Exposure vs. Brightening (revised)

Started Apr 4, 2013 | Discussions thread
ForumParentFirstPreviousNextNext unread
Flat view
gollywop
Veteran MemberPosts: 5,786
Like?
Exposure vs. Brightening (revised)
Apr 4, 2013

_________________________________________________________________________

EDIT: This has been posted as an article and is available: here .

_________________________________________________________________________

This is a revised version of a thread I originally made on the mFT forum. This revision reflects a number of excellent suggestions made there, for which I am very grateful. Some material has been either moved to or added to a set of End Notes.

EXPOSURE VS. BRIGHTENING (revised)

Some recent threads have again raised concerns in the minds of some that the term exposure is often incorrectly defined and used to refer to notions outside its proper realm. This situation arises in part, I believe, because no alternative terminology is commonly accepted to replace the term exposure when it is misapplied. I intend to make the case here, following others that have come before me, that the term brightening provides simple, straightforward, and consistent terminology to correct this lack.

Let me begin with a simple model of camera behavior that will help motivate the suitability of the suggested terminology. Some may wish to skip the next section and go straight to the summary that follows. End Notes, denoted with a numbered *, are provided to deal with useful details not essential to the flow of the text.*1

BACKGROUND:

Properly defined, exposure is the amount of light falling per unit area on a sensor: it is determined by the scene illuminance, the f-ratio (more properly the t-ratio – the equivalent f-ratio for a lens with 100% transmittance), and the shutter speed. Note that ISO does not figure in this definition. Let us see why.

Light is made of photons, and the camera's sensor is essentially a photon counter. The photons that fall on the sensor during exposure release electrons, and it is these released electrons that make the signal that is processed, recorded, and eventually transformed into the final image.*2

In subsequent (and separable) actions, the charges of these electrons are read, amplified, and otherwise transformed in-camera (by hardware and software) as they are converted into the digital numbers (ADUs - analog to digital units) that comprise the recorded raw-data file. It is from these raw data that processed images, such as jpegs, will be made, either in-camera or with a raw processor on a computer. The camera's ISO setting determines this amplification factor in-camera. Raw processors can also manipulate these numbers, effectively allowing one to apply software-based "ISO" with a computer.

Rather generally, let me refer to any such amplification that takes place, either in-camera and/or in the raw processor, as brightening. Sometimes this brightening is referred to as ISO, but that term can be treacherous because it is only partially applicable to the total gain that may be applied and it is a concept used in very separate contexts with very different meanings (some of which are themselves inherently nebulous). In what follows, the term "in-camera ISO" will be used to refer to the camera's ISO setting.

The total brightening applied to an exposure in producing the final image, either in-camera or with a raw processor, determines the brightness of that image, a term that applies only to a final image. Thus, exposure is the input to the process of making an image -- photons falling on the sensor --, and brightness characterizes the final image that is the output -- the level of the data of the final image after brightening is applied to the exposure either by adjusting the ISO in-camera and/or by pushing or pulling in the raw conversion.

In brightening an exposure, it is possible to apply excessive amplification, either in-camera or with the raw processor, that causes either the raw data or the data defining the final image to exceed their characteristic bit-depths, i. e., to cause the data to be clipped. It seems reasonable to refer to any such clipping, whether of the raw data or final-image data, as over-brightening.

In contrast to over-brightening, we can consider the case where the sensor is subjected to so much light per unit area during exposure that some of its photosites (sensels) become exposed beyond their capacity to accommodate all the electrons released by the light they receive (blown). It seems reasonable to refer to this situation as over-exposure.*3   Both over-exposure and over-brightening will result in portions of the final image being clipped.

It is entirely possible, then, for an image to be "appropriately" exposed (no blown sensels) but over-brightened (clipped data) because of excessive brightening applied either in-camera or during processing. Such an image is sometimes inappropriately called "over-exposed" in common parlance (and even in much of the literature), but this is an incorrect use of the term. The image is correctly exposed, but over-brightened.

SUMMARY AND BASIC DEFINITIONS:

We can view the working of a camera in two steps: (1) Light from the framed scene is applied to a sensor that is able to represent that light as electrons. The density of light falling on the sensor in this step is determined by the illuminance of the scene, the f-ratio (more accurately, the t-ratio), and the shutter speed. (2) The charges of these electrons are then read, amplified, and transformed by the camera's hardware and software to produce the raw data from which an image is formed either in-camera by the camera's jpeg engine or with a computer using an appropriate raw processor.

Step (1) is exposure. It is here that the basic information from which the final image will be made is introduced into the camera through its "exposure" to the light. Step (2) is brightening. It is here that the basic exposure information obtained in step (1) is amplified and transformed by hardware and/or software ultimately into an image of the desired brightness.

Here is a flow-diagram depicting this process.

