logo Sign In

Color matching and prediction: color correction tool v1.3 released! — Page 4

Author
Time

The corrections are fully automated, so the only way to adjust the brightness/contrast is by adjusting the reference frame. For the moment I will keep the reference frame the way it is, but if you think you've got a better color grading for the reference frame, I will be happy to show how it will work out for the other frames. 

Author
Time
 (Edited)

Using the same white balanced Legacy frame, I recalibrated the color adjustment model for the -1 video sample:

Legacy white balanced versus -1

http://screenshotcomparison.com/comparison/139207

-1 before versus -1 after

http://screenshotcomparison.com/comparison/139208

and the other frames previously posted:

http://screenshotcomparison.com/comparison/139210

http://screenshotcomparison.com/comparison/139211

http://screenshotcomparison.com/comparison/139212

http://screenshotcomparison.com/comparison/139213

http://screenshotcomparison.com/comparison/139214

http://screenshotcomparison.com/comparison/139215

http://screenshotcomparison.com/comparison/139216

http://screenshotcomparison.com/comparison/139217

http://screenshotcomparison.com/comparison/139218

http://screenshotcomparison.com/comparison/139225

http://screenshotcomparison.com/comparison/139227

http://screenshotcomparison.com/comparison/139228

Author
Time

I really like the newest versions

Author
Time
 (Edited)

http://screenshotcomparison.com/comparison/139204

The whites are blown out on the storm trooper on the right and a lot of shots came out sort of yellow and too bright and the faces are a bit orange.  The reference frame looks great at first glance until you start to see these other frames and look at it again.

http://screenshotcomparison.com/comparison/139173

Looking again at the original, it is brighter so all of the other scenes will be brighter too.  Can you do comparisons with the original without adjusting it?

Edit: I posted that before seeing the -1 corrections.  My comments for those is everything looks too green and dark. R2's head doesn't look silver for example.  I wouldn't necessarily worry about it though.  I would be more interested in the bluray and the Gout.

Author
Time
 (Edited)

I've improved my method, by including the rotations archimedes is using, with the difference that it can also predict other frames.

Here's the bluray calibrated to the unaltered Legacy frame:

http://screenshotcomparison.com/comparison/139276

Here's the other three bluray frames from the Tantive IV scene, adjusted with this color adjustment model:

http://screenshotcomparison.com/comparison/139278

http://screenshotcomparison.com/comparison/139279

http://screenshotcomparison.com/comparison/139280

Author
Time

Something is definitely not right with the -1 color adjustments, blacks are crushed, possibly even murdered by this process.  Also I agree it has ended up too green.

Author
Time

These aren't as bad.  The skin tones are better.  If you can back off the green just a bit without raising the brightness it would probably be good.  I would color correct this myself if I had access to my other computer that is packed away while I am trying to sell my house.

Author
Time
 (Edited)

With regards to the -1 corrections there are a couple of things to consider:

1) It's not the process, but mostly the reference and the source that determine the outcome. If blacks are crushed in the reference or the source, they end up crushed in the correction.

2) The dark areas in the -1 preview don't have much detail, due to compression, so by definition any increase in contrast will lead to crushed blacks. 

3) The colors in this source are in pretty poor shape.

4) The correction is meant to demonstrate the potential of the method. This correction was done, based on a single Tantive IV frame, assuming the Mike Verta frame is a good reference. So, it is not going to be perfect for the entire film. The fact that it works this well is surprising to say the least, but this is not meant to be a final color correction for the -1 sample. 

Author
Time

Great discussion, and very informative! Mike is such a professional.

The Legacy frame I used has a preliminary color grading. I was aware of that fact. I personally like the more saturated look with more contrast. It also matches the Senator print photos pretty well, but the color grading is by no means definite.   

The new samples posted by -1 look terrific, so the color correction on the old video sample is outdated anyway. It mostly serves as an example what can be done with a faded source, and a limited number of reference frames.

