Why Don’t Digital Cameras Offer a Perfect ETTR exposure?
In DAP, I discuss the benefits of Expose to the Right (ETTR) for lower noise and better image quality. In essence, one exposes as bright as the sensor will take without blowing highlights, then “pulls” back the exposure. Call this the exposure delta.
Why should anyone have to figure this out manually?
As an optional setting, why cannot the camera take the picture, analyze it (don’t save it or do, optionally), then immediately take another one that maximizes the exposure to the limits of the sensor? In bright light or on a tripod, this is exactly what I’d want in many situations. It’s not appropriate for sports or action or other stuff, but so what? Turn it off in such cases, just as one disables mirror lockup or whatever other options.
The camera would record the exposure delta for proper post-shot rendering as per the exposure that was designated (e.g. the exposure is set for 1/250 at ƒ/4, the camera actually shoots 1/500 at ƒ/4, there is a -1 stop exposure delta).
In this way, the photographer could aim for the tonal balance that was desired (e.g. high key, low key or whatever), and the camera would simply do its job— make the best possible capture.
It’s nuts that such useful image quality improvements are not implemented with digital. Instead we get “scene modes” and other gimmicks that are so simple and so numerous that they become too complicated for me to understand, and have the desultory effect of diminishing the usability of the camera.
I don’t want bracketing, I want the best possible digital capture. It seems so obvious. Who will be first to do it out there?
- Bracketing is a crude film-era hedge whose time has passed. The very concept of bracketing is obsolete; the camera can just get it right, with one (1) frame. Every sensor has its maximum potential, the job of the camera is to fully utilize that potential, and do so automatically. This is not a creative option; it should just be built-in, ignored, and assumed. Sensor design could incorporate it directly (e.g. just cut off exposure just prior to overflow). Of course, the photographer needs to be able to override this behavior, as it might be demanded to use a certain shutter speed or aperture for the exposure which must not be changed. But on a tripod in aperture priority, there is frequently no reason to care whether the exposure is 1/30 or 1/18th second.
- A setting that allowed choice of the amount of specular clipping, eg. 0.1%, 0.2% could be part of the setting.
- A true RAW histogram for each R/G/B channel, something like the Leica M Monochrom, which actually shows the gamma-less distribution of the RAW data; in other words show what the camera is actually capturing on the sensor. Makes a ton of sense for a true-color Foveon sensor, but probably still makes sense for a Bayer sensor.
- The camera ought to provide a sufficiently wide gamut to encompass the actual sensor gamut (e.g., ProPhotoRGB). So that the histogram could actually be used accurately, even if this proposed feature is not implemented. Even so, inappropriate white balance messes with that analysis by the user, so the true RAW histogram is helpful.
- Maximizing signal to noise is especially important with small sensor cameras, which don’t have the signal-to-noise capabilities of full-frame DSLRs. Case in point: the Olympus E-M5 which can be quite noisy even at its base ISO of 200. Yet (for RAW), it often can take a whopping one to 1.5 stops more exposure than the autoexposure suggests.
Elaborating a bit:
- The only way one can know if the camera is using the full range of the sensor to maximize signal-to-noise is to analyze the entire frame (by taking a picture and/or analyzing a full Live View frame). But no one can analyze the frame, because the camera provides no mechanism to do so (histogram does not cut it, except on Leica MM).
- Only the camera can analyze a sensor of data (ideally, right off Live View if it is in use) and determine what is the maximum exposure, because only it can see the raw data numbers off the sensor free of misleading interpretations.
- Only the camera can choose oddball fractional shutter speeds, allowing (for example) exposures of 1/125, 1/115, 1/113, 1/103 second, to produce a perfect maximal exposure.
- The camera histogram is misleading, often giving up a full stop or more of headroom, depends on the white balance (often inappropriate) and color space and gamma thereof (pinned and clipped to AdobeRGB in all cameras today) and other camera settings. Camera vendors usually don’t even document what settings the histogram incorporates (JPEG or RAW based? White balance? Color space and gamma?), making it even more mysterious.
- The blinking highlight features in some cameras are also inaccurate, with the same errors and assumptions (color space, white balance, etc) as the histogram.
- Most cameras consistently give up a stop or more of headroom, even for shots that look perfectly exposed. A doubling/halving of S/N (e.g. 1/250 instead of 1/125) affects noise by √2. Almost always, images leave nearly a stop of headroom unused, based on what I see during RAW conversion.
- Most recently, see how the Sigma DP1 Merrill behaves for metering/histogram for an example of the misleading histogram. It is no different for Nikon or Canon; I have seen this problem for years, working around it manually via ETTR techniques.
- The entire frame needs analysis. This can only come off the sensor. In my experience, metering is far off the mark in the vast majority of scenarios in which I shoot, which is why my cameras are on manual 99% of the time.
- Spot metering won't get the job done, and requires recomposing and still requires the user to guess at how much the sensor can take, and the camera’s own metering does not inform one of anything but middle gray exposure— so it’s a guess how far to go to get to max.
Bottom line is I want a smart camera. I do not want more complexity (e.g. a waveform monitor), I want the camera to just give me its best result when I press the shutter button so that I can focus on making the image I am after. If that is not immediately feasible in some automatic way, then at least I’d like a wide-gamut histogram that allows me to accurately assess the usage of the sensor potential.
Here are a few pages which address this topic in general and specifically.
Exposing for Highest Dynamic Range and Lowest Noise (Sony RX100 in Guide to Mirrorless).
Exposure Bracketing Sequence (Sigma DP1 Merrill in Guide to Mirrorless)
Dynamic Range Analysis from -2 to +4/3 Stops (Olympus E-M5 in Guide to Mirrorless)
Pattern Noise — Expose to the Right Analysis (Grill) (Canon 1D X in DAP)
RAW vs JPEG — Dynamic Range (Fuji X100 in Guide to Mirrorless)