here:
http://www.dsp.ece.rice.edu/cs/cscamera/
and in english,
http://news.bbc.co.uk/2/hi/6263551.stm?lsm
http://www.dsp.ece.rice.edu/cs/cscamera/
and in english,
http://news.bbc.co.uk/2/hi/6263551.stm?lsm
0.001 megapixel camera concept
- Thread starter leews2001
- Start date
- Status
The mirror elements not pixels meh? Just the reverse of the TI DMD only mah.. Then how to control shutter speed?
dunno much about the concept, but you have some really amazing pictures in your flickr gallery!
the mirror elements does not act as pixel elements. They act as basis functions (random or predefined. if its simply on/off mirror, it could be like a hadamard).
the mirror elements+2nd Lens combo behave like a projection function, and the single ccd "array" records, equivalently, the projection results, which mathematically is a scalar coefficient.
to reconstruct the image via linear transformation, they need say a few thousand coefficients, thus a few thousand sequential read off from this single ccd.
in jpeg compression, we are also using only a few thousand "readings" out of the millions to get back the "degraded" image.
dunno much about the concept, but you have some really amazing pictures in your flickr gallery!
thank you for visiting.
If you have a few million mirror elements, you will need an equivalent number of "reading"s in order to get the image. Hence the processing capability of the camera is tremendous.
Not true. The concept of the above mentioned model is to achieve highly skewed distribution of computation load in the image acquisition-encoding-reconstruction system.
from TFA, no processing is done (as per element basis) on the mirror elements, since they are optically projected onto a single element sensor. effectively, the acquisition/coding "camera" system only needs to perform a few thousand readings with some projection calc. However, the bulk of the complexity is shifted to the reconstruction/display stage, of backprojecting those thousand readings into millions pixels.
therefore such model might be beneficial for a ultra-compact, low-power mobile image acquisition tool which can be deployed in mass and cheaply(?).
of course, it seems more of a academic curiousity for now, at least in the near future.
If you have only one sensor, the light collected on the sensor is no more than the result of interference of all the photons detected on the sensor. Hence in order to reconstruct the image, you need to know the quantum mechanical states of each photon detected by the sensor. From the article, I assume that each mirror transmit a photon or an ensemble of photons of similar states. So each mirror constitutes a variable to the equation, and since there are millions of mirror,it will correspond to a million discrete variables. That's why to decipher each variable, you need an equal number of readings to construct a simultaneous equation.
it is unnecessary to bring in quantum theory here, which would not be helpful either since we are still well within the domain of "macro" in this problem.
there is no need for perfect inversion/reconstruction in this set up , nor it is the objective of th e experiment. thus the image reconstruction from the 1ccd reading can be viewed as a restoration problem, which allows for an ill-conditioned system of linear equations.
there are many methods to solve such system from here on. but as discussed previously, it will be computationally expensive (for the decoding/reconstruction side).
Hmmm..., ot a bit.
IIRC, (maybe ten years back or more), when the Compact Disc Player have a new invention, 1bit D/A converter instead of 16bits (or 18bits, 20bits, 24bits) D/A converter.
Sound too good to be true. But the cost of 1bit D/A converter is much much cheaper than their multi bits brothers. And ears (for most of us, but not for those golden ears) can't distinguish the quality.
I am not following Hi-End (above Hi-Fi) anymore, so anyone know who is the winner now? 1 bit D/A converter or old multi bits D/A converter still own the crown?
Regards,
Arto.
IIRC, (maybe ten years back or more), when the Compact Disc Player have a new invention, 1bit D/A converter instead of 16bits (or 18bits, 20bits, 24bits) D/A converter.
Sound too good to be true. But the cost of 1bit D/A converter is much much cheaper than their multi bits brothers. And ears (for most of us, but not for those golden ears) can't distinguish the quality.
I am not following Hi-End (above Hi-Fi) anymore, so anyone know who is the winner now? 1 bit D/A converter or old multi bits D/A converter still own the crown?
Regards,
Arto.
Perhaps, it does not require perfect reconstruction of the image. But at least the states of the light from certain mirror elements must be determined. However, suppose we have 3 variables but only 2 equations (readings), the result is an infinite number of solutions. Which means to say, you will not be able to determine the colour (for example) that each mirror is capturing, since the colour of the light captured by 1 mirror is "dependent" on the colour of the other?
- Status
