seeking to produce images that look like photographs or by computer scientists rendering images that look like photographs show that these people know about photo-specific aesthetics.

10What specific aesthetic traits does the medium of photography offer? Everyone knows what a photograph looks like, but it is not so easy to describe what the typical photographic look is. Moreover, it is reasonable to expect that it arises from the fundamental characteristics of the medium described. I will focus on conventional photography with a lens and suggest only a few traits (including indexical detail and perspectival projection), especially those that have attracted attention in research on PRCG:

    • 9 See Wolfgang Ullrich, Die Geschichte der Unschärfe (Berlin: Wagenbach, 2009). In his first chapte (...)

    The optical system can lead to blurring or different levels of sharpness. This did not exist in painting (prior to photography).9

    • 10 See Fernando Navarro et al, “Motion Blur Rendering: State of the Art,” Computer Graphics Forum 30 (...)

    Depending on the time frame, there may be motion blur caused by movement of the object or the camera—this is something that also didn’t exist in painting before photography.10 There is at least one example of motion blur in painting, though, that precedes photography, and it is the representation of the turning wheel in Velázquez’s painting Las Hilanderas (1657).

    • 11 See Joe Geigel and F. Kenton Musgrave, “A Model for Simulating the Photographic Development Proce (...)

    The sensor can produce overexposed or underexposed images.11

    • 12 See Geigel and Musgrave, “A Model for Simulating the Photographic Development Process on Digital (...)

    The sensor can leave specific disturbances in the image—e.g., grain with chemical film or specific artifacts with digital cameras.12

  1. The optical system can produce specific artifacts, such as lens flares (see fig. 1 for an example of a photo with lens flares). Lens flares will be the focus of the next part of the text.

1. John Monroe, lens flare and river in Yosemite National Park, California, August 3, 2006.

1. John Monroe, lens flare and river in Yosemite National Park, California, August 3, 2006.

© CC BY-SA 2.5.

11The lens flare between physics and economic viability

  • 13 Matthias Hullin et al, “Physically-Based Real-Time Lens Flare Rendering,” ACM Transactions on Gra (...)
  • 14 Shane Denson, Discorrelated Images (Durham, NC: Duke University Press, 2020), 243.
  • 15 See, for example, “LensFlare Studio,” BrainFeverMedia, accessed April 12, 2024, https://www.brain (...)

12Lens flares are often used in film in the context of special effects to make them more photo-like. A 2011 article from the field of computer graphics includes the following: “In fact, nowadays the use of lens flare is every bit as popular in games as it is in image and video editing. For the production of computer-generated movies, great effort has been taken to model cinema lenses with all their physical flaws and limitations.”13 Shane Denson has analyzed in detail how the lens flare is used in post-cinema and “how it continues to oscillate between simulation and the foregrounding of the simulation’s medial conditions of possibility.”14 There are many software tools nowadays available that allow users to add simulated lens flares to their images—there are special programs for this,15 although commonly used software like Photoshop also includes such possibilities.

  • 16 Arash Keshmirian, “A Physically-Based Approach for Lens Flare Simulation” (master’s thesis, Unive (...)
  • 17 John F. Hughes et al., Computer Graphics: Principles and Practice (Upper Saddle River, NJ: Addiso (...)

13What are the basic problems with the simulation of lens flares? Author Arash Keshmirian, who published a model for lens flare simulation, remarks, “Popular methods for producing lens flare effects employ artistic representations or rudimentary non-physical simulations.”16 Models, and therefore algorithms, “may be physically based, derived from first principles, or perhaps empirical or phenomenological, derived from observations or even intuition.”17

  • 18 Hughes, Computer Graphics, 713. See Phong, “Illumination for Computer Generated Pictures.”
  • 19 Alan Watt, 3D Computer Graphics (Harlow: Addison Wesley, 2000), 171–72.

