Some things about histograms I don't fully understand

For the most part, I understand a histogram, what it is, what it does, and what it shows. For the most part.

I have watched just about every youtube video about histograms out there (how exciting), but I still have questions.

Below are the 6 Photoshop histograms from the photo below it (boring, but it shows what I want).


My understanding of each is as follows:

RGB Histogram - Thought it was an all-in-one view of each of the additive colors (R/G/B) distribution of pixels, but it doesn't look like it is.

Red Histogram - Shows the red channel's distribution of pixels
Green Histogram - Shows the green channel's distribution of pixels
Blue Histogram - Shows the blue channel's distribution of pixels

Luminosity Histogram - If my understanding is correct, is used to confirm exposure/exposure levels of the 5 sections. In most basic form, you're looking for a bell curve representation of the pixels.

Colors Histogram - This one I'm not fully sure on. It's obviously the RGB channels overlayed onto each other, but it seems to also show the pixel quantity of the combination of the additive colors (RGB, with this showing the CYM), and then there is the dark grey along the bottom.

My questions are -

1. What are the grey pixels representing along the bottom of the Colors Histogram?
2. What does it mean (in any of these histograms) when a color (or even a luminosity) goes above the visible chart?
3. Using this example, I interpret it as the red channel is blown, because the pixel representation is climbing the right side, although the image itself isn't appears to be exposed correctly (according to the luminosity histogram). How do you correct for this without changing the image's coloring, or do you leave it alone?

I'm sure new questions will pop-up, but for now, those are my starting points.

Here's a pretty good article:

http://photo.tutsplus.​com …dobe-photoshop-lightroom/

To try and answer your questions:

I believe the grey pixels represent the combination or average. It actually shows a bit odd in the graphs that you posted, but when I view images in LR the grey portions generally cover the majority of the individual color channels.

I believe this occurs when individual colors are clipped. The height represents the amount of a particular color.

That said, as you stated, the red channel that is clipped in the highlights (right side of the histogram).

What I've had to convince myself of... is that 'perfect' histogram doesn't mean good lighting or even a great exposure. It's simply a visible method (graph) to represent data (color information). There are plenty of successful photographers who will underexpose and clip blacks, just as much as they will overexpose and clip highlights - and they do it on purpose!

In question #2, I'm referring to when the peak goes above the histogram window/box. When it hits/climbs the right side you've clipped highlights/whites (colors?), and on the left side you've crushed the shadows/blacks (colors?).

It's the peaking I'm not understanding. I almost think it's just that there is so much of that color (whatever color it is, RG or B), that it just can't be shown within the confines of the small box so you'd still be able to see details of the histogram pattern. If that makes sense.

drvnbysound wrote in post #16286844
Here's a pretty good article:

http://photo.tutsplus.​com …dobe-photoshop-lightroom/

That just gives the basics. Good article, but not what I'm after.

Yes. As I understand it, the vertical axis is automatically fit to some scale to display the most information.

If I understand correctly, the vertical axis would be listed in pixels... as to the cameras sensor (e.g. 15MP). So that if you took a picture of pure black, it would be able to show a vertical line that was 15MP in height at the very left of the histogram. However, it's understood that pictures are generally not taken of pure white or black colors, but of subjects and scenes that have variations of color. That being the case, some arbitrary scale is used to best fit those scenarios.

So if a peak goes above the viewable area, all it means is that there are a ton of pixels of that value? And not that it's a bad thing?

1. The grey plot is just that, greys; areas with very low saturation.

2. Not sure... I'd think that the histogram would auto-scale and thus avoid the presumably clipped sections. Will get back to you.

3. Luminance is derived a combination of all channels. The luminance plot can be perfectly happy with a blown-to-heck red channel. Avoid it by chimping the red histogram as you shoot, assuming your camera offers the option.

You can see that the Y-axis doesn't auto scale by looking at the Red channel posted above. By selecting the red channel you are asking the histogram to display ONLY red color information. However, it's clear that the peak is above the height of the graph provided.... heck, look at the RGB and Green histograms as well, which also have values higher than the Y-axis displays.

{Warning: Self reflection below. Nothing below is meant to be directed at any person and it was not written to be offensive in any way, so I hope that it's not taken that way. It's simply myself thinking out loud...}

The histogram... does it really matter? Good luck reading lighting ratios from a histogram and you'll never find emotion or expression there. What about determining proper focus or knowing if an image is sharp? What are the questions that you ask yourself about your pictures? ...the ones for clients, friends, or family. When you view a picture of a client are you asking yourself about the histogram?

I feel like I understand the basics of a histogram, but I almost never look at it. I have the 'blinkies' turned on in my camera so I know when I blow highlights. I also toggle the clipping triangles in Lr to show me what's been clipped, but the reality is that I look at the picture to tell me what I need to know - I don't really care what shape the histogram is.

I'm an engineer by trade and I'd love to be able to say that I know everything. I'm a problem solver and inherently I'm supposed to be able to figure things out. The fact is that I know a little about a lot of different things...

I don't need to know why an inch is the length that it is - someone came up with that a long time ago and they probably had a great reason for it to be that size. But if I want to build a desk or put up a shelf, I don't need to know that... I need to know how to use a tape measure.

At some point I have to draw the line of understanding... or wanting to understand for the sake of measurebating, and realize that I really don't necessarily need to know the answer to everything. What I need are tools that allow me to produce the [Insert thing here] that I'm trying to make... in this case, it's pictures.

The more I read into this thread I asked myself if I should know this stuff already - during the middle of typing this response it clicked, and I thought of the noise... http://www.youtube.com …er_embedded&v=z​65jlY4e908

True.

