Analyzing Color Balance
Most of the time, you’ll be able to spot inaccurate color balance visually, simply by looking at your calibrated display. For example, a tungsten-lit scene will look orange when you’re using film stock that is balanced for daylight or a video camera with its white balance set to daylight.
Aside from the obvious color cast, orange light from incandescent fixtures may lend an inadvertently theatrical look because of the viewer’s association with artificial lighting. For example, the image on the left in Figure 4.23 is incorrectly balanced for daylight, and the tungsten lighting lends a warm, orange cast to it. The image on the right is properly white balanced, with whiter highlights and truer colors throughout the scene (note the blue sunlight spill in the foreground).
Figure 4.23 On the left, a tungsten-lit scene with incorrect color balance; on the right, the same scene with correct color balance.
Similarly, a daylight scene shot using tungsten-balanced film stock or a video camera with its white balance set to tungsten/indoors will look bluish (Figure 4.24).
Figure 4.24 On the left, a daylight-lit scene with incorrect color balance; on the right, the same scene white-balanced correctly.
If the filmmaker was not intending to portray a cold winter day, this is clearly a shot that would benefit from correction. Compare the image on the left in Figure 4.24, which is incorrectly balanced for tungsten, to the properly white-balanced image on the right.
Using the Vectorscope
The vectorscope measures the overall range of hue and saturation within an image. Measurements are relative to a graticule that’s overlaid on the scope, which provides a frame of reference via crosshairs, diagonal I and Q bars, and labeled color targets corresponding to 75 percent saturated primary and secondary hues. Figure 4.25 shows all of these indicators relative to the color wheel that represents the reproducible range of color and saturation.
Figure 4.25 An idealized NTSC vectorscope graticule, showing all the crosshairs and targets you might expect to use to measure a displayed graph, superimposed over a color wheel showing their approximate correspondence to hue and saturation. Typically HD vectorscopes don’t have as many reference items.
Figure 4.25 should clearly illustrate that hue is indicated by the location of a graph trace’s angle around the center, and saturation is indicated by a trace’s distance from the center.
In reality, the graticules of most software vectorscopes are considerably simpler. At the least, a vectorscope should have the following graticule elements:
Primary and secondary color targets that correspond to the top row of bars on the SMPTE color bars test pattern (Figure 4.26).
Figure 4.26 Portions of the SMPTE test pattern that correspond to vectorscope graticule elements are called out.
- Crosshairs that indicate the desaturated center of the vectorscope graph.
- I and Q diagonal crosshairs (and their –I and –Q counterparts). These stand for In-phase and Quadrature (an amplitude modulated phase 90 degrees relative to In-phase), which correspond to the purple and cyan/blue patches at the bottom of the color bars signal.
- Tic marks along the I- and Q-bars correspond to the voltage waveform that would be traced by the discrete I and Q components, while tic marks running along the outside border note 10-degree increments.
When it comes to graticules, most vectorscopes have some manner of centered crosshairs at the center, which are critical for providing a reference of neutral black, gray, and white in the signal. The “I-bar” (as I’ve come to call it) is optional, and opinions vary as to whether it truly belongs on an HD scope. I happen to think it’s still a useful reference, as I discuss in Chapter 8.
Different software scopes display different graticule elements and also draw the vectorscope graphs differently. Some software scopes represent the analyzed data as a discrete point of data on the graph, while others emulate the CRT method of drawing traces corresponding to each line of video that connect these points together. These traces aren’t necessarily adding any actual data to the graph, but they make it easier to see the different points, and so they can be easier to read. Figure 4.27 illustrates the differences in three commonly used vectorscopes.
Figure 4.27 Three excellent examples of different software vectorscopes compared (left to right): DaVinci Resolve, Autodesk Smoke, and Divergent Media ScopeBox (showing the optional Hue Vectors graticule that I designed).
DaVinci Resolve has a traditional vectorscope, the graph of which emulates a trace-drawn graph, with 75 percent color bar targets and an In-phase reference line. Autodesk Smoke has a unique vectorscope graph option that averages analyzed color as a scatter graph that consists of differently sized dots representing the amount of color at that position, which makes it really easy to read and calls attention to the outer boundary of signal that light traces might not make apparent. Smoke draws both crosshairs and 75 percent targets.
The third vectorscope shown, Divergent Media’s ScopeBox, has a more traditional graticule available, with a trace-drawn graph, but it’s also a forward-looking application that was the first software scope to incorporate the Hue Vector graticule I designed, which presents lines that are aligned with each of the primary and secondary colors to help give colorists reference points for comparison, a center crosshair that’s aligned with the warm/cool axis of naturalistic color temperature for lighting, an In-phase positioned reference line, a user-customizable reference line, and both 75 percent and 100 percent tic marks for color intensity. ScopeBox also has a peak option for the vectorscope, which shows an absolute representation of the outer boundaries of the signal, making it easy to spot signal excursions that can be hard to see with faint traces. In fact, you may notice that the peak outline shape matches the scatter graph of the Smoke vectorscope.
