Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
05-13-2010, 12:26 PM   #1
Senior Member
jaitas's Avatar

Join Date: Dec 2008
Location: Oulu
Posts: 124
New HDR tech idea ???

This is just it, I was happily driving and then my brains went, bzzzz...

I Started to think pentax in-camera HDR and how it could evolve, then My brains told me this idea that what if cameras sensor had for example five read-out channels... And camera would count bracketin just as now and adjust shutter speed with the longest exposure needed.

Then sensor would pass that light gathering data to all five read-outs... then when ev -2 would complete first read-out would stop reading the data, then when ev -1 and second would stop then ev 0 and third, ev+1 fourth and ev+2 fifth....

That would basically make a HDR image taking possible handheld with reasonable amount of light.... would be even more control than with a raw file...

But yes, now I'm hoping that some of you more technically oriented fellas would tell me why this idea is stupid and why it would not work, because if it would, it would be already in use

Anyway, that's my brainfart for today, enjoy!

-J

05-13-2010, 02:31 PM   #2
Veteran Member
ytterbium's Avatar

Join Date: Jan 2008
Posts: 1,076
Your idea is not quite clear. Did you mean using electronic shutter, to make faster bracketing, with increasing exposure time?
Eg:
-Mirror up
-Sensor reset
-Mechanical shutter open
-1/125
-el. shuttering (sensor readout and reset)
-1/60
-el. shuttering (sensor readout and reset)
-1/30
-mechanical shutter close
-el. shuttering (sensor readout and reset)
-mirror down
This would introduce various artifacts - image shear, moving object tearing, light leak (lines from bright lights as seen in video clips) - basically all the defects as in videos and even more. At full resolution you would not be able to roll shutter at 24fps but much slower. Youd get more like scanning camera.

Or did you mean a way to read the data from sensor without reseting it?
Mirror up->Mech shtr. open->Readout at 0.1[s]->Readout at 0.2[s]->readout at 0.5[s]->Mech str. close->reset->Mirror down.
If this is the case, i doubt its possible. I've too had this idea. Because at each readout the pixel is emptied, so the accumulation starts from zero, instead of continuing exposure.

Unless youre thinking of smaller output resolution (like full HD). Then, if the channels were distributed across pixels, you could have various exposure times for pixels assigned to seperate channels. But the end resolution would be MaxMpix/Bracket steps.
Then you could continue the exposure in other pixels, while pixels in one channel is being red out and reset.

It's rather complex and is heavily dependant on sensor architecture. Weither it has global reset or per channel, are the channels distributed(pixels:123123123) or in blocks(111222333) etc.
05-14-2010, 01:01 AM   #3
Senior Member
jaitas's Avatar

Join Date: Dec 2008
Location: Oulu
Posts: 124
Original Poster
Yeah I was thinking of reading the data without resetting pixel value to zero. And yes, zero technical thinkin by me, it just occurred to me and left me troubled so I had to share

I thought so too that it would be impossible to get readouts from the sensor "during" the exposure to get the lowest ev exposures same time when exposing the longest.

Yeah and I was a little bit unclear in my first post, sorry.

-J
05-14-2010, 01:56 AM   #4
Veteran Member
ytterbium's Avatar

Join Date: Jan 2008
Posts: 1,076
Still the idea is interesting. But there are many other ways to improve HDR performance before getting ill effects.

One would be firing the shutter only, without lowering mirror between bracket shots. I believe most of the time between frame is taken by mirror movement.
The speed at which you can fire just shutter is most likely depends on how fast you can read out the sensor (while shutter is closed). Here additional channels would be usefull.

05-14-2010, 02:23 AM   #5
Pentaxian
Digitalis's Avatar

Join Date: Mar 2009
Location: Adelaide.
Posts: 8,702
some scanning digital backs for large format can do this, of course they aren't good for action and moving subjects.
05-14-2010, 10:02 PM   #6
Senior Member




Join Date: Feb 2010
Location: Írebro
Posts: 207
But even if the sensor data is "emptied" at the reading of it wouldn't his idea still work?

Say you need a shutter speed of 1/100 and also want 1/200 and 1/50.

Cant you then use a global shutter and read a rather dark image of 1/200 second, and save that data.

Then read another 1/200 second, save that data (or add it together with the first data set and save that as a normal image, depending on processing speed.)

Then read another 1/100 second and save that data (see above)

And then depending on the calculation speed make your various shots as:

dark = dataset1
normal = dataset1 + dataset2
bright = dataset1 + dataset2 + dataset3


That would use more ram because you may not want to write the data to the card and then read it back for calculations and such, but still only require 1/50 shutter time instead of 4/200+2/200+1/200=7/200s=1/28.5s.
05-15-2010, 01:55 AM   #7
Veteran Member
ytterbium's Avatar

Join Date: Jan 2008
Posts: 1,076
This would impact quality in the dark areas, which would be made up of underexposed, noisy data.
It would average out a bit, but wouldnt be as good as correct exposure.

