The “spike” on the right side of the histogram of the overexposed scene indicates over-exposed areas that cannot be totally recovered in post processing. (These over-exposed highlights would also be indicated by “blinkies” on your LCD image review if you had this feature selected.) You should retake the picture using negative exposure compensation to recover these highlights, and then adjust the overall exposure of the image in post processing. Rules of thumb for adjusting exposure using the histogram: • Use the RGB histogram. The “composite” histogram and flashing highlights warning won’t necessarily warn you when you are clipping a single color channel. • The histogram should just touch the right side of the scale. This maximizes the quality of the image data without clipping the highlights If needed, lower your exposure to preserve the highlights. (Don’t clip the image data on the right of the histogram. Remember that +2 EV is equivalent to Zone VII, the lightest areas in a scene that retain visible detail) The Rest of the Story The above description of the histogram was an oversimplification (but a good starting point for understanding histograms). What I described (“relative distribution of tones in an image, dark tones to the left, bright tones to the right”) is a luminance histogram. (Or a histogram of luminosity to match the terminology in Photoshop.) (http://spiritofphotography.com/coppermine/albums/userpics/10001/luminosity_histogram.jpg) Take a look at the histogram above and tell me what you see? Is the image data “clipping?” Before I give away the answer, I want to note that this is a luminance histogram (distribution of dark and light tones), which is one of the options for display in many image editing programs. Because they are intended to reflect the way we perceive light, luminance histograms are heavily weighted towards the green channel. (So much so that some photographers were convinced the camera manufacturers were “cheating” and only displaying the green channel histogram, masquerading as a luminance histogram.) The luminance histogram shows the “lightness” value of each pixel in the image, calculated based on the relative brightness values of the red, green and blue components of that pixel. Because of this, luminance histograms are not always good indicators of when you when you are clipping a color channel (or two). You could have a very high value for one color channel (a saturated color), offset by low values for the other channels, and thus not show a high enough combined luminance value to indicate clipping. Compare this representation to the RGB histogram below. An RGB histogram is a cumulative view of the tonal distribution of red, green and blue pixels in the image. It helps to think of this as if the camera were first producing individual histograms for the red, green and blue channels, and then adding them together. Like this: (http://spiritofphotography.com/coppermine/albums/userpics/10001/RGB_histogram%2C_all_channels.jpg) This is an RGB histogram for the same image as the luminosity histogram displayed above. Obviously the RGB histogram is a better option to use for determining if any color channels are clipped (overexposed) in your image. As a point of clarification, the “composite” histogram displayed by your camera (usually shown in yellow or white) is usually a luminance histogram. Some cameras label the composite histogram as RGB, when in fact it is not. Here’s the image associated with that histogram (Please ignore the lack of artistry, this was taken for example purposes only): (http://spiritofphotography.com/coppermine/albums/userpics/10001/yellow_rose_example.jpg) You can see the “blown out” areas on the yellow petals where the red and green color channels are clipped. Seeing the red and green channel histograms associated with this image on your camera LCD would provide enough warning that you are “clipping” image data and need to adjust your exposure. That’s one of the reasons I recommend using your RGB histogram. (With caveats, as we will talk about later.) The example image above was taken at the “default” camera settings using matrix metering and Aperture Priority programmed exposure. Comparing the exposure to either a Zone System based exposure or an exposure based on examination of the red channel of an RGB histogram, this image was 1 EV overexposed. None of the raw imaging programs I tested were able to fully recover all the color or detail from the blown areas of the petals. (Close, but not quite) I took a second image of the same scene with 1/3 stop less exposure. The histogram for the second image showed that the red channel was still blown, but the green channel was not. I was able to recover the second image in post processing by reducing the exposure 0.7 EV. This leads me to two points. First, most raw processing software can “recover” lost highlights as long as only one color channel is clipped. They do this by “interpolating” the data based on the remaining two color channels. (Personally, I've never been quite happy with images that have "recovered" highlights (from a clipped channel). They just don't look quite right to me, with the color balance just slightly off, and missing tonal transitions in the areas that have been interpolated.) The second point is a bit more complex, and has to do with the spectral sensitivity of your sensor and how white balance coefficients are applied. Almost all modern camera sensors are much more sensitive to green light than they are to red or blue light (due to the RGB filters in front of the sensor). Camera makers compensate for this by using a “multiplier” to bring the signals from the red and blue channels back into line with the green to recombine the image data into something resembling reality. In the process, they also adjust these “multipliers” based on the white balance of the light on the scene. This helps compensate for the color cast from warm or cool light on the subject. For example, with the D2x, the multipliers for “daylight” (5200K) are 1.52 for the red channel and 1.69 for the blue channel. This means the in-camera image processing engine will multiply the red channel output from the sensor by 1.52 and the blue channel output by 1.69 before combining them with the green channel to produce the final image. So where am I going with this and why does it matter? It is important to understand that the histogram display is based on the in-camera JPEG processed for your preview image. Whatever tone curve, white balance and color space settings you have in camera will affect how that JPEG, and associated histogram are displayed. What you are seeing on the RGB histogram display is a representation of the red and blue channels after the white balance multipliers have been applied. In other words, even if the histogram shows the red channel clipping, it’s probably only clipping the “amplified” red channel