Reading Histograms

[Brooks Ayola]

The 16 vs. 8 thread (Big curve moves in 16 bit) has made me think about the following question. What does a good histogram look like and how is this information best used? It would be great if some of the histogram buffs around here could share some of their wisdom on histograms and what to look for. I realize that it depends on the image and the desired effect, but it would be great to see some examples of images next to screen shots of their histograms to illustrate what's good and what's bad and why. I would also like to know if the same info could be used to edit images in camera based on the histograms provided by many digital cameras like the canon 1Ds. Do histograms ever lie?

[sPECtre]

Usually , one look if there is no "peak" at the very beggining or the end of the histogram. It means, AFAIK that there is some clipping, and some tones might be lost. One tries also usually to have a "full"histogram, that goes from the beginning to the end.
( I read in REAL WORLD PHOTOSHOP that after all retouching, one has to set the Black and White point in order to ge to press. It's an excellent book I recommend, Brooks!)

In fact, I had this very same thing in mind recently:
When is a histogram "bad"?

Sometimes, it looks like a comb, but the pictures print right on a desktop printer. I think , If I remember correctly that the offset/press output is way more sensible to that...

[Kevin Connery]

The contents of a histogram reflect what was captured, or is in the file.

What was captured or is in the file reflects what was in the scene and how it was exposed.

A high-key image, even properly exposed, may well have a significant fraction of the image at 255 (See the image and histogram with the white background), while a histogram like that could easily mean overexposure for a 'normal' or low-key image.

The opposite applies for a low-key image, where a lot of image data could be legitimately at 0. (This sample doesn't have a black background, but it does have a lot of black-without-detail.)

That said, most normal images with a lot of pixels at 0 or 255, or a lot of pixels crowded near one edge of the histogram and nothing or very little near the other, usually are improperly exposed or scanned.

If you keep in mind that ANYTHING at 0 or 255 is without detail, that'll help. There aren't that many other 'rules' that are universal. (That I know of, and I like that kind of information!)

http://www.keradwc.com/outgoing_images/highkey_histogram.jpg

http://www.keradwc.com/outgoing_images/lowkey_histogram.jpg

[sPECtre]

thanks for the heads up, Kevin!

I found these pages about histograms : http://www.marginalsoftware.com/HowtoScan/tutorial_page_1.htm

[Ammar Midani]

In addiotion to what Kevin said here’s a quick run down on histograms:
The histogram is a graphic view on how the pixels in images are distributing. (image data).
It is a common knowledge that before you go and do a tonal retouch to take a look on the histogram of the image. The histogram seeing is divided to three areas”
- Left = Shadows
- Center = Midtones
- Right = Highlights

One look at the histogram could tell you in what key is the image as Kevin said if most of the image data is on the right side it means that most of the pixels are on the highlight end and the image is a high-key.
Same with the data at the left side pixels are in the shadows area and it’s a low-key image.
And for those images pixels are in the center they are average-key images.

A tonally correct image got to have image data (pixels) distributed at the same level in all levels of the histogram seeing (shadows, midtones & highlights).

Another common action is to move the side sliders in the Levels histogram to where the image data actually begins, the thing that automatically will distribute the midtones pixels.

For an illustration see help file> Checking scan quality and tonal range (Photoshop)


That’s a gorgeous screen shots Kevin.

Ammar

[Bruce Fraser]

Histograms really just tell you some things about the state of your image -- I don't think it's helpful to think in terms of "good histograms" or "bad histograms."

The most important thing the histogram tells you is whether or not the image is being clipped. A peak in the 0s or 255s is usually something you want to avoid, especially at the capture stage.

Bear in mind that different apps treat composite histograms in different ways, some showing luminosity, some showing peak values per channel. Photoshop does both -- the histogram obtained from the Histogram command shows luminosity, while the one in Levels shows peak values. So while 0/255 peaks in the Histogram histogram always show blacks and whites, the ones in Levels could also indicate total saturation in one or more channels.

Pretty much any edited 8-bit/channel image will show combing in the histogram. It's not necessarily anything to worry about. It does provide a rough index as to how much editing headroom you have left -- if the histogram is already severely combed, with gaps of 2 levels or more, you probably shouldn't contemplate a major tonal adjustment.

Similarly, histograms can help in comparing different ways to achieve the same visual end -- in Photoshop there are always five ways to do anything! If edit a produces a very similar result to edit b, but producesless combing in the histogram, I'll generally go with edit a, not because I'm attached to making pretty histograms, but simply because I know that by doing so I'm leaving more options open for the future...

My $0.02.

[Ammar Midani]

ohh i'm sooo :]
Well Brooks sure you're a lucky lucky one to get your answer from Mr. Fraser.

Welcome abraod Guru nice to have in Greg's place.

Ammar

[Brooks Ayola]

Thanks everyone. I also want to thank Bruce and let him know that I learned a lot from him at the DIFP seminar here in Los Angeles a few months back. So, I understand what the ends of the histogram represent, but what exactly does the vertical scale of the histogram represent? Is it the amount of pixels that fall into that exact tonal range? What about a peak that goes off the top of the scale, is this just the way the histogram window crops the visual data? I'm assuming that this does not a representation of any kind of "clipping" in the image. As in Kevin's Low Key image, the left side shows that there is clipping of the blacks, but the fact that the data goes off the top is not important. Correct?

[Bruce Fraser]

exactly correct. the vertical scale is arbitrary and roughly proportional, so the taller the spike, the more pixels are at that level, but there's no absolute scale, and nothing gets clipped off the top

[Mindbender]

Yeah, summer photoshop allergies are a pain, I recommend a good anti-histogram ;) Sorry... couldn't help it. hehe

I can't really add much that the experts haven't already explained. Basically when I first started working with histograms the way that it made sense to me is to think about it like a tally sheet. Like you're counting something. Ever play those silly road trip games as a kid? Count how many red cars go by and the other person counts the blue ones? Image it like that. Each time a blue pixel goes by, add one to the blue collumn. Each time a red pixel goes by, add one to the red collumn. Now just do that for 256 ps levels in each channel based on luminocity. hehe

$0.02

[sPECtre]

Another explanation of histograms (in general) is available at :
http://www.luminous-landscape.com/tutorials/understanding-series/understanding-histograms.shtml