3D

I don't know if this has to be a request for FF4 or a suggestion for FF in general.
Anyway..
Let's suppose we have a picture to work on. Just a normal picture, with people and a landscape. The face of the people is nearer. The landscape is not on the same distance as the face of the people. How to apply the same effect to the whole photo keeping the different feeling depending on the distance?
We could use Photoshop. Divide it in layers and after using FF on every layer, but the result would be so so.
Any other idea?
Could a new FF4 work on different layers as they were very far ? How to say.... If we have a person as main subject of the photo, and a background as something very far... Can FF works on them and set a unique effect? I tried the effects of FF3, but, even helping with Photoshop, the result is not satisfying.
Yes, doing a lot of job with Photoshop, we can clearly get a better result, but that's not my idea. I just wonder if the new FF could do a smart work on the distances ( for example, selecting two layers, one as main and a second as far layer )

Using FF as Photoshop plug-in, it could be surely done, but I don't know if it is possible as stand alone program.

I've been thinking about something similar for years, i.e. how to "detect" the depth of field in a natural photo - with relative post processing in mind of course.

So I'd say the first problem here is how to determine the depth - I thought about somehow using correlation of frequency changes in different color spaces. This is not easy, and my attempts so far did not produce anything useful (too much noise/errors).

Skybase - that is exactly what I meant, i.e. frequency ranges change over depth due to focal change. Lets get a challenge going

No, hold on a second: Does it even matter whether the area is out of focus or just lacking sharp edges?

If the effect is dependent on how sharp the image area is, it might not matter whether the area is out of focus or just without much high-contrast detail, in many cases the results would likely be visually similar.

Which makes me think that some High-pass-related solution might be adequate. In other words, where the image result run through the High-pass component is close to mid gray the effect would have a larger radius, and the further from mid gray the High-pass result the smaller the radius.

Sounds simple enough.

Sort of like unsharp mask or smart blur.

Okay, so here's another thought:

Most images have a certain type of perspective, the most common being a ground plane with a sky (or back wall). In this scene there can be individual objects.

You could have a few simple inputs for the basic layout, such as horizon height and back wall distance, perhaps focal distance. This is the background set.

And then, wherever there are areas that have sharper details than a given threshold (different enough from expected background focal values) you assume there to objects. All the rest is background.

Perhaps this method could be used to construct a basic depth channel with just a couple of inputs?

This thread has become ThreeDee talking to himself thread. D:

YUVsoft seems to just basically put all what's being discussed here into a package. I really want to try their products since it really adds to what I do. But they probably cost a few thousand bucks.

Here is one of the strange bi products of the experimentation. The input image is the Ladybug image

Ooohhhh ho hoho..... these results are look seriously promising. So I've been wondering... how long does it take to process the image?

With less smoothing.

ThreeDee - that looks good! We both seem to have a problem with the "fill" though. Here is the same algo and image with some other settings.

You could then force the z-channel for foreground objects to be lighter than (for instance) 50% and background objects darker than 50% gray value. The actual value could be adjusted based on focal depth.

Quote
CFandM said: You could always do this for masks....It the way I had to do for some filters..

I'll be darned. Exactly that -- thanks! Didn't think to try that method. External selection never worked, so I had given up on it.

It was that well-hidden "Use transparency" option that I missed

Faking it even further by blending some of the brightness channel information and blurring the mask for additional roundness: