View Full Version : Video camera sensitivity
September 1st, 2010, 08:54 AM
It is very common for the "sensitivity" of dedicated video cameras to be stated in terms of the minimum scene illuminance (ambient lighting) under which they can deliver a "usable" image, and to state that illuminance in the SI unit, lux.
I thought it would be interesting to compare this with the terms in which we commonly think in the still camera world.
A common rating for dedicated video cameras is 0.5 lux. In APEX terms, this corresponds to Iv -7.1. This is said to be about the illuminance afforded by the light of the full moon under good conditions. My incident light exposure meter provides scale markings down to 0.125 lux, with good readability down to 0.5 lux.
Under the "standard exposure equation" with scene luminance known (often described as the "incident light metering" form of the equation), for a camera sensitivity of ISO 400, this scene luminance would call for a photographic exposure of EV -0.1 (about f/2.0 at 2 sec).
By way of comparison, an exposure of 1/30 sec with an aperture of f/2.8 is Ev 7.9.
Thus, it would seem, a dedicated video camera with a rated sensitivity of a scene illuminance of 0.5 lux would seem to be 8 stops more sensitive that a still camera at ISO 400, 1/30 sec, and f/2.8.
Lest my math have gone wrong, or my favorite exposure calculator, I had my light meter do the same reckoning, with the same result.
I am startled by this result.
To take a properly exposed still picture with a scene illuminance of 0.5 lux (assuming the normal range of scene item reflectance) would seem to require perhaps this setup:
ISO 3200, f/2.0, 1/2 second.
How can this be?
September 1st, 2010, 09:30 AM
I have always wondered about this. Perhaps the frames are considered as a subunit of overlapping sliced sections of a cube so that data on each feature is accumulated.
September 10th, 2010, 03:52 AM
Hello Doug and Hello Asher,
I read the original post a couple of days ago typed a reply - session timed out and I lost the response so here I am again!!
First of all I find the original question most interesting.
The initial finding that a video camera is 8 stops (x 256) more sensitive than a still camera puzzling and prompts one to get answers to some basic questions (of the manufacturer) about the video camera itself.
How do the manufacturers define sensitivity ?
How is a video image described? ... Picturing a static scene and integrating a number of exposures? ... I don't know.
The video camera system "has it all over" the single frame cameras in that it can coherently integrate many images to reduce the effect of noise - now to do this will make an unfair comparison!!
Very simplistic noise analysis:
The video camera has 3 sensors. So the noise power (uncorrelated) in an image pixel is a weighted sum of the separate noise powers. But in a single sensor still camera each image pixel is the result of the combination of nearest neighbor sensels. This implies that there are more contributory noise sources in the case of a still camera.
The video cam is looking increasingly better in low light. But we are still not comparing apples with apples. To do that we really much more info from the manufacturers.
September 11th, 2010, 10:08 AM
I hope you guys can get more information on this. It is important as to where one might best spend limited funds in getting videos in low light. One needs both increased sensitivity but also protection against blowout of the bight areas. For me that's the relatively low lit stage for music or ballet performance with lights from above giving harsh specular highlights on the forehead and face, while eye sockets are relatively dark.
September 11th, 2010, 05:19 PM
I'll take an " action item " and have a closer look at this issue of low light and see how it applies to our situation.
Interestingly there is a a huge body of knowledge "out there" on the use of video cameras being used to make a still image, this is the specialist field of " Lucky Imaging ". It gets its name from the statement, " You are lucky if you get a good image !"
For any interested readers I'll refer you to some very sophisticated and free (there are some very talented and generous people "out there") software named REGISTAX :
( http://www.astronomie.be/registax/ )
September 11th, 2010, 06:38 PM
This goes along with my idea that videos are really cubes of image data where the same thing is imaged over and over again with some movements needing alignment and various translations being made of angular movements. Essentially, I imagine a video as not a series of still images but rather a series of overlapping stacks of image data. So there's incredible potential for increasing resolution and decreasing noise.
but is that anything to do with how images are processed for video?
September 12th, 2010, 01:02 AM
Re: So there's incredible potential for increasing resolution and decreasing noise.
Answer: Definitely !
When one talks about stacking that's what the eye/brain combo does. Once, I remember looking at a video image thinking how good it looked but a single frame looked so noisy that it was disappointing.
I was looking at some of the newer Sony (v.expensive) and Canon (somewhat cheaper) video cams and the advertising blurb showed very very good low light performance but they did not tell the whole story this is where I would like to dig deeper.
BTW. I take issue with the way the marketing departments of the manufacturers twist terminology to make things sound so much better.
Your mail box is full - does not accept PM's
September 12th, 2010, 05:55 AM
Once, I remember looking at a video image thinking how good it looked but a single frame looked so noisy that it was disappointing.
Indeed, a very important point, and one that may in fact be pivotal to the issue in this thread.
September 13th, 2010, 03:06 AM
Yes its got back to the problem of signal to noise ratio with the added complication of many many frames being used to constitute an image. Your original post got me intrigued and prompted me to do some "fact finding". As my mother (bless her heart, 88 and going strong!) would say "He's got a bee in his bonnet".
I'll take a close look at the subject of video images and how we perceive them etc etc. etc.
Then try to marry this with image quality.
I'll keep you posted.
March 15th, 2011, 09:42 AM
The reason why video cameras boast abilities to film in almost complete darkness is mainly commercial. At the minimum illumination levels, the image is usually very noisy and almost devoid of colors.
However, the basic sensitivity of video cameras is usually relatively high, because:
-they have relatively big pixels, since they have few of them on relatively large chips (*)
-the 3 chips models lose less light than a bayer array
-their lenses are usually pretty fast (f/1.6 is common)
Sensitivity levels are not directly comparable, because they depend on the gamma curve (which can be adjusted on the better video cameras), but one can try to simply compare readings between a DSLR and a video camera. I tried to do that with a Sony AX2000, and I find a sensitivity of about 640 ISO at 0 dB. A quick search on the net showed that other people found similar values for similar cameras, for example the Canon XH-A1 is reported at 340 ISO at 0dB. For simplification, I will take ISO 400 as a representative sensitivity at 0dB for video cameras.
Video cameras have a function called "gain", which doubles the apparent sensitivity every times the gain is increased by 3 dB. So we have:
GAIN - ISO
0 - 400
3 - 800
6 - 1600
9 - 3200
12 - 6400
15 - 12800
18 - 25600
21- 51200 (maximum for the AX2000)
At 21 dB, the image is still usable, the colors are still visible, but the resolution is visibly compromised by noise reduction. If one needs better images, it is better to limit oneself to, say, 9 or 12 dB gain.
Small consumer models may go to even higher gain values, typically 24 dB (102400 ISO), and silently reduce the shutter speed to 1/25 instead of 1/50s (1/30 vs 1/60 in NTSC), effectively doubling another time the apparent sensitivity if one does not check the shutter speed.
Now, if we had a camera with 102400 ISO, f/1.6 and 1/25s, we could indeed take pictures in very dark places indeed. Just as with a camera with 3200 ISO, f/2.0 and 1s, which is about what was found in the first message of this thread (we have a factor of 2, which is basically the accuracy of the various calculations).
(*) On large chips: please consider that DSLRs discard typically 4 lines out of 5, effectively reducing the active size of their chips by the same factor. A 2/3" video camera thus has the same active surface as a four thirds sensor, but typically a maximum of 2 millions pixels per chip (HD).