With the preceding model in mind, we make the following definitions of basic concepts:

Exposure: the amount of light falling per unit area on a sensor: it is determined by the scene illuminance, the f-ratio (more properly the t-ratio), and the shutter speed.

Brightness: the values of the data that characterize a final image after amplification is applied to the exposure either by adjusting the ISO in-camera, or by pushing or pulling in raw conversion. This term refers only to the final image.

Brightening: the process of applying gain to an exposure, either through ISO in-camera or tonal transformations in processing, to increase or alter the values of the raw data and/or the final-image data.

Over-exposed: an exposure in which at least some of the sensor's photosites (sensels) are exposed beyond their capacity (blown pixels).*3

Over-brightened: an application of gain that leads either the raw data or the final-image data values to exceed their bounds (clipping).

EXTENSIONS:

These basic notions allow for unambiguous descriptions of higher-order concepts, such as:

Brightness vs. brightening: brightness refers only to the level of the values of the final-image data, while brightening refers to any amplification that affects the values of either the raw data or the final-image data.

Over-exposed vs. over-brightened: over-exposure occurs when the bounds of the sensor are exceeded; over-brightening occurs when the bounds of the data are exceeded -- either the raw data or the final-image data. Over-exposure necessarily leads to clipped raw data and, hence, to over-brightening. While it is possible for an image resulting from over-exposure (blown sensor pixels) to be processed so that the final-image data do not exceed their bounds (for example, pulling back a blown sky in a jpeg so its values are less than 255), there will be a pile-up of data on the right-hand side of the histogram that indicate and betray the over-brightening. The converse, however, as we have noted, need not be true: it is possible for a properly exposed image (with no blown sensor pixels) to be over-brightened (having clipped raw or final-image data values).

Raw-data clipping vs. final-image data clipping: Over-exposure and/or excessive application of in-camera ISO can lead to clipped raw data. Likewise, excessive pushing of unclipped raw data can lead to clipped final-image data. In either case, we have over-brightening.

Recoverable highlights: clipped highlight values whose clipping has resulted from over-brightening by the jpeg processor of otherwise properly exposed and brightened (i.e., not clipped) raw values. Alternative tone curves applied to such raw data could result in a final image of appropriate brightness but without these highlights being clipped.

SHOOTING CONSIDERATIONS REDUX:

Here are outline statements of various shooting considerations employing the above terminology. Some may find them to have pedagogical value.

Achieving desired brightness shooting jpeg:

The jpeg shooter has two means for establishing the brightness of his/her final image: exposure and brightening. Given the scene, there are two camera controls that affect exposure, f-ratio and shutter speed, and there is one that affects brightening, ISO. For most jpeg shooting, the desired brightness is best achieved first by maximizing the exposure (while avoiding over-exposure) and then adding the minimal amount of ISO required, if any. The combined average effects of the two can be assessed roughly via the camera's metering. The trade-off in exposure between f-ratio and shutter speed must be resolved by considerations of DoF (depth of field) and acceptable motion blur and/or camera shake. An appropriate exposure may also be influenced by other artistic considerations.

The jpeg shooter should also be aware that achieving a desired degree of image brightness (average tonal value) can be accompanied by over-brightening (clipped highlights). Any such over-brightening should be assessed using the histograms and/or blinkies. This situation can only be resolved by some compromise, either using less exposure or brightening or by accepting the clipped highlights. Some cameras also incorporate tone-curve (highlight/shadows, D-lighting, auto-lighting, and the like) settings that can be used to alieviate this situation.

A jpeg final image that has been brightened to the desired brightness without maximizing the exposure (while avoiding over-exposure) may be said to be under-exposed. Under these circumstances it might be difficult to tell from the image itself that it was under-exposed, but if the exposure had instead been maximized, an image of the same brightness could have been achieved with less noise.

A jpeg final image that has been shot with maximum exposure (while avoiding over-exposure) but otherwise looks too dark is not under-exposed: it is under-brightened.

Achieving desired brightness shooting raw:

The raw shooter also has two means for establishing the desired brightness of his/her final image: exposure and brightening, but there are differences from shooting jpeg. Given the scene, exposure is still set using the camera's f-ratio and shutter speed, but brightening can take place either in-camera via the ISO setting or in the raw processor.

Barring overriding artistic considerations, the raw shooter should typically consider exposing to the right (ETTR). ETTR is an attempt to maximize the exposure at base ISO by choosing exposure settings that push the histogram to its right-hand edge without clipping. This has the benefit of maximizing signal and minimizing relative shot noise. If this results in an image that is too bright, one simply pulls the brightening back in raw processing.

Shooting considerations may make ETTR impossible. DoF considerations, for example, may dictate a high f-ratio, and camera-shake considerations may dictate a fairly high shutter speed. Together, these settings may entail an exposure that falls short of ETTR. What to do?

