With the tone curve established, this blog entry talks about how we can match the color rendition of the two cameras. As anyone familiar with video cameras knows, no two brands of cameras look the same straight out of the box - much to the delight of a DIT's employment security! First, let's look at the color rendition of the Hero 3 camera with the "Cam Raw" white balance versus the "5500k" white balance (I'm using an approximate 5500k light in this experiment).
As you can see (clicking on the picture to enlarge it might help) the "Cam Raw" white balance leaves the picture quite a bit desaturated from the 5500k white balance. It appears that the folks at GoPro are doing a bit of image processing on the non-"Cam Raw" white balance settings, which makes sense if you read the instruction manual. Let's look at the vectorscope plots to see what's really going on.
The vectorscope shows that not only is the video greatly desaturated by the "Cam Raw" setting, but GoPro also adjusted the overall color rendition because the outline of the color coordinates on the plot changed. This tells me that they added a 3x3 matrix adjustment with normal white balance settings. Why does this matter? Well...because with these fragile 8-bit codecs we're using on these cameras I'm looking to do most of the camera matching in-camera. That's where the video is still better than 8-bit. From experience I can tell you that the 5500k white balance setting will likely be a better place to start the camera matching because it will require less adjustment later - but we'll get to that at the end of this article.
Next, let's look at the color rendition of the DSC CamAlign chart using the neutral picture style on the 5D with contrast set at -3. The vectorscope plots below show the color rendition with the saturation set from -1 to -4. I didn't bother with the 0 to 4 settings since the color actually gets over saturated at a setting of 0, but maybe I'll save that for a follow on article. Notice how the overall color vectors maintain their angles, but the amplitude decreases with decreasing saturation settings?
Again, why does this matter? Well, I'm looking for the best match of saturation with regard to the Hero 3 and the 5D. I think at this point I may have to skip the technical explanation and just say that the best fit is a saturation value of -3 on the 5D. How did I figure this out? Well...that would require explaining the spreadsheet shown below in detail and who wants to read through that? (Augh!! implied...even I got a headache at one point last night) The bottom line is that the 3x3 transformation matrix we are going to apply later in Premier Pro maintains a unity gain with the 5D saturation set to (-3). ...and yes, that's a virtual vectorscope I wrote in Excel!
So what's the next step. This is where things get a bit technical, but the spreadsheet shown above calculates a new 3x3 matrix that is applied to the GoPro RGB values to transform the vectorscope color coordinates you see at the beginning of this article (Hero 3 Black 5500k WB) into the 5D color coordinates with saturation set to -3. Cutting to the chase once again; here is the matrix...
[ 1.00 -0.23 0.19] [R-gpro] [R-5d]
[-0.01 0.90 0.07] x [G-gpro] = [G-5d]
[-0.07 0.22 0.86] [B-gpro] [B-5d]
No doubt, anyone familiar with image processing is yawning about now. So how do we implement this in Premier Pro? Easy. Use the Channel Mixer effect. It implements the matrix multiply, just like we would in a pro camera that's being calibrated. See the picture below for details on how to implement the effect, but all it's really doing is the matrix multiply shown above which was calculated from my spreadsheet. Easy, right? Notice how now the vectorscope plot has gained the same shape as the 5d neutral vectorscope plots shown above.
Now...what happens if we have the 5D set to saturation values of -1,-2, or -4? Isn't that going to be difficult to match? Nope. That's just a matter of adding a Fast Color Corrector effect in Premier Pro. You can then scale the saturation to match saturation settings on the 5D. Empirically, I found the following settings to be approximately correct.
- 5D saturation -1 means set the GoPro saturation to about 120%
- 5D saturation -2 means set the GoPro saturation to about 110%
- 5D saturation -4 means set the GoPro saturation to about 90%
One thing to note here is that the magenta is still a little bit off in the "After Matching" picture. The process I've described isn't a perfect match by far, but it gets us very close to matching. For me, perhaps sufficiently close.
The process I'll talk about in the next article involves transforming the GoPro color rendition into a semi-calibrated rec.709 color rendition. I developed a rec.709 custom picture style for my Canon 5D mark II, which I use quite often on projects. I know not everyone has access to fancy test charts and knows how to calculate color matrices, so I thought it would be best to start with something anyone sufficiently familiar with video editing can do. Stay tuned for the exciting episode kids.
P.S. Don't you think GoPro should give me a free camera for this? ;) It's just about Christmas...just saying...