Judging Color Balance Using a Vectorscope
Since the center of the vectorscope graph represents all desaturated, neutral values, it follows that if a graph is uncentered and the image is supposed to have neutral tones in it, a color cast is present.
In Figure 4.28, the vectorscope graph to the left is suspiciously lopsided, leaning heavily toward yellow-green. This may not necessarily be wrong, but it should at least cause you to look at the source image a bit more closely to make sure this makes sense.
Figure 4.28 Comparing an off-center graph (left) and an image with a centered graph and image (right).
The vectorscope graph to the right corresponds to a neutral version of the same image. Notice how this graph is much more evenly balanced relative to the center crosshairs of the graticule, with arms stretching more prominently toward several different hues. Again, this is no guarantee that the color balance is correct, but it’s a pretty good indication that you’re in the right ballpark if the image on your broadcast display looks right.
Judging Saturation Using the Vectorscope
Judging the relative amount of saturation of an image is easy, since more saturated values extend farther away from the center of the scope than do less saturated values. In the following low-saturation image, the vectorscope graph is small, hugging the very center of the vectorscope graticule (Figure 4.29).
Figure 4.29 A low-saturation image with a correspondingly small vectorscope graph.
Take a close look at the graph. There are in fact excursions (parts of the graph that extend in various directions) that stretch toward the R(ed) and B(lue) targets, but they’re small, indicating that while there is color within the image, there’s not very much.
Most vectorscopes have the option to zoom into the graph, allowing you to see the shape of the graph with more clarity, even if the image is relatively desaturated (Figure 4.30).
Figure 4.30 Zooming into the vectorscope graph from Figure 4.29 makes it easier to see more detail in the graph of an image with low saturation.
The high-saturation image in Figure 4.31 yields a much larger vectorscope graph, with arms stretching out toward the various color targets that correspond to each hue.
Figure 4.31 A highly saturated image with a correspondingly large vectorscope graph stretching farther out toward the edge of the graticule.
In the more highly saturated image in Figure 4.31, notice how the abundance of red reads as an arm of the graph that extends toward the R(ed) target, while the blues in the man’s clothing appear as another arm of the graph that extends toward the B(lue) target. An abundance of yellow and orange creates a cloud in the vectorscope graph stretching toward the Yl (yellow) target. Finally, two conspicuous gaps in the graph, in the direction of the G(reen) and Mg (magenta) targets, tell us that there’s very little of either of these two hues present in the image.
Using the RGB Parade Scope
The parade scope shows separate waveforms analyzing the strength of the R, G, and B components of the video signal. This is a composite representation, even if the original video is Y’CBCR-encoded. By showing a comparison of the intensity of the red, green, and blue components of the image, the parade scope makes it so you can detect and compare imbalances in the highlights (the top of the graph), shadows (the bottom of the graph), and midtones for the purposes of identifying color casts and performing scene-by-scene correction.
Recall that the whitest highlights and darkest blacks of an image are nearly always desaturated. With that in mind, red, green, and blue waveforms with tops at or near 100 percent/IRE and bottoms at or near 0 percent/IRE should typically align very closely.
In Figure 4.32, we can see that the lighting outside the window is a cool blue, the lighting on the wall behind the woman is fairly neutral, and the shadows are deep and black.
Figure 4.32 An evening scene for analysis.
Each feature can be seen within the parade scope, and the relative height of the corresponding graphs indicates the color balance within that zone of image tonality. For example, the blue window can be seen in the elevated spike at the left of the blue waveform (Figure 4.33). The woman’s face corresponds to the elevated spike in the middle of the red waveform. And the neutral wall can be confirmed by the equally level shape of all three color channels at the right of all three waveforms.
Figure 4.33 The parade scope analysis for Figure 4.32.
By learning to identify features within the parade scope graphs, you can quickly spot where unwanted color casts appear and get guidance as to where within the image you need to make corrections.
Learning to Read Parade Scope Graphs
The RGB parade scope is essentially a Waveform Monitor that displays separate graphs for the red, green, and blue channels of an image. To understand the parade scope’s analysis, you need to learn how to compare the shape and height of the three Waveform graphs to one another.
Similar to the Waveform Monitor, each of the parade scope’s graphs presents a left-to-right analysis of the tonality in the scene. The difference is that while the Waveform Monitor measures the luma component, each graph in the parade scope represents the individual strengths of the red, green, and blue color channels.
In Figure 4.34, the generally accurate and neutral color balance of the scene is evidenced by the relative equality of the heights of the red, green, and blue channels, especially at the top and bottom of each waveform.
Figure 4.34 An image with an RGB parade scope analysis showing evenly balanced highlights and shadows.
