Sony revolutionizes the sensor world?

noirist said:
Yes, ideally you would get a real measurement for every point. BUT THEY CAN'T.
Click to expand...
Well they can - for static images. It gets true values for R,G,B values for every photosite, it just gets them one after the other rather than all at the same time. On a ststic image then there is no question that it will output true "4:4:4" resolution equivalent to the sensor dimensions, and by direct readout.
noirist said:
I don't think there will be terrible edge effects on motion because the sampling rate is so much higher than the shutter.
Click to expand...
And yes indeed to that. In principle, with such a sampling rate, then movement shouldn't be a problem either. Each sample of each colour will be many samples accumulated spaced throughout the exposure time.

The principle is sound - whether it's possible to engineer it quite another matter! :)
 
harddrive, if they can't get all three rgb values for each pixel at the same time for the same duration of time, then they are sampling. I guess I should have said, "Yes, ideally you would get a real measurement for every point [in every time slice]. BUT THEY CAN'T.) Since that sensor is sampling the RGB values, then the final image will necessarily require interpolation.
 
noirist said:
harddrive, if they can't get all three rgb values for each pixel at the same time for the same duration of time, then they are sampling. I guess I should have said, "Yes, ideally you would get a real measurement for every point [in every time slice]. BUT THEY CAN'T.) Since that sensor is sampling the RGB values, then the final image will necessarily require interpolation.
Click to expand...

OK, so given 2 images, one where the samples are taken for the full shutter speed time (say 1/50) and one where there were 1000 samples (or pick your number) distributed between R,G, and B over the same time period and then averaged, where do you think you would see a difference in the 2?
 
Well, for one, if the sensor could really measure all three color values over the full shutter speed then it would capture significantly more light (eg. 3x) which means much better low light performance. And if there was no interpolation then the spatial resolution would be higher and there would be less aliasing.
 
noirist said:
Well, for one, if the sensor could really measure all three color values over the full shutter speed then it would capture significantly more light (eg. 3x) which means much better low light performance. And if there was no interpolation then the spatial resolution would be higher and there would be less aliasing.
Click to expand...

How do you derive the aliasing conclusion?
 
noirist said:
Well, for one, if the sensor could really measure all three color values over the full shutter speed then it would capture significantly more light
Click to expand...
Yes, that's true, and it's why I'm sceptical of SOME of the claimed performance advantages for it. If the fundamental sensor dimensions are (say) 1920x1080, I've no problem it will give an effective resolution equivalent to a 3 1920x1080 chip design. But because of the time slicing it won't be as sensitive. For a given colour and photosite position the effective exposure time will only be a third the amount. Compared to bayer, the photosite size will be bigger (so an advantage) but the accumulated exposure time per frame less (so a disadvantage) - I'd expect them to cancel out.

But if the basic chip is 1920x1080, it will give full 1920x1080 "4:4:4" colour resolution with minimal processing - so in that respect is comparable to a 3 1920x1080 chip design. If they can get the movable CFA to work at all.......
 
Aliasing is a more general phenomenon. Moire is a type of aliasing that has a spatial characteristic.

Anyway, the Sony sensor only samples one color per pixel per time slice. So how is it going to fill in the missing colors in each pixel? Currently our best guess is that it fills in the missing colors from the same pixel in adjacent time slices. So if you're taking pictures of a stationary scene, where the color values of the pixels don't change at all, then you won't notice anything. But if the color of a pixel changes across time slices, the sensor won't interpolate correctly, and you will likely see something strange.
 
What I gathered from the description was a large number of samples per pixel per frame per color were being taken, which is why I think this essentially equivalent to concurrent R,G, and B exposures. Given a change in what is being sampled during the shutter period you should have essentially the same effect whether a single exposure or a summation of smaller exposures. Interesting approach at any rate.
 
noirist said:
So if you're taking pictures of a stationary scene, where the color values of the pixels don't change at all, then you won't notice anything. But if the color of a pixel changes across time slices, the sensor won't interpolate correctly, and you will likely see something strange.
Click to expand...
As J Michael says - you would indeed see something strange (coloured fringing on motion) if you only got one each of red, green, blue "slices" per frame - but the theory is that the time of each slice is very short compared to the period of a frame. Or in other words, you get many slices for each of the colours during a frame.
 
The minimum exposure for the camera is three time slices. Imagine you're taking a picture when a flash goes off very briefly, just enough for the first time slice. Then all the pixels sampled in the first time slice will have a different white balance than the pixels sampled in the second and third time slices. The red pixel values from the first time slice will be different than the red pixel values from the second and third time slice. Same for green and blue. And you will get weird inconsistent colors in your exposure. Aliasing is inherent in the reconstruction of a sampled image. The new Sony sensor is sampling the pixel values. It's going to have aliasing. I'm surprised that this is controversial.
 
But for each colour, the charge that gets accumulated each slice will simply go into a charge "bucket" with that from all the other slices. So not a lot of difference from the normal charge accumulation process over the course of a normal frame - it will just happen in a hundred or so bursts rather than continuously. The idea is three buckets per photosite - one for each colour - which get added to cyclically, one after the other.

As far as "imagine you're taking a picture when a flash goes off very briefly....", then yes, good point in principle, but the time slices are supposed to take 1/16mS each, or 3/16mS for a complete R,G,B cycle. I believe a typical electronic flash is about 1mS, in which case equivalent to about 5 full R,G,B cycles, so any wild colour differences due to a flash should be minimal. (And only last one frame anyway.) The nett end result could likely be better than the result of a current rolling shutter response to electronic flash.
 
You must log in or register to reply here.
Back
Top