Hi,

I've been playing with the dual photography concept:
http://graphics.stanford.edu/papers/dual_photography/

And I've come to these images:

1.- Primary image. Standard light and camera
2.- Reconstructed dual image. Shine the light over a tiny portion of the
scene, create a lot of 1x1 pixel images and put them together as a mosaic.
Camera and light layout as in #1
3.- Reference dual image. Camera and light exchanged

Ideally, 2 and 3 should be identical. 3 looks somewhat distorted... there
must be something I didn't take into account. It's also clipped, because
the "photodetector camera" was not wide enough to capture light from every
possible direction. And it's too noisy, that could be due to incomplete
antialiasing.

I'm sending the source code to p.b.scene-files.

-- 
light_source{9+9*x,1}camera{orthographic look_at(1-y)/4angle 30location
9/4-z*4}light_source{-9*z,1}union{box{.9-z.1+x clipped_by{plane{2+y-4*x
0}}}box{z-y-.1.1+z}box{-.1.1+x}box{.1z-.1}pigment{rgb<.8.2,1>}}//Jellby

Post a reply to this message

Attachments:
Download 'primary.png' (28 KB) Download 'reconstructed.png' (10 KB) Download 'reference.png' (5 KB)

Preview of image 'primary.png'

Preview of image 'reconstructed.png'

Preview of image 'reference.png'

Among onther things, myself wrote:

> 3 looks somewhat distorted... there
> must be something I didn't take into account. It's also clipped, because
> the "photodetector camera" was not wide enough to capture light from every
> possible direction. And it's too noisy, that could be due to incomplete
> antialiasing.

I meant "2", I had to change the numbering because knode sorts the
attachments alphabetically...

Anyway, there was something missing, when creating the reference image, the
light should be projected through the same view field the camera has in the
primary image, here is the corrected "reference.png", much more similar to
"reconstructed.png"

-- 
light_source{9+9*x,1}camera{orthographic look_at(1-y)/4angle 30location
9/4-z*4}light_source{-9*z,1}union{box{.9-z.1+x clipped_by{plane{2+y-4*x
0}}}box{z-y-.1.1+z}box{-.1.1+x}box{.1z-.1}pigment{rgb<.8.2,1>}}//Jellby

Post a reply to this message

Attachments:
Download 'reference.png' (4 KB)

Preview of image 'reference.png'

Among onther things, myself wrote:

>> 3 looks somewhat distorted... there
>> must be something I didn't take into account. It's also clipped, because
>> the "photodetector camera" was not wide enough to capture light from
>> every possible direction. And it's too noisy, that could be due to
>> incomplete antialiasing.
> 
> I meant "2", I had to change the numbering because knode sorts the
> attachments alphabetically...
> 
> Anyway, there was something missing, when creating the reference image,
> the light should be projected through the same view field the camera has
> in the primary image, here is the corrected "reference.png", much more
> similar to "reconstructed.png"

... and with +R9 in the antialiasing options a much better result is
obtained for "reconstructed.png" itself. Now the checker structure is
evident, it works!

The problem is... is this useful at all?

-- 
light_source{9+9*x,1}camera{orthographic look_at(1-y)/4angle 30location
9/4-z*4}light_source{-9*z,1}union{box{.9-z.1+x clipped_by{plane{2+y-4*x
0}}}box{z-y-.1.1+z}box{-.1.1+x}box{.1z-.1}pigment{rgb<.8.2,1>}}//Jellby

Post a reply to this message

Attachments:
Download 'reconstructed.png' (9 KB)

Preview of image 'reconstructed.png'

> The problem is... is this useful at all?

I can't see an immediate use for this in the raytracing-world, but it's
interesting anyway :)


cu!
-- 
ZK AKA SaD
http://www.povplace.com
"Train you will I. Train you I will. Yes! Step one: touch your tongue to
mine."

Post a reply to this message