Even though the graphs look similar, closer inspection reveals that the peaks and valleys of the parade scope’s three graphs correspond to various features in the picture. While strong highlights, shadows, and desaturated elements often have components of equal height in each graph, saturated subjects will certainly vary.
For example, splitting apart the red, green, and blue channels of the image in Figure 4.35 and superimposing the red, green, and blue parade scope waveforms shows the correspondence between individual features within the image and the strength of each parade scope waveform. Keep in mind that each individual color channel is merely a grayscale image and that the corresponding waveform is simply an amplitude measurement of that channel.
Figure 4.35 In this image, the red channel is significantly stronger (elevated) all the way through the graph, while the green channel is the next strongest. This indicates a strong yellow/orange (the secondary combination of red and green) color cast throughout the shadows, midtones, and highlights of the image.
Looking closely at each waveform reveals that, while the highlights corresponding to the pillar and window sill are of equal height, the portion of the red waveform corresponding to the faces is stronger than in the green and blue channels, which we’d expect. There’s also a spike in the red channel that lines up with the brick wall, which we’d also expect.
By identifying a particular feature within the graph, you can check its color balance. Generally speaking, color casts are the result of one or two of the color channels being either too strong or too weak. Whatever the problem, it’s easy to see which color channels are at fault using the parade scope. In Figure 4.36, a bit of detective work might reveal that the white balance setting of the video camera was incorrectly set relative to the lighting of the environment. If you’re dealing with a film image, a film stock may have been used that was inappropriate for the lighting.
Figure 4.36 In this image, the red channel is significantly stronger (elevated) all the way through the graph, while the green channel is the next strongest. This indicates a strong yellow/orange (the secondary combination of red and green) color cast throughout the shadows, midtones, and highlights of the image.
Whatever the reason for the color cast, simply knowing that one of the channels is inappropriately strong is a starting point. A closer examination of the parade scope’s graph will also tell you exactly what you can do about it.
In Figure 4.37, the bottom of the blue channel’s graph is significantly lower than those of the red and green, even though the top of the blue channel is higher (providing the strong bluish highlights for this night scene). This is your cue that the deepest shadows (blacks) of the image are imbalanced, which lends an odd, washed-out look to the image.
Figure 4.37 A low-light image with a color imbalance in the shadows.
Keep in mind that balancing shadows using the Lift control can be a tricky operation that, if not done precisely, can cause more problems than it solves if you inadvertently add a different color imbalance to the blackest parts of your image.
Most scopes have an option to zoom into the graph so you can get a closer look at how closely the shadows of the parade scope waveforms are aligned, making it a lot easier to do this critical black balancing.
In Figure 4.38, we can clearly see after zooming into the parade scope that the blue channel is weaker in the shadows than the red and green channels.
Figure 4.38 Zooming into the bottom of the parade scope makes it easier to align the blacks of the image.
RGB Parade vs. RGB Overlay
An RGB parade scope and an RGB overlay scope both display the same information, but they differ in their presentation. As we’ve seen previously, parade scopes display discrete waveforms of information side by side so that you can see each waveform independently and in its entirety. Overlay scopes, on the other hand, superimpose all three waveforms over one another so that you can see how they align more interactively.
Which is better is completely a matter of preference, but here’s a hint on how to spot where the red, green, and blue waveforms line up, and where they don’t, on an overlay scope: Modern overlay scopes usually have the option of displaying each of the three color-channel waveforms with the color they represent and the three graphs combined additively (Figure 4.39). This means that, where the three waveforms align perfectly, the resulting traces in the graph turn white (since equal red + green + blue = white).
Figure 4.39 RGB overlay scopes.
Many software scopes provide the option to turn color on and off, on the premise that the colors can be a distraction in a darkened suite. While parade scopes can still be read with the graph colors turned off, color is essential to being able to make sense of an RGB overlay scope, so make sure it’s turned on.
Where the waveforms don’t line up, the discrete colors of each waveform are more or less clearly visible in the region of image tonality where the incongruity occurs, making offsets more visible.
Different applications also present individual histograms for the red, green, and blue channels. Similar to a luma histogram, each color channel histogram shows a statistical analysis of the number of pixels at each level of image tonality. The results are somewhat similar to the RGB parade scope in terms of seeing the comparative strength of each color channel in the highlights, midtones, and shadows of an image.
Unlike the RGB parade scope, there is no way to correlate an individual feature or subject within the frame to the rises or dips on any of the color channel histograms. Large rises indicate a lot of color channel pixels at that range of image tonality, while dips indicate fewer color channel pixels.
Depending on the application, RGB histograms can be either presented in parade mode or overlaid over one another. Sometimes histograms are oriented vertically, as in FilmLight Baselight (Figure 4.40, left), while other applications present them horizontally (Figure 4.40, right).
Figure 4.40 Two RGB histogram graphs compared. FilmLight Baselight is on the left; SpeedGrade is on the right.
RGB histograms are very good, however, at allowing you to compare the overall strengths of each color channel within each zone of image tonality.