Better to underexpose or overexpose?

<p>Hello! I've been trying to find the answer to this for several months but all the information I'm finding is quite contradicting. It seems like most people agree that it's better to underexpose because in digital, once the highlgihts are blown, they're gone. But at the same time, I'm reading articles that show the distribution of the bits over the tonal range of the file and they all consistantly show that the most data is held in the brightest stop of data while the least information is found in the darkest tones of the image. Can someone please sort this out for me? I don't understand how the area with the most information would be the most sensitive to clipping. Thanks!</p>

<p>It's like this, you have a number of containers with a different amount of water in each one. You chose a measuring cup so you can measure the amount of water. But you have to blindly dump the all of the contents of each container into the cup, if it overflows, you can only record a full cup, you have no idea how much was spilled. That is clipping. </p>

<p>The left-to-right distribution of values in a histogram is determined by the lighting in the scene. In a low key or underexposed sample the distribution weighs farther to the left (lower light values). In a high key or overexposed the bias goes right. It's possible to "clip" either end (reach a point where there is no visible detail registered). The nature of digital sensors is to increase noise in the shadow values and wash out highlight values. People seem to choose where they bias their histograms by what they want most to avoid - murky, low-detail shadows, or blank white highlights. The goal seems to be filling the histogram from left to right without going off either extreme end. That's difficult to do with the limited dynamic range of some sensors without using fill light or in some way lowering the contrast of the scene's elements.</p>

<p>Also, it's not absolute; it's relative. It's not so much to expose for the very brightest highlights as it is to compromise in such a way that you can retain some highlight detail, while still being able to recover some detail in shadows.<br>

When the subject contrast is really very great, <a href="http://en.wikipedia.org/wiki/High_dynamic_range_imaging">HDR</a> may be necessary. In more typical settings you can do what is jokingly called the "<a href="../casual-conversations-forum/00Qwkz">Ozone</a>" system (by analogy to the "Zone System") of using ACR or "Shadow and Highlight" in Photoshop to get detail in highlights and shadow. ;)</p>

<p>There is often much more detail present in the slide or digital image than the human eye will register. That's why HDR may often be overkill.</p>

<p>Here's an example - A Kodachrome slide of Gay Head that was exposed for the sky, making the shadows very deep. On the right, is a 'development' of the image in ACR to give some shadow detail in the brick.</p>

<p> </p><div></div>

<p>If you overexpose a digital image, you cannot fix the highlights to restore detail. Raw conversion software has ways of <em>simulating </em>highlight recovery -- to some extent, but it's done by creating detail using what is in adjacent pixels. At the other end of the scale, there is shadow data there to be used if necessary, but it comes at a price - it will have noise somewhat as if you had used a higher ISO rating, and often, it ends up looking a little odd in comparison to the rest of the picture. So, depending on what you want the picture to look like, you neither want to overexpose it or underexpose it.</p>

<p>If you have a scene in which there is too great a difference between the shadows and the highlights to allow detail in both, it's better to expose it in a way that preserves highlight detail, because you can always bring up shadow detail afterwards. It may not be perfect, but it may be adequate for the purpose.</p>

<p>In other words, you can recover more shadow detail in a raw image than you can highlight detail, so if you have to, it's better to NOT overexpose. </p>

<p>Very good question! There's some guy who's claimed you should overexpose a bit, because a RAW converter can recover a bit of blown out information. While this capability does exist, It is not a good solution and, in my experience, does not result in quality highlights. I consider highlight recovery a fail-safe feature, and I'm sometimes glad it's there. However, I think the object (for me) is not to have to use it. I try not to live life that close to the edge.</p>

<p>Digital images clip one channel at a time, and that leads to abrupt hue changes in highlight areas. Photograph a sunset with just a touch of overexposure, and you'll see what I mean. In my experience, the most pleasing highlights occur when the response curve rolls off just a bit, forming a tiny shoulder. (I'm using terminology describing film response curves if you don't know what it means and want to google it.) By having a roll-off, there are no sudden color transitions, so much as a smooth loss of saturation and transition into white.</p>

<p>Don't be afraid to blow out highlights that are unimportant to you. If you are a strict follower of the ETTR method or the expose-a-fraction-of-a-stop-beyond-ETTR method, you might not get the best exposure for what really matters to you in the photo. This is especially true of a very contrasty scene. What I often do is to spot meter the brightest object in the scene where I want detail and expose 1 - 2 stops more, depending on what it is. I suppose this is an extremely dumbed-down zone system approach. Another approach might be to expose as much as possible, but dial back the exposure if you see flashies over any important part of the image (e.g. have a flashing sky, but not a flashing face). If in doubt, I'll do a bit of exposure bracketing and select my frames in post. I will then apply a curve with a bit of shoulder to roll the highlights gracefully into white.</p>