14This points to the fact, mentioned already above, that in the process of PRCG, “rudimentary” approaches are used that do not in a strict sense simulate the visual phenomena. These algorithms work without being based on a sophisticated theoretical description of the phenomenon that is simulated. One important example of this kind of empirical, phenomenological, or rudimentary (these notions can be used here interchangeably) approach is Phong shading.18 This is an early algorithm for shading developed in 1975. Its goal is to give computer-generated visual objects the appearance of being lit by a light source, thereby helping to suggest a three-dimensional shape. It is not a physical simulation but based rather on an intuitive but computationally cheap approach, as is pointed out in a computer graphics textbook from 2000: “Despite the subsequent development of ‘global’ techniques, such as ray tracing and radiosity, Phong shading has remained ubiquitous. This is because it enables reality to be mimicked to an acceptable level at reasonable cost.”19 The point is that such approaches “mimic” or “imitate” the visual phenomena, which is not the same as simulating it. This is the economic aspect of rendering.

  • 20 See Jens Schröter, “What is a Virtual Image?,” in Virtual Images: Trilogy of Synthetic Realities (...)

15On the other hand, there are simulations which are (approximately) accurate in physical terms that require more computer resources—ray tracing and radiosity, as discussed in Kittler’s paper “Computer Graphics: A Semi-Technical Introduction,” are examples of this. In PRCG, such techniques are especially interesting since they are about simulating concrete aspects of the photographic dispositif. Virtual photography is real photography (in a structural sense), but not actual.20 The data from which the models, and therefore the algorithms, are constructed are either derived from measurements of real photographic apparatuses and/or from general physical models of light transport etc. This is the physical side of rendering.

  • 21 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 1.

16In the following, I want to focus on an informative paper by Matthias Hullin, Elmar Eisemann, Hans-Peter Seidel, and Sungkil Lee—an instructive example of the rendering of lens flare, especially as they give a good overview of other approaches. Their work starts by looking at the same tension that was mentioned by Keshmirian: “The problem of rendering lens flare has been approached from two ends.”21 They present a new approach that is conceived of as realistic enough and efficient enough. In the first figure in their article (fig. 2), they compare an actual and a virtual lens flare. The argument is that the algorithms used are good enough to reproduce the “personality” of the flare—i.e., its specific aesthetic originating in a certain lens system (see also fig. 4). The aim of the comparison is to show that the rendered image looks like a photo.

2. Comparison of the lens flare generated by a Canon zoom lens (left) and by Hullin’s virtual rendering system (right)

2. Comparison of the lens flare generated by a Canon zoom lens (left) and by Hullin’s virtual rendering system (right)

Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 1.

© Association for Computing Machinery

3. Photo of a bright light source (left) and its virtual reproduction by means of a chromatic Fourier transform of an aperture transmission function (right)

3. Photo of a bright light source (left) and its virtual reproduction by means of a chromatic Fourier transform of an aperture transmission function (right)

Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 4.

© Association for Computing Machinery

4. Lens flare of various lens systems—lens layout (top row); aperture fully open (middle row); aperture reduced by 4 f-stops (bottom row)

4. Lens flare of various lens systems—lens layout (top row); aperture fully open (middle row); aperture reduced by 4 f-stops (bottom row)

Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 7.

© Association for Computing Machinery

  • 22 They discuss, amongst others, Keshmirian, “A Physically-Based Approach for Lens Flare Simulation. (...)
  • 23 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 2.

17After having reviewed several approaches to the problem of mimicking or simulating lens flares,22 they present their model of the optical system of a photographic apparatus (fig. 2). The optical system is understood as a set of “building bricks”: lens design, materials, aperture, AR (= antireflective) coatings. Each of those bricks can be described mathematically. Each of the elements corresponds to one or more of the phenomenological photo-specific properties discussed in part 2. However, “depending on the requirements of the application, some effects can be skipped to simplify the model and increase the performance.”23 Photography does not need to be constructed as a whole—it is understood as a modular assemblage of elements: some can be simulated, some can be mimicked, and not all of them have to be used at all.

  • 24 Hullin, 3.
  • 25 Hullin, 3.

18In Hullin et al., the simulation of the lenses is based on a textbook on lens design and the geometry of the lenses, the form and behavior of the aperture, and the properties of the glass. Some aspects of the optical system are ignored for good reasons: “All optical glasses partially absorb light that passes through them. However, this is a weak effect (with typical light loss of a few percent across the entire lens system) and of low frequency (global attenuation). We therefore chose not to include it in our model.”24 But the anti-reflective coatings used in photographic optical systems to avoid lens flare are modeled, since “even the best available coatings are not perfect”25 and also add to the lens flare.

  • 26 Hullin, 3.
  • 27 See Mark Fox, Quantum Optics: An Introduction (Oxford: Oxford University Press, 2006).

19The behavior of light reflected in the optical system between the different surfaces of the lenses is modeled by understanding light as rays: “Every time a ray of light hits an interface between two media, a part of it is reflected, and the rest transmitted. It is the reflected part that gives rise to ghosting artifacts, which we seek to simulate.”26 This is interesting in itself: strictly speaking (with current physics), light is waves that can also be seen as particles, and therefore a physically correct description of light would have to draw on the mathematically complex field of quantum optics.27 But that is mathematically far too complicated and too costly to be rendered, even if it were possible.

  • 28 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 3.

20This is also the case if we are speaking only about the wave properties of light (without the quantum aspects), diffraction in particular: “In our system, we encounter two typical occurrences of diffraction: the starburst shape centered around the image of the light source and the subtle ringing patterns around the border of each reflection ghost. In a wave-optical framework, both can be computed exactly by evaluating the so-called diffraction integral for all points on the sensor. This is very costly and by no means possible in real time. Instead, we are interested in a computationally cheap approximation. As it turns out, we can convincingly reproduce the above effects by precomputing a small set of textures using the popular Fraunhofer and Fresnel approximations to the diffraction integral, respectively.”28

  • 29 Hullin, 4.
  • 30 See also Hullin, 6: “For a brute-force evaluation, most ghosts [= lens flares] are well represent (...)
  • 31 Hullin, 4.

21Again, in the fifth figure in Hullin’s article (fig. 3), we are shown a photo on the left. The simulated diffractive effect, the “starburst shape” is presented on the right. The shape is precomputed, using a simplified algorithm, kept in storage, then added to the light source in the rendering process and thus not costly to compute in real time. The following quote is somewhat technical but necessary to show that even with basic levels of programming, economic considerations come into play: “During runtime, we center it at the projected sensor location of the light source, and scale it in size (w, h) and intensity I as follows: where # is the f-number (as in ‘f / #’), and IRGB the radiant flux entering the lens expressed as an RGB vector.”29 This describes the basic parameters necessary to render the pattern. The f-number is a description of how much light can enter through the optical system, depending on the relationship between the focal length and the aperture: N = f / DIRGB describes the light that enters the optical system, differentiated into red, green, and blue components.30 But then they again stumble upon problems and opt for a pragmatic solution: “We are not aware of a simple way of obtaining the accurate scaling constants w0 (‘size’) and I0 (‘intensity’) that are specific to the optical system. Instead, since both parameters are rather intuitive, we leave this choice to the user, allowing them to either recreate the appearance of a given optical system, or to amplify and resize the starburst as they desire.”31 The diffractive starburst pattern can partially be specified by the users, based on their “intuition” and “desire”—i.e., how they want the final result to look. After the optical system is mathematically modeled, the rendering pipeline is presented. This is the name given to the concrete sequence of steps to be taken to render a certain type of image.

  • 32 Hullin, 5.

22At this point, Hullin et al.’s article explains the concrete steps used to compute a lens flare. I cannot explain every step in the chart in detail, but it shows how the light is modeled: a system involving a form of ray tracing (handling light as beams that traverse the optical system) is used. Again, decisions are made to make the rendering as effective as possible—for example, higher-order flares are not rendered since they are weak. Moreover, only a reduced, “sparse” bundle of rays is used: “The set of rays is sparse and, each ray taken by itself, would deliver insufficient quality. Our goal is to interpolate information from neighboring rays to estimate the behavior of an entire ray beam.”32 In the ray tracing, the “lens prescription,” meaning the geometric description of the chosen lens system, plays a role, while the “AR coatings”—i.e., the special coatings that should avoid unwanted reflections in the lens system—are modeled. On the right side, the extra steps are shown that are used to model the starburst texture which is finally added in the “framebuffer” to the rendered remainder of the lens flare pattern on the left side. After this, further techniques for acceleration are discussed. Finally, the results are presented—different lens systems that result in different lens flares (fig. 4).

