Max Lyons Forums

[Buho]

Hey all. I'm toying with an idea I haven't done before, but perhaps someone else has.

I've got a 180º x 120º panorama with architectural elements on the left side and nature on the right. I am strongly resisting projection in cylindrical because of the horrible distortion. However, rectilinear restricted to 120º or even 140º doesn't show everything and the resulting composition just doesn't work artisticly.

I'm toying with blending rectilinear on the left with cylindrical on the right. Is there anything special I have to do? Is it possible? Will it blend correctly? Are there examples of such I can look at to figure this out better?
_________________
Eric

[Jim Z]

Eric,

If you split the 'construction' in the center, the halves would match, but wouldn't the perspective of the left half be with the vanishing point at the right edge of the building? I'm thinking that the cylindrical and the rectilinier match at the vanishing point, perspective center, of the rectlinier projection.

I have many panoramas warped and merged by eye in PS, before PTassembler, with nothing close to correct perspective or projection, and they work very well. Correct perspective can some times be, to my eye, harsh in a natural scene.

I have a panoramic photo of a grand, 80 year old, log built lodge at Lake Tahoe, assembled from five images, about a 100deg FOV. The faciade isn't straight but ahs inside and out side corners towards the camera. More or less, my old pano keeps each side, or wall, correct with straight lines, but doesn't keep correct over-all perspective, but is an artistic, 'painting' perspective, verticals are right but the VP isn't. I just re-assembled the same photos in a correct rectilinear projection. It's a give or take which is better. The funny thing is that the most obvious roof eve in the photo looks worse in the correct projection, because the roof droops, for real (old age), towards the camera. The best rectilinear had that part 'morphed' up to be 'correct.'
_________________
Jim Z

[GURL]

I would try that, too:
1) in a first project stitch the whole pano to cylindrical.
2) in a second project, use the cylindrical previous result as the (single) source image of Lens Type = Panoramic and use various "lens" Horizontal FOV (the Step 2 parameter). If those FOV values are lower than the real one (the Step 5 Horiz FOV of the first project) you should get various intermediate results between the "correct" rectilinear an the "correct" cylindrical projections.

I imagine you could end blending two half parts produced using different (fake) horizontal FOV...

Because I know you are using a 3 monitors setup, I would suggest making 3 separate panos using rectilinear projection and Horiz FOV = 60°: one pano for each of the monitors. If you want to make prints, use cylindrical projection but fix the result on a half cynlinder carboard mount and ... place your head in the center: this will correct the so called "distortion" (this is not a joke, a friend of mine saw some panoramas shown this way in a gallery and found the result being very pleasant.)

Besides that, showing us some thumbnail(s) of the pano could help...
_________________
Georges - - Stay tuned, this is just the beginning of digital photography

[johnh]

It isn't difficult to do a quick test. This has the left side in rectilinear projection and the right side in cylindrical. I had to use edit->transform->scale to match the images reasonably at the join. I haven't spent any time trying to get the join invisible. A blue sky helps.

John

[Pablo]

I have seen a poster (called Squaring the Circle in Panoramas) about this topic at the ICCV conference:

http://www.vision.caltech.edu/lihi/Demos/SquarePanorama.html

They also explore the multiple projections in one Panorama like in the example by john, as well as using locally different projection centers, for example to represent peoples faces. Painters sometimes paint the buildings and faces with different viewing directions, so that the faces look undistorted even if they are far from the center of the image.

In the examples, the authors virtually center the camera on each head for example, and remap it. Then they have some holes in the background, which they fill using some image resynthesis algorithm.

ciao
Pablo

[Buho]

[GURL]

Why pictures look right when viewed from the wrong place
...to which title was lately added
(and sometimes look wrong when viewed from the right place)

[Jim Z]

Pablo, excellent link.

I guess the reason the rectilinear and the cylindrical projections don't match at the vertical center lines of each is aproximately the equivalent of the differance between a Mercator and a geographic map. The rectilinear (the same as 'perspective') projection has the same vertical distortion at the center of perspective as the Mercator projection has uniformly and a cylindrical projection is the same as the geographic projection... I think.(?)

Now I know that I my own backwards way I've been doing 'Multi-Plane' projections...
_________________
Jim Z

[johnh]

[Jim Z]

[johnh]

[Buho]

I gave the hybrid a shot last weekend.

1. Output cylindrical. Use Enblend or blend in PS. Do not crop.
2. In a new project, reproject cylindrical to rectilinear. Do not move yaw/pitch/roll. Set lens parameters to 0. Set the HFOV manually (annoying warning in PTA when doing this, Max).
3. Take both into Photoshop.
4. Mask out half of the rectilinear image, exactly over the vertical half line. Place overtop the cylindrical and line it up.
5. Edit -> Transform -> Scale until rectilinear half matches the cylindrical half. No blending nescescessary, as they should line up perfectly.

The results are a little smoother than one of these articles posted, since straight lines transition from a straight line smoothly to a parabolic curve instead of intersecting at angles. The technique is quite liberating, though, compositionally-wise, for ultra-wide angle scenes.

I'll post my pano later. It's quite wild to flip between full rectilinear and hybrid. It feels like my viewpoint is "bending" somehow, something I've never experienced in real life.
_________________
Eric

[Jim Z]

Eric,

Thinking about this, you could join any number of rectilinear images if the joins between the two adjacent pairs are equal distance between the respective perspective centers. (And if you can survive the roller coaster horizontals.)

The join of a cylindrical and rectilinear projection at the (perspective) center of the rectilinear is a verticle line, but if you join the cylindrical and rectilinear images off of the center of the rectilinear image and vertically scale the images to match top and bottom equally above and below the horizon, and if I have this figured out correctly, the join line is the arc of a circle, covex towards the cylindrical projection, vertically tangent at the horizon line...
_________________
Jim Z