BTTR (brightening to the right) is a concept similar to ETTR except that ISO above the base level may be applied in camera to help place the histogram against its right-hand edge. If ETTR can be achieved (at base ISO) additional ISO would clearly be counterproductive, simply resulting in portions of the image being clipped. But if ETTR is not possible, one would typically still want to maximize exposure (i.e., push the histogram as far to the right as possible at base ISO) and then add brightening.

Whether the brightening should be added in-camera or during raw processing (or both) depends on the "ISO-nature" of the camera. With an ISOless camera (one whose read noise does not change with the camera's ISO setting), one could do either, but there are advantages to shooting dark (letting your image remain unbrightened) at the base ISO and brightening during raw processing. This will typically result in a final image with better IQ and less chance of clipped highlights. With a non-ISOless camera, the benefit is in favor of brightening with added in-camera ISO, which will typically result in less read noise than shooting darker and pushing in raw processing. Some cameras are semi-ISOless, becoming ISOless only after reaching a given ISO level, say 800 or 1600. Here there are benefits from increasing ISO in-camera, if needed, up to this level and then effecting any further brightening, if required, during raw processing.*4

Various constraints can come into play in determining an optimal exposure. These include DoF considerations affecting the f-ratio, motion-blur and/or camera-shake considerations affecting shutter speed, any desired amount of blown highlights (often, but not always, none), any desired shadow noise (or lack of it). Sometimes these constraints are incompatible and cannot all be achieved at once. In this case the raw shooter must make compromises. The maximal exposure satisfying these compromises while avoiding unwanted over-exposure is the optimal exposure, even if it is not ETTR. A shot based on any lesser exposure may be said to be under-exposed.

Assessing over-exposure in the camera:

Over-exposure occurs when there are blown photosites (sensels). We can't see the sensor, so there is no direct evidence of this condition. However, blown photosites will manifest themselves at base ISO as clipped raw data. Unfortunately, cameras also do not provide accurate indicators of the raw data. Until this happens, the only way to know for sure if raw-data clipping has occurred is to examine the raw file in a (free) program like RawDigger.

We can, however, get a rough idea of over-exposure by examining the histograms and/or blinkies of a shot taken at base ISO. Some cameras show these indicators in live view before the shot is taken. Virtually all cameras show them after the shot is taken. The post-shot indicators are more accurate than the live-view indicators, although the differences are often (but not always) minor. These indicators do not directly reflect the raw data, but rather, are based on a jpeg rendering of the raw data. When shooting raw, the camera's histograms provide the most accurate depiction of the raw data when using UniWB (a useful refinement, but not essential) and the camera's widest color space, usually Adobe RGB. Thus, the best, but not perfect, in-camera indication of over-exposure is a post-shot (at base ISO) histogram that indicates clipping or the presence of post-shot blinkies.

Caution, however, must be exercised when trying to assess over-exposure using the camera's jpeg-based color histograms when shooting ETTR or BTTR. Because the levels for the different underlying raw color pixels can differ greatly, attempting either ETTR or BTTR could lead to some raw color channels being clipped and others not. This situation might go undetected by the camera's histograms because the resulting demosaiced jpeg pixels need not be clipped but could nevertheless have the wrong color because of the one or more clipped raw channels. Conversely, unless the raw shooter is employing UniWB, the jpeg data will reflect the application of the WB multipliers. This could lead to the camera's color histograms, particularly in the blue or red channels, to indicate clipping that does not exist in the underlying raw data.

END NOTES:

1. An earlier draft of this paper was posted on dPreview.  It has benefited from the numerous excellent comments made there. I specifically wish to express my gratitude to Great Bustard. His communications have been particularly helpful and I am extremely grateful for his input. And, at the risk of ignoring several others who shouldn't be ignored, let me mention bobn2 who, along with GB, has been instrumental in helping to disabuse the photographic community of some fundamental misconceptions. I hasten, however, to add the usual disclaimer that I alone am responsible for this content and any possible erroneous constructions.

2. Not all the photons falling on the sensor produce a an electron; the proportion that do defines the sensor's quantum efficiency (QE), the average number of incident photons required to release an electron. For example, a QE of 50% means that two photons, on average, are required to release one electron. So, a sensor with a QE of 50% would require only half the exposure to record the same amount of light as a sensor with a QE of 25%. QE is an important element in characterizing a sensor's response to exposure.

DxO Mark measures a camera's base ISO as inversely proportional to the exposure to light applied directly to the sensor (illumination times exposure time) needed to achieve sensor saturation. For two sensors with equal saturation capacity, but one with twice the QE, it would take only half the exposure to saturate the sensor with the higher QE, and hence it would have twice the base ISO. If, however, that sensor with twice the QE also had twice the saturation capacity, then both sensors would saturate at the same exposure, and so their base ISOs would be the same -- but, in doing so, the one with the higher QE would have acquired twice the signal, entailing approximately a 41 percent betterment in signal-to-noise (or, if you prefer, a 30 percent reduction in noise-to-signal), and, thus, would display roughly one stop more DR.

