photoguy99
No longer a newbie, moving up!
- Joined
- Mar 20, 2014
- Messages
- 1,485
- Reaction score
- 313
Yeah something like that. Absent a lens each pixel calculates the convolution of the ambient light falling on the sensor with a sinc() or something that looks like a sinc() close enough to be a basis vector in the frequency domain or close.
By varying the properties of the gratings etc across the array you get different sinc()-like functions.
Apparently this is sufficient to effect a change of basis from a straight up luminance array (considered as a vector) to something pretty close to a similar 2d array (also considered a vector) in some frequency domain space - or close enough.
With real sinc() functions this is just Fourier analysis. These things are not quite sinc() but whatever. The ideas are the same, the details probably get a bit mucky.
I honestly don't really understand what it does with a lens included in the system. The blending of spatial and frequency domains makes my fragile model collapse.
By varying the properties of the gratings etc across the array you get different sinc()-like functions.
Apparently this is sufficient to effect a change of basis from a straight up luminance array (considered as a vector) to something pretty close to a similar 2d array (also considered a vector) in some frequency domain space - or close enough.
With real sinc() functions this is just Fourier analysis. These things are not quite sinc() but whatever. The ideas are the same, the details probably get a bit mucky.
I honestly don't really understand what it does with a lens included in the system. The blending of spatial and frequency domains makes my fragile model collapse.