I wasn't sure where to put this. It's not at all Mars-specific but it is 'tech general'. I was wondering if a camera could be designed to respond to light levels in a way more similar to the human eye. Why are all our current cameras linear and therefore highly intolerant of wide brightness differences? Is this not a most unfortunate handicap in space where the surface brightnesses of objects are so hugely variable?

http://www.globalspec.com/FeaturedProducts...an_eye_/36317/0

Could there be a useful way forward here? Has this come up before? If so could somebody bring me up to speed?

Well - high dynamic range is the buzz phrase in the animation world - and even with a bog standard camera you can get HDR output by doing multiple exposures at multiple exposure lengths and then adding them all up to get a cool image. The sort of thing you see being used for an example is the inside of a cathedral, and you can drag the brightness up and down from seing all the cathedral lit in dark candle light, and then drag it as far as totally black in the cathedral but you can see clouds outside the windows.

How much that might apply to something worthwhile or useable for spaceflight...I don't know.

Doug

Thanks for the response Doug. I've seen the exposure-adding trick used to very good effect on total solar eclipse images, but it seems cumbersome and indirect. The advertisement I linked to seems to say they have a chip with a logarithmic light response. What isn't clear (at least to me) is whether they mean they have a logarithmic-responding substance as such or whether they are doing fancy processing within the 'chip' to get the log output.

I understand that any logarithmic camera would not make the most efficient use of every incoming photon and therefore you wouldn't want to use it for most space telescope applications. I'm thinking of targets like the solar corona, Saturn's rings, planetary surfaces generally, seeing into shadows at Victoria Crater or - an even more extreme case - on the Moon. The question is whether these logarithmic 'chips' can be reduced to pixel size and arrayed CCD fashion to do real imaging. I have no idea, but it's exciting to imagine one day having cameras that give us natural-vision images.

ngunn, to have an image sensor with log response would be great for everyday applications, I often notice how my eyes are able to cacth details in highly contrasted subjects that my DSC camera is absolutely not able to reproduce, unless making multiple exposures...
This is a feature of "active pixel" CMOS sensors under study now (they make amplification and ADC conversion inside each pixel, which is not possible with a CCD).
Anyway, for most scientific applications in your examples, I think is sufficient to have an high dinamic range sensor, with more than classical 8bit output precision. This is what normally happens, for example MER use a 12bit ADC and, before transmission, images are dinamically compressed through a "look-up table" which makes some kind of square-root/Gamma correction and reduce lenght to 8bit while mantaining a good photometric precision at low light levels; isn't a logarithmic sensor but is a good compromise IMHO.

Very interesting, dilo. The 'active pixel' does begin to sound more like a retinal cell in its function. My experience in the very different field of seismic recording introduced me to floating point amplifiers in which a constant bit resolution was available for any size of signal just by scaling the whole shebang up or down by powers of 2 before quantising. Of course this depended on the analog part - the microphones - having a stupendously wide dynamic range in the first place. I'm not sure how far we are from having an optical equivalent. I think it's wonderful how good the MER cameras are but as you say they're still not logarithmic. Many of us spend longer nowadays looking at photographic images than at reality and we have learned to 'translate' mentally. 12 bits, though, is still less than 4 orders of magnitude in brightness. With my naked eyes I can look at a sunlit iceberg or the Andromeda galaxy. How many bits is that?

A back-of-the-envelope answer to my own rhetorical question - about 24.

You cannot see the andromeda galaxy and an sunlit iceberg at the same time. the effective simultaneous dynamic range of the eye is somewhere around 16 bit.

but if you want to make an animation from night to day in antarctica in the full dynamic range of the human eye including all its adapted states you might want to have it stored in floating point.

/Mattias

Very interesting info, thanks.
Regarding the quoted comment I suggest that the animation scenario you mention is not hugely unlike cruising around Saturn's ring system taking images from various angles - including straight into the sun - of objects ranging from icebergs to faint wisps of gas. Are you aware of any optical imaging system at the present time that does use floating point?

I think what Malmer was getting at is if you want to capture a physical property of the scene such as radiance, I/F, whatever, you're better off with 32 bit floating point. It has 6 or 7 decimal digits accuracy plus an exponent so you can store a large dynamic range. That doesn't mean floating point cameras -- you could still take your antarctica animation with 12 bit DN depth, but store the calibrated I/F into the animation using floating point so you don't lose any dynamic range a 12 bit quantization would impart on the whole night-day brightness range. This would still leave each frame essentially as 12-bit (or you could do that fancy HDR multiple-exposure thing to bring that up to say 24 bits), but those bits would be optimized for the current scene brightness.

Yes that's more or less what I understood. To simulate what the eye is doing you would need a system with 16 bit resolution that could operate over a dynamic range of 24 (or 32) bits. Think of a helmet-mounted camera worn by an astronaut exploring in and out of shadows in a crater near the lunar south pole, or a Europa rover crossing sunlit ice but also peering into deep caves and crevasses. Having some kind of non-linear process at the analog (light receiving) end would be a great help but I realise that is a separate issue from floating-point analysis of the signal.

After some more digging:

http://www.kip.uni-heidelberg.de/vision/pr...dr/divichi.html