  • 33 Gilles Deleuze, Difference and Repetition (New York: Columbia University Press, 1995), 209.
  • 34 See Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 8. The aspect of lens-system desig (...)
  • 35 Hullin et al. published additional material online, which not only allows a detailed look into th (...)

23In an additional chapter, they discuss “artistic control.” The virtual optical system to some extent approximates the real optical system. Deleuze makes an interesting point: “The reality of the virtual is structure.”33 It is in this sense that Hullin et al. discuss the possibilities of their model—since the optical system is a mathematically described virtual structure, it can be changed at will (as was already visible with the star-shaped diffraction pattern, discussed above); they discuss the possibility of weirdly formed apertures and the strange lens flares they would produce. But it is also possible to change the symmetries or colors. After discussing some possible areas of application—CG movie production, computer games, image and video processing, lens-system design, and deflaring34—they conclude their paper with a review of the limitations of their approach.35

24The approach of Hullin et al. is, of course, only one approach to the rendering of lens flare as a typical photo-specific visual trait, but their dense and informative paper is illuminating in several respects, as I have tried to show: it shows the permanent oscillation between physics and economic viability in the rendering of lens flares (figs. 5 and 6 show two further examples of rendered lens flares: a simple and a complex rendering).

5. Nimish Ajmani, example of computer-generated lens flare to give a more “real” look to CGI, February 20, 2008

5. Nimish Ajmani, example of computer-generated lens flare to give a more “real” look to CGI, February 20, 2008

© CC BY 3.0.

6. Lens flare generation and starburst effect of an eight-blade zoom lens

6. Lens flare generation and starburst effect of an eight-blade zoom lens

Based on the technique developed in Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics 30, no. 4 (2011): 8

© CC BY-SA 3.0.

  • 36 As an early example, see D. H. Kelly, “Systems Analysis of the Photographic Process. I. A Three-S (...)

25PRCG is a piecemeal process, as my reading of the paper by Hullin et al. tries to demonstrate. In this process, photography is itself understood algorithmically: as an operation that can be disassembled into separate steps. This is clearly described in an early paper published in 1960 by D. H. Kelly: “This report presents a set of methods for the quantitative characterization of photographic processes, based on a functional model of the types of transformations which are assumed to intervene between the initial exposure of a photographic emulsion (to an optical or ‘contact’ image of any kind) and the final density pattern (which is measurable after the film has been developed).”36 Such quantitative characterizations can be transformed step by step into algorithms for a computing system (if so desired).

26This process oscillates between, on the one hand, workarounds and crude approximations, which are cheap in terms of computer resources, and, on the other, physically accurate simulations—although these are often only partial and approximate because simulations of the wave properties of light, for example, are very costly. And are really accurate simulations necessary? It can be assumed that most of the audience for whom effects like lens flares are rendered (e.g., moviegoers and computer gamers) cannot decide if a given lens flare is an “artistic” workaround or a physically correct simulation. Perhaps an artistic workaround looks more satisfying than a physically correct simulation. The “look” of photography is produced with mixed, partial, and approximate strategies even in producing only one photo-specific trait—and, moreover, not all photo-like properties need to be included to produce an image that looks like a photo.

  • 37 Kittler, “Computer Graphics,” 32.
  • 38 Kittler, 36.
  • 39 Peter Trawny, Medium und Revolution (Berlin: Matthes & Seitz, 2011), 27 [translated].