You need to accumulated certain amount of charge in a photosite to overcome the base noise level.

There would not be a ram problem, i think, because currently bracketing takes it up anyway.
05-17-2010, 10:23 AM   #8
Senior Member
jaitas's Avatar

Join Date: Dec 2008
Location: Oulu
Posts: 124
Original Poster
So this kind of "interval" reading during exposure from sensor would be sensors feature, and with todays tech impossible... Too bad, hopefully somewhere in the future this kinf of implementation would become possible. :ugh:

05-17-2010, 01:17 PM   #9
Veteran Member
ytterbium's Avatar

Join Date: Jan 2008
Posts: 1,076
I think making better sensor is more conveniet way for user. A sensor that is high dynamic range itself. Producing very low noise and good highlight preserving capability. A way to allow HDR image from single shot.

Maybe a hdr capture mode. Currently sensors operate in linear mode. They might incorporate a modifier circuit, that non-lineary bleeds high charge when enabled. This would preserve highlights.

Research has been done in this area but no devices practically usable for dslr. Here you can see a nice sample from one such camera:
07-07-2010, 03:00 AM   #10
Pentaxian




Join Date: Nov 2008
Location: Bangalore
Photos: Gallery
Posts: 2,447
Interesting idea indeed and I think it's very much doable - I am sure someone will file a patent
07-07-2010, 04:48 AM   #11
Veteran Member




Join Date: Jul 2008
Location: Var, South of France
Photos: Albums
Posts: 1,071
Well, the ultimate HDR sensor would be a time-based sensor, where each photosite records the time needed to attain full capacity...
But you'd then have to wait a little more for dark photosites to fill... And if they don't fill, then the information would be lost.

Other possible solutions are:
- the Fuji SR system,
- the "new" Kodak RGBW filter array (where instead of having two green photosites for a red and a blue, you have one green and one white, the white being more sensitive to light).
- A multiple reading using the CCD architecture (do a reset of the sensor during the exposure so as to obtain 2 images, the first one being pushed in the masked area then read, while the second one is kept safe at the end of the exposure by closing the mechanical shutter).
- locally-adapted CMOS photosites (mimicking the actual behavior of the eye where a light receptor is aware of its surroundings and its immediate history).

It's to be noted that the most efficient way to do HDR so far seems to be by using multiple images, so maybe we will go back to CCDs after all!
07-07-2010, 04:56 AM   #12
Pentaxian
Lowell Goudge's Avatar

Join Date: Jan 2007
Location: Toronto
Photos: Gallery | Albums
Posts: 15,396
I think the ultimate HDR sensor, and let's assume I said this first and can patent the idea, is to have a different sensor design all together.

Let's assume we have a sensor with for example 3 times as many cells, and we can program adjacent "like cells" to be processed differently.

One example would be to make a single pixel out of an average surrounding it, to reduce noise, but another would be to make 3 images at the same instant, each with a different exposure via changing the iso setting. this way you get 3 images at the same instant, therefore no motion blurr, and no impact of subjects moving between HDR frames. You know, the little things like trees and flags moving in the wind, that can't be done in HDR any way or people walking thorugh the frame, waving their hand etc.

You can't really resample the sensor while the shutter is open for different gains. you would need too much time.
07-07-2010, 05:24 AM   #13
Veteran Member




Join Date: Jul 2008
Location: Var, South of France
Photos: Albums
Posts: 1,071
QuoteOriginally posted by Lowell Goudge Quote
You can't really resample the sensor while the shutter is open for different gains. you would need too much time.
There are patents using this method, actually...
In a CCD sensor, at the end of the exposure, you shift the charges in a masked area (electronic shutter), then read them at your time.
But meanwhile, the main imaging area (reset by the shift) can still acquire photons!

So, if you keep the sensor exposed after the charge shift, a "second" exposure will take place.
You then have to close the mechanical shutter to prevent further accumulation once the desired exposure time is reached, while the first image is still being read from the masked area.
Then, you shift charges a second time from the imaging area to the storage area, adjust the gain, and read them...

You'll have to take into account the curtain traveling speed for the first shift, so as to have identical exposure time for all the lines...
Granted, this method could be problematic with shutter speeds above the sync speed (still possible, but with highly precise synchronization between curtain travel and charge shifts).

But in the end, for a 1/200 equivalent exposure, you'll end up with, say, a 1/160 total exposure time with a 1/200 image and a 1/800 image.

Last edited by dlacouture; 07-07-2010 at 05:35 AM.
07-07-2010, 05:24 AM   #14
Pentaxian
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,863
Ramos

QuoteOriginally posted by jaitas Quote
But yes, now I'm hoping that some of you more technically oriented fellas would tell me why this idea is stupid and why it would not work, because if it would, it would be already in use
jaitas, you didn't explain your idea extremely well but I think I got it (didn't read further comments though).