<p> </p>

<p>[[it's better to expose it in a way that preserves highlight detail,]]</p>

<p>It is better to define what constitutes a highlight detail to begin with (as JDM points out) for any given photograph. Just preserving the detail of the brightest area in a scene is quite often the wrong choice.</p>

<p>[[in other words, you can recover more shadow detail in a raw image than you can highlight detail]]</p>

<p>I don't think you are comparing apples to apples here. If your shadows are clipped you have no information. You can't recover detail in clipped shadows any more than you can recover detail in completely blown highlights.</p>

<p> </p>

You have to learn when to underexpose (beach, snow, whitewater) and when to overexpose (white subjects in bright sunlight, light subjects in dim shade). Or use your spot meter.

 

EV compensation is the most important control on a digital camera, I feel. The Canon G11 and Nikon P7100 have excellent dials for +/-. Most current cameras autofocus and set shutter speed, aperture, and ISO about as well as they can, but metering can be fooled by certain scenes.

 

Adding to Pierre's comment, you can recover more shadow than highlight detail in most JPEG images, so this property of digital sensors is not unique to RAW mode. However always underexposing is a bad idea because it increases noise and reduces detail.

 

ETTR is a lot of trouble considering that many fool-the-meter images are improved by small EV adjustments in the .33 to .67 range.

<p>PL: <em>"...Raw conversion software has ways of simulating highlight recovery -- to some extent, but it's done by creating detail using what is in adjacent pixels..."</em></p>

<p>Often, only one channel, often the red one, is blown. Highlight recovery uses the non-blown channels to estimate a value for the blown channel. </p>

<p>Information from adjacent pixels could also be used to say, estimate the hue, saturation or luminosity of a given blown pixel, but it wouldn't be a very robust approach since often large areas of continuous color have one channel blown and their hue, saturation and even luminosity would also be in error. </p>

<p>JDM, re "Ozone" ... IMHO, it's a bit risky to use that term, even tongue-in-cheek, since there is a commercial plugin with the very same name that is used for exposure control? It used to be sold for stand-alone use by a small company, Digital Film Tools, and then Tiffen bought the rights, and it is now part of their extensive DFX plugin filter package:</p>

<p>http://www.outbackphoto.com/workflow/wf_a118/essay.html<br>

http://photoshop.pluginsworld.com/plugins/adobe/376/ozone/rPage.html<br>

http://www.tiffensoftware.com/products/dfx-photo-plug-in</p>

<p>Tom M</p>

 

<blockquote>

<p>I'm reading articles that show the distribution of the bits over the tonal range of the file and they all consistantly show that the most data is held in the brightest stop of data...</p>

</blockquote>

<p>There are folks out there, who should know better, but don't don't understand the difference between counting and analog to digital (A/D) conversion. A sensor assembly converts analog levels of light into discrete numeric values. The numeric values are in binary (base 2). Many sensors output 2^12 bits, which is equal to 4096 discrete levels (per color channel). All the levels are 1/4096 of the total output range. Think of this as a flight of 4096 evenly spaced steps.</p>

<p>Now for the counting confusion - If a pixel's brightness puts it on step 16 (decimal), in binary that is 10000. The "most significant bit" (MSB) is the fifth column from the right and equal to 1. Moving to the left each column is double in value in binary (and ten times as much in decimal), but that is an artifact of counting in binary and has nothing to do with the step it's on.</p>

<p>><br>

There's some guy who's claimed you should overexpose a bit, because a RAW converter can recover a bit of blown out information. While this capability does exist, It is not a good solution and, in my experience, does not result in quality highlights.<br>

><br>

Do you mean Thomas Knoll, Andrew Rodney, or Jeff Schewe?<br>

ETTR (Expose To The Right) doesn't mean Over Expose To The Right.</p>

<p>Quoting from Jeff Schewe's article about the topic:</p>

<p>"This is not intended to be used in every photographic situation, only those situations where the scene's contrast range is lower than you camera's sensor dynamic range. Clearly if you have highlights whose texture and detail is important, you wouldn't want to increase the exposure to the point where the highlights are clipping."</p>

<p>More here:<br>

<a href="http://schewephoto.com/ETTR/index.html">http://schewephoto.com/ETTR/index.html</a> </p>

<p>So take a moment and read the article. I thought it is very informative.</p>

<p> </p>

<p>Never mind reading about it here or anywhere else, my post or anyone else's. All you need to do is experiment for yourself, and look at the results. It doesn't take long to form an opinion about it, with the camera YOU are using.</p>

<p>Sarah's mentioned a sunset above. I think that, and sunrises too, are indeed an excellent example. The internet is full of highly-rated, highly-favorited, and horribly bad, ugly sunrise/sunset pictures, because the sun or the bright parts near it are totally blown, nothing but 255 white, and the transition to them is extremely abrupt (hence the ugliness of it all). Expose to the right for that and there is little you can do. Highlight recovery looks just like that, highlight recovery. Underexpose a bit, and you can make that look very smooth and film-like. On the other hand, underexpose too much and you don't have much room left to boost shadows and/or midtones without obvious noise. This is why my original reply says you should neither over or underexpose... as a rule. Now, it's pretty obvious that both over and underexposure can be creative decisions. I just assumed that was understood by anyone smart enough to be on photo.net.</p>

<p>There's a famous photographer who has example pictures on the internet showing how well his camera's intelligent metering renders a snowy landscape type scene, all by itself. I look at those, and I have to assume he likes his snow to be snow-white. Me, I don't like his pictures as far as that goes, because I like to see the twinkleness of the snow crystals on the surface rather than just white. So, while I would do the same scene by underexposing it a bit (that is, underexposure according to the histogram). It would give me more leeway to do what I want. That's an example of a creative decision.</p>

<p>> Never mind reading about it here or anywhere else,<br>

Oh really? </p>

<p>Don, long ago I read a lengthy article by some guy (can't remember his name, or I would have stated it) who believed EVERY exposure should extand a certain fraction of a stop beyond the righthand end of the luminance histogram (i.e. ETTR + a fraction of a stop). As I recall, it was something like 1/2 or 2/3 of a stop -- the limits to which he believed highlight recovery could be performed. His approach was not as reasoned as the quote you cite (Schewe's). It also did not take into account such matters as the saturation of the blown-out highlights. (Saturated highlights blow out worse and more easily, with more disastrous consequences, than unsaturated highlights.) I do not agree with his method.</p>

<p>As I said, I do not expose my photographs my way. What I do works very well for me. As Pierre aptly points out above, there is no single, absolutely correct way to expose a photo. It's always a creative decision.</p>

<p>OK, Don, I've read (skimmed, really) Schewe's article. I take no issue with anything he says. It's not the article to which I referred, which as I recall was found on an ugly website with a yellow background. Perhaps you haven't run across it. The article was all the rage for a month or two, and I haven't heard about it since.</p>

<p>I suppose there are two ways to interpret the term "overexpose." One is with reference to average luminance, as the camera would ordinarily meter it. If I were trying to take the best possible picture of a uniformly illuminated white wall, I'd want to expose as many as a couple of stops more than my camera would do it -- essentially an ETTR approach.</p>

<p>The second interpretation of "overexpose" would be in the sense of clipping at least one color channel. I sometimes do this too, but I'm careful about how I do it, and many times I will not do it. The guy to whom I referred advocates for the this variety of overexposure in each and every frame. As I recall (and my memory is a bit foggy on this point), he does not actually use the term "highlight recovery," but that is the feature he is exploiting.</p>

<p>I do not use any of these exposure techniques all the time for all situations. My exposure technique could be called EFTS (Expose For The Situation). ;-)</p>

<p>If you consider either "overexpose" and "underexpose" as getting areas of the image <strong>with no recorded information</strong> you can not judge which one is the better, because <strong>both are bad. </strong><br>

If you talk about deviating from the <strong>ideal</strong> EV value to record the scene, both luminance and color rendition/saturation wise, then you should go to the right because you still have the information there and gain in the shadows, that are more likely to generate noise if underexposed.<br>

The point is that when you look at the histogram you see the both the luminance and RGB representation in a JPEG image, but the RAW file has already some more information there that you can use when you convert it. This makes the exposure of JPEG images more demanding and as you only have 8 bits per channel you can not afford to go out of the dynamic range of your sensor, unless you do it on purpose and you want or you don't mind if you get completely black or washed out areas in parts of the image. </p>

<p>Us old film people (JDM... I can remember High Speed Ektachrome at 160) knew that to blow the highlights on slide emulsions was to lose them completely so if there was time, read the highlight value and go with that. Sarah's EFTS could be used to describe the process. To me the value of ETTR is to reduce the shadow noise by moving the whole histogram to the right (not overexposing the highlights) and then dialing back the the brightness in post processing. Don't know if this is useful to others but I also lower the in camera contrast and crank it up in post.</p>

<p>FAIW, Randy, I have my cameras set to low contrast picture styles too, but that doesn't really affect the RAW data. All it affects is the way the jpeg (including the thumbnail jpeg) is rendered. Because the histogram is taken from the thumbnail jpeg, which is what you view on the LCD, the contrast adjustment also impacts the shape of the histogram. The righthand edge will remain the same, and blown-out is still blown-out (at the same exposure). However, the lefthand edge shows more shadow detail, as though the X axis has been extended to lower values. Basically all you get is a more informative histogram in the LCD playback.</p>

Jeff, if you are still following this, could you cite the source that said "the most data is held in the brightest stop of data while the least information is found in the darkest tones of the image" please? Thanks, I'm curious.