27It is obvious that the algorithmic rendering of photo-like images oscillates between the poles of the physically and mathematically possible and the economically possible. PRCG is a “typical digital compromise between engineers and management experts.”37 However, this also means that the following statement is not entirely correct: “It is only in the name of impatience that all existing computer graphics are based on idealizations.”38 It is not merely impatience, it is the costs that decide which algorithms will be used. The history of computer graphics cannot—as Kittler suggests—be seen as the history of abstract models that are subject to purely mathematical considerations or the individual impatience of individual programmers, but rather as the history of those military, mass media, and other industries that produce PRCG. It would be more correct to define the medium as the “unity of capital and technology.”39

  • 40 Tagg, The Burden of Representation, 63.
  • 41 The programmable computer is a relatively un-specific machine—see Jens Schröter, Das Netz und die (...)
  • 42 Roland Barthes, Camera Lucida: Reflections on Photography (New York: Hill & Wang, 1981), 88.

28“Photography,” at least as implicitly modeled in PRCG, is a heterogeneous assemblage on several different levels. Does that mean, to come back to the questions discussed above, that “photography as such has no identity”?40 On the contrary, it has an identity in its recording of electromagnetic radiation, but this does not result in a simple, monolithic essence but rather a field of basic properties (as listed in part 2), not all of which need to be present, but at least some of which are present in each case. Compared to this, computer graphics has no real identity, because in computer graphics—in principle—any visual property can be realized that can be algorithmically represented and computed with the available resources.41 It is all the more striking, then, that photography has for so long been the model on which computer graphics is based. This obviously complicates simplifying notions like “digital revolution,” which imply a rupture caused by the introduction of digital technologies. Perhaps it is “the advent of the Photograph [sic] which divides the history of the world”42—in the sense that the photo-graph is the type of image that has formed visual culture in such a pervasive way that even the malleable and programmable algorithmic digital media have to follow it. In the photographic-looking visual mass media, this comes as no surprise. In that sense, many algorithms of computers today follow the algorithms of photography.

Top of page

Notes

1 Bruce Gooch and Amy Gooch, Non-Photorealistic Rendering (Natick, MA: AK Peters, 2001), 1. To avoid misunderstandings, it should be said that images produced with PCG and AI systems can, of course, be combined. Such hybrid approaches will be the standard in the future, I suppose.

2 Konrad F. Karner, Assessing the Realism of Local and Global Illumination Models, PhD diss. (Vienna: Oldenbourg, 1997), 26. See Gary W. Meyer et al, “An Experimental Evaluation of Computer Graphics Imagery,” ACM Transactions on Graphics 5, no. 1 (1986). The text describes an experiment to evaluate the photorealistic nature of computer images: a video image of a simple, real-world arrangement of objects and their lighting is presented to test subjects alongside a video image of a generated graphic representing that arrangement. If the subjects can no longer say with certainty which the image of the real arrangement is, the image is considered realistic. The video image, although electronic, is here counted among the photographic media insofar as it is based on the storage, processing, and transmission of light through lens systems similar to photography.

3 See Alan Watt, “Rendering Techniques: Past, Present and Future,” ACM Computing Surveys 28, no. 1 (1996): 157.

4 Friedrich A. Kittler, “Computer Graphics: A Semi-Technical Introduction,” Grey Room 2 (2001): 30. See also 36: “Computer graphics are differentiated from the cheap real-time effects of the visual entertainment media by a capacity to waste time that would rival that of good old painters if its users were just more patient.”

5 Bui Tuong Phong, “Illumination for Computer Generated Pictures,” Communications of the ACM 18, no. 6 (1975): 311.

6 Kittler, “Computer Graphics,” 37. A more detailed technical account of ray-tracing and radiosity algorithms is given in Andrew S. Glassner, Principles of Digital Image Synthesis (San Francisco: Morgan Publishers, 1995), chaps. 18 and 19.

7 “Diskurse der Fotografie,” Suhrkamp | Insel, accessed January 6, 2024, https://www.suhrkamp.de/buch/diskurse-der-fotografie-t-9783518291993 [translated].

8 John Tagg, The Burden of Representation: Essays on Photographies and Histories (Houndmills: Palgrave Macmillan, 1988), 63.

9 See Wolfgang Ullrich, Die Geschichte der Unschärfe (Berlin: Wagenbach, 2009). In his first chapter, Ullrich mentions the foggy scenarios of Turner and other romantic painters as examples of painterly blurriness, but most of the book is reserved for blur and out-of-focus-ness in photography. A scenario clouded by fog is not blurred in the sense that an out-of-focus photograph is blurred. See, on blurring images in computer graphics, Craig Kolb et al, “A Realistic Camera Model for Computer Graphics,” SIGGRAPH Proceedings (1995).

10 See Fernando Navarro et al, “Motion Blur Rendering: State of the Art,” Computer Graphics Forum 30, no. 1 (2011)

11 See Joe Geigel and F. Kenton Musgrave, “A Model for Simulating the Photographic Development Process on Digital Images,” SIGGRAPH Proceedings (1997).

12 See Geigel and Musgrave, “A Model for Simulating the Photographic Development Process on Digital Images.”

13 Matthias Hullin et al, “Physically-Based Real-Time Lens Flare Rendering,” ACM Transactions on Graphics 30, no. 4 (2011): 1. With their mention of computer-generated movies, they are referring to the films made by Pixar.

14 Shane Denson, Discorrelated Images (Durham, NC: Duke University Press, 2020), 243.

15 See, for example, “LensFlare Studio,” BrainFeverMedia, accessed April 12, 2024, https://www.brainfevermedia.com/lensFlareStudio.html.

16 Arash Keshmirian, “A Physically-Based Approach for Lens Flare Simulation” (master’s thesis, University of California, 2008), 1.

17 John F. Hughes et al., Computer Graphics: Principles and Practice (Upper Saddle River, NJ: Addison Wesley, 2014), 2. See also 713, where they differentiate between empirical/phenomenological models, measured/captured models, and physically based models.

18 Hughes, Computer Graphics, 713. See Phong, “Illumination for Computer Generated Pictures.”

19 Alan Watt, 3D Computer Graphics (Harlow: Addison Wesley, 2000), 171–72.

20 See Jens Schröter, “What is a Virtual Image?,” in Virtual Images: Trilogy of Synthetic Realities I, ed. Lars C. Grabbe, Patrick Rupert-Kruse, and Norbert M. Schmitz, Yearbook of Moving Image Studies 2020 (Marburg: Büchner, 2021). Simulations of photographic media belong in that sense to “virtual optics”: see Jens Schröter, 3D: History, Theory and Aesthetics of the Transplane Image (New York: Bloomsbury, 2014), 37–51 and chap. 10.

21 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 1.

22 They discuss, amongst others, Keshmirian, “A Physically-Based Approach for Lens Flare Simulation.”

23 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 2.

24 Hullin, 3.

25 Hullin, 3.

26 Hullin, 3.

27 See Mark Fox, Quantum Optics: An Introduction (Oxford: Oxford University Press, 2006).

28 Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 3.

29 Hullin, 4.

30 See also Hullin, 6: “For a brute-force evaluation, most ghosts [= lens flares] are well represented with only three wavelengths (RGB).” Focusing only on three wavelengths (which nevertheless add up to white light) again reduces computational costs.

31 Hullin, 4.

32 Hullin, 5.

33 Gilles Deleuze, Difference and Repetition (New York: Columbia University Press, 1995), 209.

34 See Hullin, “Physically-Based Real-Time Lens Flare Rendering,” 8. The aspect of lens-system design is especially interesting: “Even in live-action cinematography, lens flare is sometimes considered a desired effect. […] Certain lens systems (such as the Lensbaby) are designed for creative use of aberrations and other optical effects. In combination with a traditional lens design tool, our algorithm could help lens designers to preview and optimize the flare characteristic of an optical design.” The virtual optical systems can become a tool for the development of real optical systems—that is not surprising as this is one of the main uses of computer simulation, to anticipate and pre-construct the real. See Jens Schröter, “Computer/Simulation: Kopie ohne Original oder das Original kontrollierende Kopie?,” in OriginalKopie: Praktiken des Sekundären, ed. Gisela Fehrman et al. (Cologne: Dumont, 2004).

35 Hullin et al. published additional material online, which not only allows a detailed look into the code but also contains video clips that vividly demonstrate the results of the rendering. See https://dl.acm.org/doi/10.1145/2010324.1965003.

36 As an early example, see D. H. Kelly, “Systems Analysis of the Photographic Process. I. A Three-Stage Model,” Journal of the Optical Society of America 50, no. 3 (1960): 269.

37 Kittler, “Computer Graphics,” 32.

38 Kittler, 36.

39 Peter Trawny, Medium und Revolution (Berlin: Matthes & Seitz, 2011), 27 [translated].

40 Tagg, The Burden of Representation, 63.

41 The programmable computer is a relatively un-specific machine—see Jens Schröter, Das Netz und die virtuelle Realität: Zur Selbstprogrammierung der Gesellschaft durch die universelle Maschine (Bielefeld: transcript, 2004), 10.

42 Roland Barthes, Camera Lucida: Reflections on Photography (New York: Hill & Wang, 1981), 88.

Top of page

List of illustrations

Title 1. John Monroe, lens flare and river in Yosemite National Park, California, August 3, 2006.
Credits © CC BY-SA 2.5.
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-1.jpg
File image/jpeg, 460k
Title 2. Comparison of the lens flare generated by a Canon zoom lens (left) and by Hullin’s virtual rendering system (right)
Caption Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 1.
Credits © Association for Computing Machinery
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-2.jpg
File image/jpeg, 143k
Title 3. Photo of a bright light source (left) and its virtual reproduction by means of a chromatic Fourier transform of an aperture transmission function (right)
Caption Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 4.
Credits © Association for Computing Machinery
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-3.jpg
File image/jpeg, 84k
Title 4. Lens flare of various lens systems—lens layout (top row); aperture fully open (middle row); aperture reduced by 4 f-stops (bottom row)
Caption Illustration taken from Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 7.
Credits © Association for Computing Machinery
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-4.jpg
File image/jpeg, 143k
Title 5. Nimish Ajmani, example of computer-generated lens flare to give a more “real” look to CGI, February 20, 2008
Credits © CC BY 3.0.
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-5.jpg
File image/jpeg, 130k
Title 6. Lens flare generation and starburst effect of an eight-blade zoom lens
Caption Based on the technique developed in Matthias Hullin et al., Physically-Based Real-Time Lens Flare Rendering, “ACM Transactions on Graphics” 30, no. 4 (2011): 8
Credits © CC BY-SA 3.0.
URL http://journals.openedition.org/transbordeur/docannexe/image/2443/img-6.jpg
File image/jpeg, 147k
Top of page

References

Electronic reference

Jens Schröter, The Lens Flare. Photorealistic Computer Graphics between Physics and Economic ViabilityTransbordeur [Online], 9 | 2025, Online since 26 February 2025, connection on 13 January 2026. URL: http://journals.openedition.org/transbordeur/2443; DOI: https://doi.org/10.4000/13dww

Top of page

About the author

Jens Schröter

Prof. Jens Schröter has been chair of media studies at the University of Bonn since 2015. Recent publications: Medien und Ökonomie, Wiesbaden: Springer, 2019; (together with Christoph Ernst): Media Futures: Theory and Aesthetics, Basingstoke: Palgrave, 2021; (edited together with Andreas Sudmann et al.), Beynd Quantity: Research with Subsymbolic AI, Bielefeld: transcript, 2023.
Jens Schröter occupe la chaire Media Studies à l’université de Bonn depuis 2015. Parmi ses publications récentes figurent Medien und Ökonomie (Springer, 2019) ; avec Christoph Ernst, Media Futures. Theory and Aesthetics (Palgrave, 2021) ; avec Andreas Sudmann et al. (dir.), Beyond Quantity. Research with Subsymbolic AI (Transcript, 2023).

Top of page

Copyright

CC-BY-NC-ND-4.0

The text only may be used under licence CC BY-NC-ND 4.0. All other elements (illustrations, imported files) may be subject to specific use terms.

Top of page
Search OpenEdition Search

You will be redirected to OpenEdition Search