Oct 2013

About HDR

Introduction to HDR

High-dynamic range imaging (HDRI) is a method of digital photography, where the limited dynamic range of a digital camera sensor is overcome by combining multiple exposures of a one scene/subject. Compared to the human eye, digital sensors have inferior dynamic range, which means that shots with large contrasts bear those almost iconic burnt-out whites or drowned blacks. Burnt-whites or drowned blacks are sensors readings as maximum and minimum values of the incoming light, retaining no information. If sky is tried to capture within the sensors dynamic range, sky being brighter than other parts of a usual outdoors scene, the other pats of the shot are underexposed or even cut to black. If then the shadow detail is desired to be captured into the image, tihen the sky is blown out. The usual outcome of camera’s automatic calibration is a solution, where something is given off from both of the extreme ends of the dynamic range. A bit of the sky is blown out and bit of the shadows are cut to black. This is called Low-Dynamic Range imaging or LDR for short, and images even when shot in RAW-format, are LDR-images. In HDR-photo there are atleast three exposures of one subject: one where shadow detail is visible, one with the “standard” settings, or the best estimate of cameras automatic calibration, and one for highlight detail. When these three exposures are combined into one image, the resulting photo is a 32-bit floating point image. Normally image is 8-bits. The bits refer to the color-depth, as to what extent the colors can be reproduced. The 8-bit image is limited to hundreds of thousands, and 32-bit counts in the trillions, in almost infinite number of colors. When in 8-bit or 16-bit depth images, every pixel has a certain value assigned to it, in the floating-point (where the name comes from) every pixel can have a possible value from a range of values, making it a superior format in color and exposure precision. Though it is very limited where full 32-bit images can be shown, so a process of tone-mapping is applied into the image, when it’s converted from floating-point format to either 16-bit or 8-bit. The tone-mapping process combines all detail from the different exposure values of the HDR to one LDR-image when it can be displayed normally over a range of services and devices.

Misconceptions About HDRI

There are many varying misconceptions as to what high-dynamic range imaging essentially is. Most common misconception of all is propably of the images where there is extreme local contrast due to tone-mapping, which is somewhat uneasy to look at for longer periods of time. It almost seems like that this style of processing makes the images look flat, and miss all shadow and highlight. It is because the HDR holds within great ranges of colors, processing can lead to a point where highlights are brought down and shadows are brought up, and the image becomes hyperreal. Usual features of a HDR-image are ringing or halo artifacts, which are the glowing auras between boundaries of areas holding greatly different contrast. The other features include overvivid colors, too dark highlights and too light shadows. Also excessive glow is present, all this coupled with overmodified local-contrast, which is the key point in bringing the “HDR-look”: it brings out large scale details in the image. All of these, however, are results of faulty image processing. A true HDR is a subtle, not necessarily differing from ordinary images that much. However, in the end, the purpose of a HDR-image is to capture the dynamic range of a scene or subject as accurately as possible, usually keeping the human-eye as a reference, and alter the scene visually as little as possible. There are, in fact, sites for “true HDR’s”, which tries to clean the somewhat faulty reputation of the term HDR and prohibits submission of overprocessed images.

However, the overprocessing-technique can be used to one’s advantage. When using in a good sense, the tone-mapping can bring out details in the image, which would otherwise be much more unnoticeable, and can increase the visual appeal of the image. Though there is also a dividing line here, and some people do not accept even mild amounts of tone-mapping.

Good example to take in here are the urban explorers. Urban explorers in short are people who access or “infiltrate” as it is often referred to, to abandoned, destroyed, decayed, ruined or structures otherwise unaccessible to public, and take photos from these locations. There are generally two schools of urban explorers: The ones who take documentative photographs, which usually includes less to no image-editing, and the others who take images from these because often the urban-decay settings are moody and atmospheric, and are willing to bring forth that feeling by editing the image. And this is where the tone-mapping process comes along. It is sometimes referred to as Grunge Dynamic Range or GDR for short, which means to bring out the dirtiness and trash usually present in these scenes, which essentially makes the image and the effects of urban-decay all the more striking.

Overprocessed image showing all the typical features

Mildly tone-mapped image for visual appeal

HDR-image processed only to preserve the whole dynamic range

Shooting HDR

There are many ways how one can shoot HDR’s. Nowadays even some cellphones are claimed to be able to produce HDR-images, though this is largely doubtful as to what is their true quality. HDR-imagery does not require bleeding-edge equipment, but it does need a camera where one can set the properties manually, eg. shutter speed, iso setting, aperture and so on. And usually DSLR’s are the ones providing these options. There are external devices for HDR-bracketing, but if one happens to own a Canon DSLR, then I strongly suggest downloading and installing the Magic Lantern customized firmware, which can be downloaded and used free of charge, and it installs loads of supremely useful and interesting features to your DSLR.

