In my previous post I’ve introduced the Histogram that will be available in kdenlive 0.7.8.
Next in the list of the Color Scopes I added are two closely related scopes: Waveform and RGB Parade. Both of them do the same, Waveform for Luma, RGB Parade for the RGB components. Therefore I will not always explicitly point out that properties hold for both scopes and speak of Waveform only.
More in the full article »
The Waveform is kind of a 3D Histogram.
That’s nothing to do with the fact that the above image looks kind of 3D. So where are the three dimensions?
The Waveform looks quite impressive here. But what can we learn about the image?
That is where the RGB parade drops in now.
What we see is very slight crushing for blue on the tree, and more or less equal parts of red and green as well. So if you thought the tree were green, your brain fooled you. It is not. (Actually there is no green at all in the image.) Also, blue nearly doesn’t clip at all, even in the sun. So it is still a little bit orange, which is great.
Now, didn’t the Histogram show clipping and crushing as well? Yes, it did. But it did not show where. That is a big benefit. As in some cases, if you see that there is some clipping on a light bulb, you perhaps don’t mind increasing the overall brightness of the image. It does increase clipping as seen on the Histogram, but the Waveform shows that only the light bulb is clipped a little more, which you can afford in our imaginary case.
Wait … imaginary, I said? Actually I do have a short clip with light bulbs. Which is quite interesting, not only due to the wrong white balance.
The RGB Parade reveals two things at first glance. You will at least recognize the first one as well.
To achieve proper white balance here, we make use of a new effect called SOP/Sat (if you want to know more about it: It implements the ASC Color Decision List. More to read also in the source code comments of the effect.) Curves would work as well (actually curves could do everything), but let’s use a new effect here.
The first thing to decide is how bright the darkest spot should be. This can be controlled with the Offset parameter. Reference is again the left part of the Parades, the soon-to-be white ceiling. You can use your mouse to make the Waveform or RGB Parade draw a horizontal line and display the value there. I lifted all channels to 48 in this example:
Second step is Stretching the channels. This is done with the Slope sliders. The goal is again to find a neutral spot in the RGB Parade and use it as reference. We could again use the ceiling on the left, but after some testing I found out that the little wave in the middle of the scope works as well. Its advantage is that it is slightly brighter than the ceiling, allowing to correct the color cast more precisely.
And voilà, exposure and white balance are corrected.
When taking a look at the full-sized image (or when trying it yourself with the sample clip available for download at the end of this article) you will notice color waves in the image. This is the result of the stretching; We have blown up the ceiling on the left from 25 px height to nearly 100 px. If the clip had been exposured and white balanced correctly when shooting, we would have 100 distinct values there, but now there are only 25 different blue levels, causing these steps. Also in the histogram the image looks torn apart. (Another reason might be the high compression of the Nikon D90 clips.) This effect is called Posterization.
This is one of the reasons why more expensive cameras (and with that I mean really expensive ones, like the RED One, to name an extreme example) record videos in higher bit depth. Perhaps all clips you’ll ever encounter only store 8 bits per channel, so there are 28 = 256 possible values for each channel. Having for example 10 bits per channel would already result in 210 = 1024 possible values. This would already have solved our problem.
But before you buy a RED one now – kdenlive does not support more than 8 bit yet.
Looks perfect. Good exposure (says the eye and the Waveform), beautiful colors.
Nevertheless, there is some clipping. It is just hidden by the Luma calculation: For Luma, only pixels that are totally white are at the top of the Waveform.
Although the green leaf looks much brighter than the red tip, it is the Red channel which clips at the tips. The thing is that our eyes are most sensitive to Green, less to Red, and even less to Blue. That is also the reason why the beam of a green laser is visible in the night sky, but a red one is not (unless you’ve got a really strong one.)
This is a classical High Key shot. Bright subject (but not clipped yet), white background.
Interesting detail: When color grading, some colorists prefer scopes with neutral colors, and basically with neutral everything. Just greyscale, except for the video itself. Why that? Quick answer: Afterimages. If you look at a green surface and then immediately color grade an image by eye, there will be too much green in it.
To explain the last point a little more in detail: A black/white gradient draws a line from the bottom left to top right on the Waveform. When changing the colors, e.g. with the SOP/Sat filter or with Curves, the line will change. (Attention, effects working on the saturation will not have any effect on a grayscale gradient!)
In this gradient above I have applied a SOP/Sat effect to give the blacks a blueish touch and the mids and highs a warm touch. You can play around with the gradient file and some color correction effects on the gradient file as well. What the above is good for will be in the next part.
Waveform and RGB Parade are mighty scopes. Especially the RGB Parade. Correcting the exposure is easy, with these scopes you can always keep track of the levels of each color component. It is also possible to do white balance by adjusting blacks first and whites afterwards because the horizontal axis in the Scopes correspond to the horizontal axis in the video which allows to detect spots that should be neutral.
Thanks for reading!
Please drop your comments below.
Simon A. Eugster (Granjow)