3. I use the term over-exposed here advisedly because it is so encumbered with inappropriate connotations that I fear its use will always engender confusion. In an earlier draft of this work I used the term over-saturated instead in an attempt to sidestep this issue, but, of course, that term is also confusing, perhaps even more so, because of its long-established use in a wholly different context describing color intensity. So, in light of the fact that a major impetus for this work was to reclaim the word exposure properly to mean the light density falling on the sensor, I decided it would be silly not also to reclaim it to describe the situation where that light density exceeds the sensor's bounds as over-exposure.

Thus, when used in the term over-exposure, the word exposure must be regarded as defined above in the text. Over-exposure occurs when so much light falls on some photosites (sensels) that more electrons are created than can be retained (the sensel's full-well capacity is breached.) In the vernacular: these pixels are blown.

4. Whether a camera is ISOless, non-ISOless, or semi-ISOless has to do with how the read noise changes as the ISO increases. With an ISOless camera, the read noise remains essentially constant with increasing ISO. With a non-ISOless camera, the read noise falls as ISO increases. With a semi-ISOless camera, the read noise falls up to some limit, say 800 or 1600, and is essentially unchanging thereafter.

You can check out the ISO-qualities of many cameras at http://www.sensorgen.info. The variation of read noise with ISO is depicted in the first chart given for each camera. The Nikon 7000 is a good example of an ISOless camera. The Canon D5 exemplifies a non-ISOless camera, and the Olympus OM-D, E-M5 is a good example of a semi-ISOless camera (up to ISO 800).

Let's see how the ISO-quality of a camera affects the choice of where to do brightening when shooting raw. I am assuming here that you have established and set an optimal exposure at base ISO (see above: Achieving desired brightness shooting raw) that, for whatever reason, is less than ETTR -- so you could consider BTTR, i.e., adding some brightening to bring your image up to an acceptable brightness. [Clearly, if you could attain desired brightness by ETTR (at base ISO), there is no need to add any further brightening, either in-camera or out.]

With a non-ISOless camera, read noise decreases with increasing in-camera ISO, so there are advantages to brightening the image using in-camera ISO. Recall that I am assuming that the exposure is optimal and remains fixed as you alter ISO. The shot noise will therefore remain the same, but read noise will diminish as you increase brightness. This is good. But you must be aware that as you increase ISO you also increase your chances of inadvertently clipping data through over-brightening, so care must be taken (such as it can be).

With an ISOless camera, by contrast, when shooting raw there is no advantage to ultimate image quality to increasing in-camera ISO since the read noise is not reduced as a result. In fact, you are better off in terms of ultimate image quality shooting dark (letting your image remain unbrightened) at base ISO and doing all the brightening in your raw processor. A good raw processor can do a better job of it and, even more importantly, give you control in preventing clipping (which can happen without your knowing it when you apply ISO in-camera).

Now, this is not to say that there may not be reasons to use in-camera ISO despite the camera's being ISOless. For example, because the shot at base ISO may need brightening, the LCD review image may appear too dark, and you might desire to increase ISO to simply to brighten it. Or you might be shooting raw plus jpeg and wish to be sure that the jpeg has appropriate brightness regardless of any raw-processing considerations.

With a semi-ISOless camera, read noise decreases with increasing in-camera ISO only up to some point (usually 1600 or less), and thereafter read noise is relatively unchanging. Further increases in brightening, should they be needed, may therefore be done as well or better in the computer than with additional in-camera ISO -- and this will also allow better control over clipping.

5. Additional reading: A supurb explanation of read noise and its relation to in-camera ISO by dosdan is to be found at http://forums.dpreview.com/forums/post/51160056.

An equally informative read on comparing noise and resolution among sensors with differing pixel sizes is to be found in Dan Browning's (briander) piece, Myth busted, small pixels are bad", http://forums.dpreview.com/forums/post/32064270 .

A classic monograph dealing with the concept of "equivalence," the proper way to compare lenses and their settings across systems with differing sensor sizes (mFT vs. FF, for example) is to be found at http://www.josephjamesphotography.com/equivalence/index.htm. This is no short read. An excellent summary of the main notions is to be found at GB's thread http://www.dpreview.com/forums/post/51203390 .

These are all highly recommended reading in the context of this article.

-- hide signature --

gollywop

ForumParentFirstPreviousNextNext unread
Flat view
Post (hide subjects)Posted by
ForumParentFirstPreviousNextNext unread
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow