A photosensor is (roughly speaking) sensitive only to the intensity of light falling on it, and not the angle the light comes from. A plenoptic camera (this is what Lytro are selling) trades off resolution against the ability to discriminate between angles.
Take an ordinary camera sensor, with (let's say) 16MP in a 4000x4000 array. Group them into (let's say) 8x8 blocks, of which there are 500x500. Now put a little lens in front of each block, with focal length equal to the distance from lens to sensor.
A light ray reaching one of those blocks will end up on one of the block's pixels; which ray depends on (not the exact position at which the ray meets the array of lenses, but) the angle at which it's travelling.
Now, imagine an image that's not quite focused correctly on the camera sensor. What that means is that for each point of the object you're imaging, you get a cone of light rays that are converging towards some point either in front of the sensor or behind it. With an ordinary camera, that just gives you a circular blur and you're screwed. With a plenoptic camera, you can tell what angle the light in that circular blur was coming in at, which means you can determine where it would have gone if the sensor had been further back or further forward, which means you can reconstruct what you'd have got if the focus had been different. (What if it is focused perfectly? Well, you still get to know the distribution of angles from which the light is reaching the sensor, which means you can work out how the image would have been blurred out with a different focus.)
The main price you pay for all this is a severe loss of resolution: your output "pixels" are the blocks of pixels on the sensor. So the physical size and noise level of the sensor, and the size of your raw image files, are those of (in my example) a 16MP sensor, while the final image is (in my example) only 1/4 MP.
So would you say this is more of a gadget camera rather than an everyday use camera? I have just paid out for a pretty decent point and shoot, would I replace it with one of these camera's or opt to use both?
I do love the concept of them but if all I am ever going to get out of them is a 6x4" pic to print at the end of the day then it's only really useful for "playing" with?
I can't see a Lytro camera replacing a "pretty decent point-and-shoot"; they're very different.
The post-hoc refocusing feature is neat but I can't think of many practical or artistic purposes for which it would actually be more useful than having way more pixels. But I'm not an expert photographer and could easily be wrong. One kind of situation in which it might be useful is where an object is moving rapidly towards or away from the camera, but not moving much laterally. Then the freedom not to worry about rapid and accurate focusing might be useful. As soon as your object is moving laterally too, though, the limited resolution is going to bite you: you've avoided having to locate the object accurately in z at the cost of needing to get x,y right, so to speak.
There may be non-gimmicky applications for which Lytro is The Right Thing. Right now, I can't think what they'd be. Depth measurement, perhaps.
Spatial resolution will increase at the same rate the process allows the sensor elements to shrink, so, in a couple years, a Lytro-like imaging sensor will have the same resolution as a top-of-the-line DSLR.
I'm not sure if I understand this concept correctly. Can you say that if they take 4 pixel blocks the depth of field interpolates over 4 pixels? So each screen-pixels has 4*4 corresponding sensor-pixels?
A photosensor is (roughly speaking) sensitive only to the intensity of light falling on it, and not the angle the light comes from. A plenoptic camera (this is what Lytro are selling) trades off resolution against the ability to discriminate between angles.
Take an ordinary camera sensor, with (let's say) 16MP in a 4000x4000 array. Group them into (let's say) 8x8 blocks, of which there are 500x500. Now put a little lens in front of each block, with focal length equal to the distance from lens to sensor.
A light ray reaching one of those blocks will end up on one of the block's pixels; which ray depends on (not the exact position at which the ray meets the array of lenses, but) the angle at which it's travelling.
Now, imagine an image that's not quite focused correctly on the camera sensor. What that means is that for each point of the object you're imaging, you get a cone of light rays that are converging towards some point either in front of the sensor or behind it. With an ordinary camera, that just gives you a circular blur and you're screwed. With a plenoptic camera, you can tell what angle the light in that circular blur was coming in at, which means you can determine where it would have gone if the sensor had been further back or further forward, which means you can reconstruct what you'd have got if the focus had been different. (What if it is focused perfectly? Well, you still get to know the distribution of angles from which the light is reaching the sensor, which means you can work out how the image would have been blurred out with a different focus.)
The main price you pay for all this is a severe loss of resolution: your output "pixels" are the blocks of pixels on the sensor. So the physical size and noise level of the sensor, and the size of your raw image files, are those of (in my example) a 16MP sensor, while the final image is (in my example) only 1/4 MP.