Is it possible to reconstruct two 2D images (Front side and left side) to 3D image ?

Something like these....

Answer 1

[Image Analyst]

You can use repmat() to replicate the side image a bunch of times, say N times. Then use bsxfun() to mask the volumetric image with the front view. Here's a start

% Replicate side view N times into 3-D.
image3D = repmat(sideView, [1,1,N]);
% Mask the image.
final3DImage = bsxfun(@times, image3D , cast(frontView, 'like', image3D ));

It might take some adaptation, but that's the basic idea if you have two binary images and want to project in 3-D.

[dhwani contractor]

Hi, I am working on something similar. I have the binary images of flower from 0 to 360 degree with degree difference of 1.8 degree. How can I make 3D model from the boundary information?

P.S I have tried detecting boundary and then rotating it but this concept is forming wrong pointcloud. Any help on this would be really appriciated. Thank you.

[Walter Roberson]

If you have boundaries extracted at different rotation angles, then your situation should, I suspect, be equivalent to doing CT reconstruction. If I recall correctly, that uses the radon transform; https://www.mathworks.com/discovery/image-reconstruction.html

[Dynamic]

@dhwani contractor can you please share your code for forwarding projection which helped you reconstruct the 2D projection from the 3D object?

Answer 2

[Walter Roberson]

No, it is not. You can see from this demonstration that two side views cannot possibly be enough to reconstruct the entire object.

[Watcharaphong Yookwan]

What if I need to reconstruct only 2 sides ?

[Walter Roberson]

The only two sides you can reconstruct with that information is the two sides you start out with -- in other words, no reconstruction at all.

[Image Analyst]

You could get A 3-D object from 2 views by extruding the 2-D shapes along each direction. It would not necessarily be the right one though. If it were right, then there would be no need for CT and MRI and those Nobel prizes for those would never have been granted. Like if you had the E and the G you could get a volume but if you didn't have B, the projection from above would look like a square instead of a B. Likewise, even if you had 3 views, it still won't be correct along some other direction, even though it may be "right" along those particular, exact three projection angles. That's why CT uses hundreds of projections instead of only 3.

Answer 3

[Matt J]

For a general object, no it's not possible. For an object with some known structure, like a sphere, then yes: measure the radius of the projection and you're done.