Bracketing means the process of shooting atleast three or more exposures of the same scene, and usually it is automated, for the reasons that selecting exposures by hand which are equally for example two EV’s (Exposure Values) apart is quite difficult and also time-consuming compared to the automated alternatives. Fast HDR-bracketing is crucial especially when shooting moving objects, for example clouds.

Choosing the exposure value count

As with almost all things which have considerable user-base, the variables of certain process or subject of interest are tried to be operationalized in some way. HDR-imagery, maybe the most popular way of “telling the quality” of the image is to shoot it almost as many EV’s as possible. The truth is, it depends on the scene. For example, let’s take a scene with forest landscape and overcast sky atop. The skylight is dispersed through the clouds equally into the woodlands, and the resulting scene does not contain many peaks of highlight or drops of shadow. The contrast of the scene is low. So then the one does not need to shoot 7 shots of two EV’s apart (this would be called 7-step 2-EV HDR), but the basic 3 shots of two EV’s apart (3-step 2-EV HDR) is quite sufficient. But let’s take an interiors scene, which is lit only by natural light through windows or an opening of sorts. There is a smooth gradient from dark to light blue as day is nearly turning to night. All these should be captured into one image. Then considering 7 exposure values would be a good point. The sky-gradient alone is a difficulty, for when color-depth of the image is not sufficient to reproduce the smooth sliding from one color value to next, then something which is called color-banding appears to the sky, and then the sky seems as to be composed of circles slightly darker or lighter than the one before. And in addition to that, the shadows in the corners and maybe the room with not direct light coming in, and only some light from light bounces, it is a demanding one. But with 7 exposures, enough information is provided to compile an HDR-image to hold all the different color-intensities.


3-step 2-EV image


7-step 2-EV image

Compiling the Exposures

There are many programs with which one can compile the exposures into a 32-bit floating point image, both free and commercial. I use Photoshop’s HDR Pro, which has proved to be very versatile and working option, with great ghosting removal tools. Ghosting is the artifact of HDR’s where the image-processing algorithm combines the different exposures, and something differs between the exposures, for example moving clouds in the sky. Then the algorithm tries to interpolate the differences between the frames, which usually leads to nasty artifacts, usually carrying same characteristics with chromatic aberration. Green and violet features, for example cloud puffs having green and violet aura surrounding it, as a result of unsuccessful interpolation between the differing features.

HDR-Image Post-Processing

When the image is processed into an HDR from several exposures for example in Photoshop’s Merge to HDR Pro, it can be tone-mapped there, or after the image is combined. This is where the user decides on what effects are applied, when the 32-bit floating-point image is converted to regular 16-bit or 8-bit image. And to mention it for clarity, tone-mapping is in the heart of HDR image-processing. It should not be confused, that every tone-mapped image is one of those overprocessed, visually exhausting pieces, but tone-mapping is always applied when the HDR-image is converted, or it can be applied to it while keeping it in floating-point format. If looking for visual realism while maintaining all intensities of the dynamic range, a method called “Highlight Compression” is propably the best way to start. It does nothing else to the image, than brings highlights down. If your image looks blown out in some parts, applying this technique brings equalizes the histogram in a way, that all color-intensities are visible. Sometimes the Hightlight Compression compresses too much, and it might seem that the extreme highlights like the sun at it’s surrounding areas are too bleak, then Levels settings can be applied to bring back some of the glare into the sun and the sky.

The method for shifting the images into a more artistic realm is the Local Adaptation. That is the process where large-scale details are pumped out from the image by modifying local contrast. Now this can be powerful tool if used right, it is really easy to spoil the image and render it completely useless. One should be careful with the settings, and it might get some time to use to.

Uses of HDR-imagery

HDR-images are used for variety of purposes. Propably the most general one is for artistic reasons, as like with tone-mapping mentioned above, or just to capture the full-fidelity of a scene in photograph. Also HDR’s are used in computer graphics. When HDR’s are shot in a manner that a panoramic image can be compiled, the computer graphics programs (3DS Max, Maya, etc.) can reproduce the lighting-conditions present in the image, and use it as a mean of illuminating a object or scene in the program, thus bringing realistic lighting, shadows, and reflections which would be really hard or nearly impossible to reproduce within the program. Maybe widest application of this is the visual effects industry, for example in movies where often computer-generated subject needs to be inserted into the live-action plate. Then the HDR is captured from the location where the plate was shot, and thus the CGI-subject is easy to integrate into the footage. Also one very popular subject is architectural visualization, where realism is keypoint in depicting structures and their properties, both for visual appeal and later building-process in real life.

Single Image False-HDR

The “HDR-look” can be achieved using only photograph shot in RAW-format. By exposing the picture correctly, one can bring all the color-intensities back into the image, which makes it essentially look tone-mapped. While the replication might be pretty accurate, it can’t be circumvented that the RAW-format simply does not hold as much color-depth as the floating-point format and some artifacts may occur, as color-banding in smooth gradients, or colors “breaking”.