Well, just to straighten up some language: a read-out channel connects a pixel with a pin on the sensor chip. I.e., each pixel is connected to only one channel. So, what you mean is something different than a read-out channel, let's call it demultiplexer.

The idea you describe would need alteration to work. In the K-7, to read-out a pixel takes ~50ns. Reading out 14.6MP therefore takes 0.7s. Which is why the K-7 reads out 4 pixels in parallel and therefore only needs 0.18s. In theory, you could indeed read out 5 values corresponding to 5 different levels of light collection if you started collection asynchronously over 0.7s. But even though an individual pixel would be exposed a short period of time only, this would create an inacceptably awful rolling shutter effect. And it would require additional electronic shutter circuitry on the sensor which is known to create further problems. dSLRs use a mechanical shutter for a reason.

So, it wouldn't work this way.

However, what would work is this:

1. To read-out/reset a pixel value and do an A/D-conversion right in the sensor. Either per pixel or per column.

2. Having an A/D-converter per column and 4 read-out channels would accelerate the read-out operation to 0.04ms (1/25000s). Now fast enough for synchronous operation. Sony Exmor sensors already sport column A/D converters, so this is becoming standard now.

3. Now (and this is new), add 8kB memory to each of the 5,000 column converters (a total of 40MB) plus one adder to keep adding up true 24 Bit raw values until the shutter closes. Not just 5 values but as many values as you want. You may typically want to add 1000 values to add dynamic range from 14 bit to 24 bit. Even at 1/8000s, you could still add 2 values and gain 1 stop DR!

4. Read out the 40MB of 24 bit raw data using a normal memory interface and normal operation speed.


The latter approach creates a sensor with arbitrarily low ISO and arbitrarily high dynamic range. And never again blown-out highlights. The -4,-2,0,2,4 EV bracket at ISO100 corresponds to this chip configured to ISO 6. It's coming. Processors typically now have about embeded 8MB L3 caches. 40-50 MB would still be too expensive for a consumer product. But it is feasible now and affordable soon enough. Some Itanium processors already have 24MB L3 cache. Start saving for CMOS sensors with embedded image memory. I call it RAMOS (Random Access cMOS image sensor). It will really have an impact on artistic freedom!

This idea either is patented somewhere or prior art (because I already described it somewhere else ).


P.S.
Once the above is done, you can add a little bit of extra circuitry to digitally add the output from neighbouring column converters (binning) and create a truely downsampled video signal right in the sensor. No more subsampling. It will be a totally new level of video quality which would even outperform a current RED camera. And the rolling shutter effect would be gone too, of course. And EVFs which start to overwhelm become feasible. And contrast AF starts to beat phase AF to death. After FF, I predict RAMOS to be the next big thing making entire SLR user bases upgrade their equipment.


P.S.2.
If you carry this idea further, you'll get one A/D-converter per pixel first and then you'll end up with a counting photon detector in each sensor cell.

Last edited by falconeye; 07-07-2010 at 06:19 AM.
07-07-2010, 10:52 AM   #15
Site Supporter




Join Date: Mar 2008
Location: Prince George, BC
Photos: Gallery
Posts: 2,295
Thanks, Falk. Now I know where we are heading. After viewing Doctor Bussard's 2006 lecture yesterday, it now seems that physics and engineering has the capacity to pull our species out of its own deeply dug holes.


**edit for proper year**

Last edited by jbinpg; 07-07-2010 at 11:13 AM.
Reply

Bookmarks
  • Submit Thread to Facebook Facebook
  • Submit Thread to Twitter Twitter
  • Submit Thread to Digg Digg
Tags - Make this thread easier to find by adding keywords to it!
brains, camera, data, dslr, ev, five, hdr, idea, photography, sensor
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
New Tech to change digital? ozlizard Pentax News and Rumors 6 03-25-2010 12:06 PM
HDR Software - Dynamic-Photo HDR mithrandir Digital Processing, Software, and Printing 3 03-03-2009 08:46 PM
Aqua Tech RussellW Pentax Camera and Field Accessories 5 11-20-2008 04:31 PM
Sigma 17-70 tech question jgredline Pentax SLR Lens Discussion 9 01-28-2008 06:55 PM
Happy Tech Geek from IL sketchin13 Welcomes and Introductions 6 05-24-2007 10:20 PM



All times are GMT -7. The time now is 02:23 AM. | See also: NikonForums.com, CanonForums.com part of our network of photo forums!
  • Red (Default)
  • Green
  • Gray
  • Dark
  • Dark Yellow
  • Dark Blue
  • Old Red
  • Old Green
  • Old Gray
  • Dial-Up Style
Hello! It's great to see you back on the forum! Have you considered joining the community?
register
Creating a FREE ACCOUNT takes under a minute, removes ads, and lets you post! [Dismiss]
Top