 

Bottom line: highlights are the Achilles heel of digital sensors, which is why a lot of ASCII characters have been wasted talking about it. My solution: Fuji EXR.

<p>Thanks everybody for the responses. I already understood that it's better to not under or over expose at all; I guess my question was more about the technial aspect of the RAW image file rather than a subjective choice to either over or under expose. I hope that made sense.</p>

<p>Bill, here's one of the articles I read about the bit distribution. <a href="http://www.earthboundlight.com/phototips/nikon-d300-d3-14-bit-versus-12-bit.html">http://www.earthboundlight.com/phototips/nikon-d300-d3-14-bit-versus-12-bit.html</a></p>

<p>Jeff, there's not much magic to the RAW file. The more important question, IMO, is to ask how the RAW file corresponds to the histogram and the flashies, so that you'll know how to really get the exposure you think you're getting. After quite a bit of testing with my various Canon cameras (might be different for Nikon et al.), I determined that a pegged RGB histogram always accurately corresponds to maxed out channels in the RAW file with either normal or low contrast picture settings. I didn't test high contrast, because I'm not interested in that. With a prior camera that only had a luminance histogram (my 10D), I found that a color channel might peg out before the luminance histogram or the flashies would reflect the problem. This occurred with relatively more saturated highlights, such as a bright blue sky or a bright red setting sun, necessitating maybe 2/3 stop underexposure via the ETTR method. Again, this is only as reflected by a luminance histogram, not an RGB histogram. Finally, I found there is a whole lot of additional information to the left of the displayed histogram, available through a raw converter. You can see more of this information by dropping the contrast as far as possible.</p>

<p>A fascinating question and still one difficult to answer. I can speak from the film era. Spent many years learning the zone systems and attempting to apply the same principles to color film, primarily slide film. Not always easily done. I have spent time using a hand held meter with digital cameras. In specific situations, I achieved my desired results, but after going back to practice even more, I found that I could have gotten the same results with a greater understanding of both the digital image I was going to end up working with (i.e. 12 bit versus 14+ bit), and what I had hoped to achieve in the final image. Clipping shadows or highlights is not only common among many, many photographers find expanding the dynamic range is sometimes very difficult. I have to take my concept of this from my work with film, some of it applies, a lot does not apply, so I'm left dealing with taking an empirical approach with my subject matter. For me, with digital, bracketing works well, and I'll spend time really looking hard at the exposure. I have to be the one to decide where I 'clip' my image, i.e. what's more important, the shadow detail, the detail in the highlights, or something of an overall balance. This, to me, applies to each one's eye and to the subject matter as well as the desired effect of the final image.</p>

Thanks for the citation, Jeff. I think Bruce Rubenstein did a good job explaining the mathematical fallacies in the article, which is in other respects very useful and even-handed.

 

Mark, I find that when bracketing without a tripod, the third picture is usually blurred or framed differently. Is this me, or my camera doing vibration reduction? I don't know. It's maddening, because the third (-EV) is usually the one I want! Advice on holding steady would be welcome.

<p><em>You have to learn when to underexpose (beach, snow, whitewater) and when to overexpose (white subjects in bright sunlight, light subjects in dim shade). Or use your spot meter.</em><br /> Confused thinking here - you need to give (about 2 stops) MORE exposure than an indicated integrating reading (which you can call "overexposing" if you wish) to predominantly light-colored subjects (such as beach, snow, whitewater) and give less to predominantly dark subjects (e.g. a black cat on a pile of coal). It is totally irrelevant whether the subject is in open sunshine or shade - obviously the lighting level is higher in bright sunshine and so is the contrast, but the subject tonality (light/medium/dark) is exactly the same.<br>

In response to the OP's question - think of clipping as going off the scale at the high end. Highlight detail needs the most attention during digital capture (and also capture on positive film) because there is no real way of restoring highlight detail not recorded during capture (as another poster observes, there is a limited facility for this with RAW images but this does not work very well).</p>

<p>I think I should have rephrased my question. Is there more information to recover in a 1 stop overexposed highlight or a 1 stop underexposed shadow?</p>