data (not the “raw” data), and can be recovered in post processing if the red channel levels are reduced. The only channel that doesn’t have a multiplier applied is the green channel; therefore, if you see any indication that the green channel is clipping in the RGB histogram, you should reduce your exposure accordingly. The previous statement applies for daylight white balance. For very warm incandescent lighting, neither the red channel or green channels have multipliers; however the blue channel multiplier is increased to 3.98 (for the D2X, 3.0 for the D3). The point is that you should be careful about clipping the red channel or green channel when shooting under incandescent light, and don’t worry too much about clipping the blue channel. By now you should have made the correlation that if you are shooting in warm light (incandescent) the blue channel will be the least sensitive and will have a multiplier applied in order to achieve proper white balance. As you shift up the scale to cooler light (“shade” at 8000K) the red channel will be the least sensitive and will have the largest multiplier applied. In “direct sunlight” (the light conditions where we are usually the most likely to “clip” color channels), the red channel (for the D3) has a multiplier of 1.82, while the blue channel has a multiplier of 1.36. Given that information, let’s take another look at the 2nd yellow rose exposure (red channel only clipping) and see what happens when we change the white balance. Remember that for this image I had to reduce the exposure by 0.7 EV (or 2/3 of a stop) in my raw processing software in order to prevent clipping in the red channel. Now, with the exposure set at “0” on the scale, when I change the white balance to incandescent (3000K), ALL of the red channel clipping is recovered! (The "multiplier" or white balance coefficient for the red channel is lower in incandescent light, which reduces the "amplification" of the red channel enough that it no longer appears clipped.) (http://spiritofphotography.com/coppermine/albums/userpics/10001/Untitled-1.jpg) This tells me that in “direct sunlight” I have at least 2/3 stop of exposure latitude in the red channel before the raw data (without the multiplier applied) is clipped. This also tells me that if I have a high contrast scene and I want to maximize the dynamic range captured in my image data, I probably don't want to apply exposure compensation in the field in order to “recover” the red channel just because the RGB histogram implies that it is clipping. Now that I’ve examined the data, I know the “clipping” is due to the in-camera or software applied multiplier, and that I’m not really clipping the raw data. (Rough analysis indicates that I also have about ½ stop exposure latitude in the blue channel (in direct sunlight) due to the white balance multiplier.) Based on this information, my recommendation is to think twice before reducing your exposure just because you see the red or blue channels “clipping” in your RGB histogram. Some photographers even go so far as to load a special white balance preset into their cameras (called UniWB) that sets the white balance coefficients (multipliers) for the red and blue channels to "1" so that they can see a histogram of the "real" raw data and know when they are clipping one of the color channels. I think this is a little extreme (your image white balance will be way off, and requires correction in post processing) and I am comfortable knowing that if the composite luminance histogram is indicating clipping, or the green channel is clipping, then I should reduce my exposure accordingly. You should trust your meter (if you are using manual spot metering and the Zone System) more than you should trust the histogram. However, you can make your histogram match the raw data as close as possible by adjusting your camera settings to use the largest color space (adobeRGB) and the lowest contrast tone curve ("neutral"). I also know, based on my testing and analysis of the raw data, that I can come very close to "zeroing out" the white balance coefficients (multipliers) by setting a white balance of 4760K with a hue adjustment of G6. (CAUTION: These adjustments are specific to the D3 and D700 based on the spectral properties of the sensor and Bayer filters. The numbers will be different for other sensors; however, the concept remains the same.) In this case, the multipliers for the red and blue channel are both 1.09, close enough to "1" for practical evaluation of the raw data in the field. This enables me to see a fairly accurate representation of the raw data color channel histograms with very little inaccuracy induced by the multipliers. Because of the extra step involved in post processing the data (correcting the white balance and final exposure) I will only use this technique when photographing very high contrast scenes where I want to maximize the dynamic range in the scene. Before we leave the topic of histograms, I should answer my question in the beginning where I ask you to take a look at the luminance histogram and tell me if I’m clipping any image data. With all we have discussed above, you should now be able to answer “yes” without any hesitation (since you saw that the green channel was clipping). Another “hint” in the luminance histogram is the sharp peak close to the right edge of the histogram. Anytime you see a sharp peak like this, you should suspect that one of your color channels is clipping and take a closer look at the RGB histogram (or bracket your exposures to be safe). Because this has been such a long discussion, I should probably summarize the key points. 1) Set your in-camera settings to adobeRGB color space and a "neutral" tone curve. Selecting a smaller color space or higher contrast tone curve will indicate clipping prematurely in your histograms. 2) Recognize that your red and blue channel histograms can change drastically based on the white balance coefficients (multipliers) being appliled, and that these "multipliers" result in the red and blue channel histograms not being a very good indicator of when the "raw" data is clipping. 3) You typically have 0.7 EV or stops of exposure lattitude in the red channel in "direct sunlight," so don't be too quick to reduce your exposure to prevent clipping, especially if you are trying to maximize the dynamic range captured in your image file. 4) Watch the green channel. This channel doesn't have a "multiplier" applied in-camera, and so "clipping" on the green channel histogram is real. Reduce your exposures accordingly. My final recommendation is to learn how to use your spot meter and the Zone System to control your exposures. Mastery of those tools and techniques will eliminate any concerns with "clipping" your image data. The histogram will naturally fall into place as a "sanity check" that your Zone System based exposures were correct. Keith |