Go Back   Steve's Digicams Forums >

LinkBack Thread Tools Search this Thread
Old Jun 25, 2010, 10:17 PM   #1
Senior Member
ac.smith's Avatar
Join Date: Dec 2006
Location: Davenport, IA
Posts: 2,093
Default ISO and Sensor Signal Amplification

I ran across a claim recently that suggested that the amount of gain (which is what raises the ISO above the base level) in the amplification stage after the sensor is read but prior to A/D conversion is adjusted in whole stop steps such as 100, 200,400 etc. and intermediate values such as 125 or 160 are arrived at by interpolation in later processing.

Can anyone confirm or deny the claim and/or comment on the practical consequences if true?

A. C.
ac.smith is offline   Reply With Quote
Sponsored Links
Old Jun 25, 2010, 11:43 PM   #2
JimC's Avatar
Join Date: Jun 2003
Location: Savannah, GA (USA)
Posts: 22,378

Here's one thread on the subject:


From what I gather from reading threads on the subject, you're probably better off sticking with whole stop increments when setting ISO speed as not to negatively impact Dynamic Range, depending on how a given manufacturer approaches it.

In addition to intermediate ISO speeds, you also see odd issues with some Sony models like my A700 if using ISO 100, where the camera is basically overexposing ISO 200 (the sensor's actual base ISO speed) by a stop, and then pulling the exposure (decreasing the values associated with each pixel to make the exposure look correct), which can lead to loss of DR in the highlights (clipping them because the original photo before the adjustments was overexposed by a stop)..

You also see the opposite issue when you go above a given ISO speed with most cameras (since the amplification is often performed after the A/D conversion at the highest ISO speeds available).

For example, a camera's ISO 3200 setting may be underexposing ISO 1600 by a stop, then pushing the exposure (multiplying the values associated with each pixel to bring the exposure back up again in order to simulate more analog amplification), leading to loss of Dynamic Range in the Shadows.

So, depending if the intermediate ISO speeds are being "pulled" or "pushed" from the nearest whole stop increment, you could lose DR in the Highlights or Shadows respectfully, if it's being done after the Analog to Digital Conversion (and you'd probably need to analyze each camera on a case by case basis to tell, but some analysis I've seen suggests that's the case, at least with some camera models).

Personally, I use one stop increments until I get into very high ISO speeds. Then, I go to 1/3 stop increments (since the higher ISO speeds are really just underexposed images that are pushed anyway, once you get into the so called "expanded" or "boost" ranges with most cameras).
JimC is offline   Reply With Quote
Old Jun 26, 2010, 9:59 AM   #3
Senior Member
TCav's Avatar
Join Date: Sep 2005
Location: Washington, DC, Metro Area, Maryland
Posts: 13,826

Interpolation is a fairly complex mathematical process that computers don't actually do very well. The simplest and fastest way to adjust the "ISO" of an image is to bit shift. For instance, if a single pixel has a digital value of 127 (0001111111 in 10 bit binary numbers), you can double it by shifting all the bits to the left one place (0011111110 in binary, or 254 in decimal numbers.) It's easy and fast, and when you have to do it for every single pixel in a multimegapixel image, you want something that's easy and fast.

The problem with doing it that way is that you start getting color banding. For instance, if adjacent pixels have the values 126 (0001111110 in binary), 127 (0001111111 in binary), and 128 (0010000000 in binary), and you double them all, you get 0011111100, 0011111110, and 0100000000 in binary, or 252, 254, and 256 in decimal. Notice that, using that process, there will never be any pixels with values of 253, 255, or any odd number, because each pixel value is the double of an earlier value. By simply doubling the brightness values for all the pixels just one time, color banding won't be apparent, especially since digital cameras work with 12 or even 14 bit data. But when the values are doubled multiple times (i.e. at very high ISO values), the bands sometimes become apparent when an image is supposed to contain a smooth gradient. But at high ISO values, noise appears and noise reduction is frequently used, and the effect of both the noise and the noise reduction reduces the effect of the banding.

Of course, this simple method of adjusting the ISO doesn't allow for 1/3 EV changes, but other forms of binary math can be used to obtain the effect. So all the ISO operations on an image can be accomplished quickly and simply once the image data has gone through the analog-to-digital process.

It's true that many image sensors have amplifiers, but those amplifiers would get quite hot and use a lot of the battery's power to bump up the ISO of a 12MP image to, say, 12800, and that doesn't happen. So I'm inclined to think that most of the ISO processing happens in bitwise operations once the image has been converted to digital.
  • The lens is the thing.
  • 'Full Frame' is the new 'Medium Format'.
  • "One good test is worth a thousand expert opinions." - Tex Johnston, Boeing 707 test pilot.

Last edited by TCav; Jun 26, 2010 at 2:51 PM.
TCav is offline   Reply With Quote

Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On

All times are GMT -5. The time now is 4:19 AM.