1. Reminder: Please use our affiliate links to get to your favorite stores for holiday shopping!

Oh please, not another thread about dynamic range.

Discussion in 'Image Processing' started by RobWatson, Aug 5, 2011.

  1. RobWatson

    RobWatson Mu-43 Hall of Famer

    Tonal quality and dynamic range in digital cameras

    Sort of ... I found some gaps but all in all very interesting read. Particulary the parts about setting up tonal curves to match well known curves or films.

    One missing bit (pun?) was the reason why low ISO generally has higher DR is becasue that is when the sensors is "filled" to capacity. Any additional exposure will cause 'defects' in the form of blooming spikes, anti-blooming circuitry licks in and can do stange things, etc.

    One thing that I have not seen (or done myself) is a detailed work up on the noise when going from raw linear data then through a non-linear transfer curve (like a gamma or "S" type curve).

    One of these days ... raw+JPEG photon transfer curves for each ISO just to put the lid on it. If only the E-P1 could be tethered then a simple script and the task is automated.
     
  2. RobWatson

    RobWatson Mu-43 Hall of Famer

    Pixel peepers alert

    Maybe the blog section is a better place for this as it is mostly me rambling to myself to attempt to clear my head.

    I always equate dynamic range with signal to noise ratio but I suspect many do not.

    Signal range is a term I use to describe to darkest signal up to the brightest signal and some use this as dynamic range.

    Noise has many components and sources. What is the noise in a single pixel? Since many use ROI and standard deviation to evaluate noise a single pixel is lost in this type of definition - a uniform subject is required but still not quite good enough as pixel level (even sub-pixel) variations in quantum efficiency are not trivial. Best to use sets of images and compute statistics through the image stack (now a single pixel has meaning) - laborious this method is.

    Photon transfer curve is a great tool but requires many images (10 data points is good) so for the 19 ISO settings in an E-P1 that is 19 tranfer curves and 380 images to crunch (not to mention collect carefully). The same data set gives linearity (actually the full response curve) ... forgot to add raw+jpeg so 760 images to crunch. How badly do I really need to see these results?

    I read somewhere that the Canon 10D did something funny when switching from ISO 800 to ISO 1600. Basically the gain was altered in the signal processing chain for ISO up to 800 but for ISO 1600 and up they just stretched the digital data (gaps can be seen in the histogram?).
     
  3. RobWatson

    RobWatson Mu-43 Hall of Famer

    Random thought pops

    Dynamic range = signal to noise ratio under linear tranfer curves!?

    A significant distinction occurs when the transfer curve is non-linear (like with film) so dynamic range no longer simply maps to signal to noise ratio?

    Maybe this is why there is persistent confusion when comparing film to digital dynamic range? The non-linearity of film enables much wider signal ranges than is possible using a linear sensor and then doing a non-linear transfer curve (unless the digital sensor has the same or more signal range than film).

    Add to the mix the significant role played by contrast and understanding that there is a sort of conservation law in which (loosely speaking) the product of contrast and dynamic range is constant. One can actually trade contrast for improved dynamic range or vice-versa but one cannot boost both. Ponder a moment just exactly what anti-aliasing does.
     
    • Like Like x 1