Author
Time

DrDre said:

The new samples posted by -1 look terrific, so the color correction on the old video sample is outdated anyway. It mostly serves as an example what can be done with a faded source, and a limited number of reference frames.

 Yes. And Team is still working on the Reel 3 color correction. Given that it's the best color of any footage they've posted (Jedi aside, perhaps), the fact that it will be further corrected is exciting if it's done right, worrisome otherwise. Needs more of a delicate touch, rather than the complete overhaul some clips require.

Author
Time
 (Edited)

I'm now able to reference from a single frame, too. However, it only works with one rotation. 

Here's how the algorithm works. First, I match the normalized cumulative histogram of each color channel. This results in a 1-to-1 transfer of each value to the target value.  For example, the red value of 185 may change to 233. I'll save this 1-to-1 mapping and use it for other pictures. The result is really great and  the algorithm saves a lot of time with the color mapping process. 

But as I've said, it only works with simple histogram matching, because when I apply the pdf algorithm, there is 1-to-many mapping. And I haven't found a way yet, to calculate how the next value is choosen. Since the next values are usually in a range (let's say 185 changes to values between 225-240), I've calculated the mean value and used it for one to one mapping. Doesn't work, unfortunately. I've also tried to map from rotation to rotation, but even then one value is mapped to many others. 

So yeah, the mapping based on a refernce frame is a great addition, but so far it only works with with histogram matching/one rotation. It will result in a good color transfer, which is probably enough most of the time. 

Author
Time
 (Edited)

That's great. Actually, it's a many to many mapping, not a one to many mapping. You should save the rotations, and the mapping in each color space, and then apply each of the rotations and mappings to the new image, transform the individually mapped images back to your original color space, and average all the predicted images. 

Update: there's a bug in my code, so mapping multiple color spaces for different images/frames is not yet possible.

Author
Time
 (Edited)

Ive tried two ways to implement the mapping from one frame to another in multiple rotations. 

The first is the one you described. The problem here is that there are still many-to-many mappings in the rotated color spaces. Granted, the ranges are usually pretty narrow, but this has still an effect along all rotations. My algorithm selects the value which is most often mapped. So, there is information lost in each rotation. This is why only few rotations in different frames will get the correct colors for those rotations, but not for many rotations. 

The second one only works for frames of the same shot. I've noticed that each RGB color of the input frame is mapped to exactly one RGB color in the full rotated colored output frame. In my tests, there were usually about 50k-60k unique colors in one frame. So it's hardly enough to recolor a whole film based on a single frame, using this method. However, if I use the color transfers of the reference frame, I can usually correctly recolor 95% to 99% of the pixel of the same shot frames. The remaining pixels are recolored according to the mapping function obtained from simple histogram matching. 

Author
Time
 (Edited)

I think the problem is, that the histogram matching method assumes that the individual color channels can be adapted independently. Therefore each value has a unique mapping. Although this usually is a good approximation, in principle there can be multiple mappings for each RGB color.

Author
Time

Here are a few preliminary bluray color corrections. The crushed blacks of the bluray are an issue, but you get the idea:

http://screenshotcomparison.com/comparison/140626

http://screenshotcomparison.com/comparison/140627

http://screenshotcomparison.com/comparison/140629

http://screenshotcomparison.com/comparison/140630

http://screenshotcomparison.com/comparison/140632

http://screenshotcomparison.com/comparison/140633

http://screenshotcomparison.com/comparison/140635

Author
Time

I generally like the color on these ones. Highlights are just way too bright in some cases and the detail washed away. But aside from that, they look nice.

Author
Time

Looks pretty good!  They are a touch blown out, but that is because they probably should be to be representative of how it looked.  The Blu-ray preserves more details at the expense of an overall darker image.  These shots might look better on an HDR display (assuming the detail is not actually clipped).

Author
Time

Can you try it on this scene?  Mike laughs at the blu-ray on this one and I would love to see how it comes out with the automatic correction.

https://vimeo.com/124147988