HISTOGRAMS for Dummies & Experts

Earlier this week I was teaching at Reducation in Long Beach and I asked the class how many people had shot RAW on a DSLR or a Digital Cinema Camera and almost the entire class raised their hands. But then later in the day, I asked if anyone in the class (36 students) could explain what a histogram was technically representing - and only two people raised their hands raised. Even if more people actually knew, clearly they were not confident enough with what they thought was the correct answer - and this, IMO is a problem. If you really want to leverage a technilogical tool - you need to be confident in that tool and you need to understand what is technically going on.  So ... I think it's time for a quick "for Dummies & Experts" post on the Histo.

What is a Histogram?

The histogram is a representation of the number of pixels at a specific brightness indicated by the height of the line scaling from dark pixels on the left (shadows) through all the brightness levels (mid-tones) to bright pixels on the right (highlights).

The histogram is usually computed after color interpolation and the gamma/ISO curve is applied to the image so it represents what appears on the monitor.

When you change your ISO - the histogram will change it's representation - even though what the camera records will be the same.

On Scarlet/Epic-MX/Epic Dragon - when in RAW view - the cameras monitor paths and histogram are now* locked to ISO 800. RED could (and hopefully will) optimize how the histogram is mapped when in RAW view as the RAW view's purpose is to try to give the most unprocessed representation of what the camera is capturing - which is a bit of a challenge since the camera is capturing and recording linear light.

* In DSMC firmware builds v5.1.47 and later - RED has re-calibrated the RAW view mode to RLF (REDlogfilm), RC2 (REDcolor2), 5600K, 800 ISO, 0 TINT. 


Can you describe it more technically?

To generate a histogram the camera looks at each pixel in the image and increases a counter in a list from 0  (Black) to 4096 (bright) corresponding to the brightness.

The values in this list are then normalized by the maximum value to the display height and width and drawn from dark (left) to bright (right) using the value in the list to draw the height of the line.


Huh? Gimme the "for Dummies" version?

A histogram is a representation of the distribution of color and tonality in an image.

Why is it Red/Green/Blue?

In an RGB histogram it’s broken down into R, G and B components (Represented by a red, green and blue curve) because each pixel has a red, green and blue brightness, which makes up its color.

In a monochrome histogram only the average brightness for each pixel will be used. You can select a monochrome (Luma) histogram on the Scarlet/Epic/Dragon - Menu>Settings>Display>Modes - "Histogram: Luma"

What are the narrow red lines on both sides of the histogram?

Additionally to the histogram RED counts the pixels that are either completely black or completely white (Clipped at the top or the bottom) and show them as bars representing the percentage of total pixels clipped for the whole image. These are known as the "GOAL POSTS".

I'm still kinda confused … Can you show some examples?

Histogram of a grey chart. Because it’s the same brightness over the whole image with the same color it will show up as a narrow spike.

Grey chart with different grey levels. As there are two primary brightness levels here we see two bumps in the histogram. They are wider then before due to the different grey scales.

Very dark image. The bumps in the histogram have moved to the left. We also see a few clipped pixels at the bottom.

Very bright image. The bumps have moved to the right and we see clipped data at the top. (notice the red in the GOAL POST on the right)


Reddish image. The red bumps are brighter then the green and blue bumps. (Making red the more visible color)

Bluish image. The blue bumps are brighter then the green and red bumps. (Making blue the more visible color)

Histogram of a colored scene. As there are only a few distinct colors in this image the show up as spikes in the histogram.


Great … So what can I use this for?

Exposure

Monitoring the histogram while setting the exposure for a scene helps the operator to understand the tonal distribution and dynamic range of a scene.

Exposure falls into two categories: Either the scenes dynamic range does not exceed the sensors dynamic range (or in other words the histogram shows that there is room on the left and/or right of the data in the image) or the scene’s dynamic range exceeds the sensors dynamic range (which in the histogram will appear as data up to and exceeding the edge which will lead to clipping indicated by the red bars - aka GOAL POSTS).

If the scene exceeds the dynamic range of the sensor it is up to the operator to either add lights to the scene to reduce the dynamic range or make a choice if it is preferable to clip the highlights or the lowlights. Often the best choice is to clip extreme highlights at the top (Specular reflections, lights) and if still needed clip some of the darker areas of the image. Other tools in the camera (the EXPOSURE tool or the false color VIDEO tool) will help identifying the areas that are too dark or too bright.

If the histogram is only partially filled it is usually best to center the data as best as possible in the mid-tone range of the histogram (And raise contrast in post processing to get a more balanced result.)

Having said that, it might not always be the best strategy to just center the histogram. A very bright scene on the beach or with snow should probably be exposed to look bright while a dark scene at night needs to appear dark to look correct. (Note that this can either be archived during exposure or in post production by adjusting the scene brightness through gain - ISO/FLUT/CURVES, etc.)

Contrast

How much of the histogram is filled with data determines the contrast of an image. Auto contrast algorithms usually determine the lowest visible value of the picture from the histogram and subtract that value from all pixels. (For example - try loading a REDlogfilm TIFF into Photoshop and hitting "command - L" and hitting the Auto Levels and see what happens.) After they determine the brightest value and scale all values in the image so that the brightest value aligns with the top of the histogram. (Also known as histogram equalization). Of course this can also be done with manual controls (Curves, Shadow/Gain controls)

White Balance

In a predominantly grey-tone scene white balancing can be judged from the histogram. The process of white balancing tries to remove any color cast introduced by colored light so that white/grey areas appear without that color cast in the final image. 

In a camera the histogram can be used as an indicator that the white balancing/ color temperature might be off and needs to be adjusted.
(Note that in the final grading of the footage a perfect white balance might not be what is desired.)

More In-Depth Info:

http://www.red.com/learn/red-101/red-camera-exposure-tools
http://www.cambridgeincolour.com/tutorials/histograms1.htm
http://www.luminous-landscape.com/tutorials/understanding-series/understanding-histograms.shtml
http://photovian.com/learning-center/histograms-easy-important/