The equation for figuring Luminance is G=59%, R=30%, B-11%. There is a strong weighting to the green and this can hide clipping in the other channels, especially blue. Its historic origin was in the fact that B/W film, like present day sensors, was most sensitive to green light. However, WB application cancels the high green sensitivity and thus makes the Luminance histogram unreliable.

Absolutely. If you were shooting at the beach and you saw that the histogram is all to the left of center you would know that the photo is underexposed, but you would have seen that already in the image review. But if you shot a blank evenly lit wall in an automatic mode and the histogram was peaked at 40% you would know your camera's automation is working well. I am constantly using the real time histogram in Live View as my light meter but only looking at the right end to get it as close as possible to the max. All the rest of it is immaterial because if I have maximized the brightest pixels all the others will be adjusted in LR.

Thanks Elie!

drvnbysound, to a huge degree, I agree with you. I only use the histogram as a guide when shooting/editing to see if things are (unexpectedly) crushed or blown.

I'm just one of those people that likes to know how it all works, and what it all means. Kind of drives my wife nuts.

KirkS518 wrote in post #16286769
My questions are -

1. What are the grey pixels representing along the bottom of the Colors Histogram?

In the "Colors" histogram display, when the R, G and B histograms overlap, they produce the resulting "mix" of the overlapping colors - so, for example, if G and B curves overlap, they produce C (cyan) - if the R and G curves overlap, yellow and if R and B overlap, magenta. If all three (R, G and B) overlap they produce gray - i.e., equal parts R, G and B. This does not mean that there is any of the mixed colors present in your image, per se.

The Y axis is pixel count - there is no explicit upper bound, but when the particular histogram bin hits the top of the plot, it means "a lot of that pixel" it's not a bad thing. Imagine, however, if you had a lot of one particular value but wanted to see details about the other values across the histogram. If the Y axis scaled to the value of the greatest bin, then you would not be able to see any of the other bins' details. It would be like plotting a bar graph of 10 people and their age - if one person was 102 years old and the rest were infants of various ages between birth and 1 year of age, you would not be able to tell the difference among all the infants if the Y axis (age) was scaled to 105 years of age (max axis value) to depict the 102-year-old person's age.

The red channel is likely fully saturated and clipped due to the content of the image. Your camera's meter "sees" the reflected light from the scene a lot like the Luminosity histogram shows you, so the exposure was "correct" for the average gray value of the scene (or spot, or however you metered). But, in this case, there are strong bright reds and magentas in the scene and, if you want to capture them accurately, you have to consider these scene-specific elements in your exposure determination. To see if your camera actually captured the red channel fully, and if you shot raw, you can use a raw analysis tool like Raw Digger. In such an application, you will able to see if the R channel of the sensor reached saturation (actual raw data clipping) - If it did not, then the clipping in the RGB image is due to the conversion of the raw file or the use of an RGB color space that could not contain the R data fully.

Some applications can reconstruct these highlight values if the other two channels are intact, but color reproduction usually suffers because the recovery is for luminosity reconstruction. These algorithms do not, of course, know what the clipped values actually were, but they can often reconstruct the luminosity and then propagate a guesstimate of the color that those pixels were into the reconstructed pixels.

The histogram is an extremely useful tool if you are interested in image processing and automation. Sure, we can all edit "by eye" and achieve something we like. But, if there are aspects to image editing that can benefit from analyzing the image content, then the histogram is a nice way of visualizing and implementing these techniques. For example, as of PSCS6, if you use Auto in "Enhance Brightness and Contrast" mode (auto levels, auto curves, auto brightness-contrast) the auto algorithm does not simply bring in the black and white points to the edge of the histogram, within the tolerance you specify - it compares the histogram of YOUR image to a database of image histograms and applies an automatic correction to YOUR image that was deemed optimal for the database image. For auto curves, it will apply a multi-point curve, not simply a straight-line black-white point adjustment. That is pretty slick and may provide a starting point that would have taken minutes of "by eye" correction to achieve.

kirk

The issue with the Y axis of the histogram charts is that it actually uses a linear scale. If the scale was set to show the full range of the highest peaks then you would probably see very little detail in the rest of the graph. With an 18 Mpix sensor you could quite probably have the peak measuring over 1 million pixels. The average sizes of the values would probably range from a few tens of thousands of pixels to maybe a couple of hundred thousand pixels. The scale seems to be picked to show the most detail.

Of course the best thing to do would be to use a Logarithmic scale on the Y axis. This allows you to see detail for levels that have just one pixel to levels with several million pixels all on the same axis. I guess the problem is that these days, as we no longer have to use slide rules for multiplication very few people now understand logarithms and logarithmic scales how useful they can be. Thus they would be baffled by the strange scale.

Alan

BigAl007 wrote in post #16287692
The issue with the Y axis of the histogram charts is that it actually uses a linear scale. If the scale was set to show the full range of the highest peaks then you would probably see very little detail in the rest of the graph. With an 18 Mpix sensor you could quite probably have the peak measuring over 1 million pixels. The average sizes of the values would probably range from a few tens of thousands of pixels to maybe a couple of hundred thousand pixels. The scale seems to be picked to show the most detail.

Of course the best thing to do would be to use a Logarithmic scale on the Y axis. This allows you to see detail for levels that have just one pixel to levels with several million pixels all on the same axis. I guess the problem is that these days, as we no longer have to use slide rules for multiplication very few people now understand logarithms and logarithmic scales how useful they can be. Thus they would be baffled by the strange scale.

Alan

Agreed - some applications permit one to switch between linear and log.