Projective Transform (Four point transform)

Here is a moderate example with clipped x and y:

Sphinx,

This is awesome to say the least! It will open a whole world of possibilities in mapping images onto 3D surfaces.

As to the interface, it is workable as it is. The 'inverse' checkbox should work the other way though, IMO. Right now it is checked when you do the original transform, and unchecked for inverse transform.

For a possible improvement, it would be helpful to be able to show just a tranformed grid overlayed on the input image. This would be particularly useful for alignment when removing perspective with inverse transformation.

TD

Quote
The 'inverse' checkbox should work the other way though, IMO. Right now it is checked when you do the original transform, and unchecked for inverse transform.

I guess that makes sense from how you see it visually. Theoretically or codewise it is the other way around though (i.e. it uses the inverse matrix when Inverse is checked). Maybe there is a better term for that input.

Quote
could the clip function have some sort of invert choice as well? With some settings, the transformed image splits and goes through the opposeite edges of the image. (Or is this the way it is supposed to go?)

Well, the thing is that for correct output the transform only support convex quadrilateral polygons. I was not sure how to handle this and decided to leave it as is - any ideas on what to with concave settings?

Quote
Sphinx. wrote: I guess that makes sense from how you see it visually. Theoretically or codewise it is the other way around though (i.e. it uses the inverse matrix when Inverse is checked). Maybe there is a better term for that input.

You could call it 'Inverse Transform' and switch the checkbox polarity.

Quote
Sphinx. wrote: Well, the thing is that for correct output the transform only support convex quadrilateral polygons. I was not sure how to handle this and decided to leave it as is - any ideas on what to with concave settings? Hmm, I could constrain the points so that you can't make a concave polygon.. the sliders would then stop having any effect if you make a concave polygon..

I see. Then contraining the points might be a good idea. It is not all that important, more of a cosmetic handling. Now that we've gone there: should there be some way of indicating when the points are clipped? Such as a red exclamation point at the corner (which you can turn off)? That's just a thought. In the big picture, it's irrelevant for the basic function of the filter, which is a fantastic piece of work.

Quote
SpaceRay wrote: I am sorry to ask, but do not know how this would help mapping images onto 3D surfaces unless you mean doing this inside FF in some way, although FF does not have 3D surfaces.

Yes, I mean internally in FF. For instance, if you were to make a script that calculated the corner points of the three visible faces of a 3D rotating cube, you could use this script to render the input image on each face.

You could pretty much make any rectanglarly faceted filled 3D shape with the input image on each face with it.

Which reminds me that it would really be a good idea to write a script that calculates the transformation of any point in 3D space.

In fact, 3D transformation may already be included in one of the more advanced scripts from Sphinx. Or in Raytracing by Egret. Haven't dug into them that deep.

(By 3D transformation I mean rotation around X, Y and Z-axis.)

Does anyone still have this snippet? Attachment goes to a "not found" page.

Thanks Sphinx!