Originally posted by Adam When there are lots of highlights the k3 does its best to avoid clipping them. That's not a bad strategy imo, as shadows are a lot easier to recover.
That's an important comment.
The k-3's high-res AE array will detect specular reflections more successfully than the 77-segment array in the K-5. I've had issues, when photographing high jump, where the strong reflection of sunlight off the shiny, narrow horizontal jump bar is blown irrecoverably (I always shoot raw), even with -1 EV Comp. With over 1,100x more segments (33x more in the vert. & horiz. directions), the chances of detecting a small, highly reflective area are much better.
So, the AE metering has better resolution (important also for the WB and AF tracking). So any possible tendency in the K-3 to "under-expose" may just be the better detection of extreme highlights.
As regards AE, what does the camera do next? The designers will probably have a percentage limit of pixels that can be acceptably clipped in a shot (e.g. 0.2%). A very small clipped region is unlikely to be noticed, but if it's on a person's cheek, forehead or nose, even this may be noticeable. The 86K AE sensor can better detect these hotspots, but what can the camera do about it? As far as I'm aware there are only 4 means available:
- Reduce the overall brightness level by lowering either the exposure (shutter-speed or aperture) or gain (ISO), depending on which AE mode you're using. This could end up making the rest of the image too dim, just for the sake of a tiny, visually-insignificant region. I believe part of the propriety/secret voodoo that each manufacturer designs in their matrix/evaluative metering system is scene recognition or how much weighting to apply to different AE metering segments in the calculation of the overall scene luminance level. I presume segments on the periphery, in general, and in the upper half (sky) in a landscape shot, are allowed to clip more, then for example, the centre.
- The user selects different Tone Response Curves e.g. Natural, Bright, Vibrant. The scene brightness distribution is no longer being linearly converted. With a "knee" & "toe" included in the TRC, and perhaps the gamma (slope of the TRC) too, the original DR in the scene is reduced, so that the overall image DR is more pleasing when reproduced within the limited DR space of a computer monitor or printer.
- Highlight/Shadow Correction. Over/under-expose/boost a region of the TRC and apply some compensating correction afterwards.
- HDR. Combine different brightness regions from multiple images taken with different exposures, and then map the tome of the image with a special TRC, either in-camera or in PP. The range encoded without clipping can be very large, but problems often occur in the tone mapping process in successfully portraying this DR, within the limited display DR, without looking unnatural.
What I don't think any camera currently offers in the TRC, but which might work, is the photographic equivalent of a single-ended audio dynamic compressor/limiter. This would dynamically vary both the location of the knee and its bend, in response to the percentage of extreme highlights metered in the scene, so as to reduce the final amount of clipped pixels.
If shooting raw, you can try to get even the brightest parts unclipped, and then fiddle with the shape of a suitable TRC in PP. The image out of the camera from a high-contrast scene will look predominately dim, but Sony Exmor sensors (as used in recent Pentax DSLRs) have very low-noise ADCs, and thus good Total Read Noise performance, which means that the shadows can be boosted a lot before noise becomes a problem.
One thing a lot of people don't realise is that, in the vast majority of DSLRs, the sensor has a fixed sensitivity (the exception being the Aptiva switchable sensitivity sensor used in the Nikon V1/J1). The sensels in the sensor should only saturate (reach Full-Well Capacity) at the true base ISO. The reason you apply ISO gain is not to increase the sensitivity of the the sensor to light. It's to boost the weaker signal from a sensor which, due to the combination of the scene luminance level and your choice of shutter speed and f-stop, is not "well filled". This boosted signal is then presented to the ADC and results in larger DNs (Digital Numbers) in the resultant digital image data. And higher DNs are rendered more brightly. So the image captured by the sensor in this situation is under-exposed, but is rendered with a good brightness level by the application of either analogue or digital gain (ISO increasing).
So, if you accept that the exposure process is the capturing of the photons (i.e. all about the light), rather than the rendering of the converted signal, then only a shot taken at base ISO has the possibility of being well-exposed. Once the combination of scene luminance, shutter speed and f-stop is no longer sufficient for the vast majority or for any of the sensels to reach FWC, then we're dealing with an under-exposed image. The signal boosting can occur either in-camera (typically analogue at low to mid ISOs; digital at high ISOs) or in PP (digital).
As the amount of under-exposure increases, the sensor output signal is weaker and weaker, and the max. levels in sensels in the brightest bits of the scene are failing further below their FWC. This is necessitating the application of more analogue gain between the sensor and the ADC. So when highlight clipping occurs in a properly designed digital imaging system, when operating above base ISO, it's not the sensor saturating that's causing it. Instead, it's the DN maxing out. (Of course, a specular highlight/reflection may be sufficiently big enough so that some sensels in the sensor reach FWC, even when the rest of the sensels are at quite low levels, and thus you'd normally use 1-2 stops of ISO boost to increase the overall rendered image brightness. But the likelihood of any sensels having reached FWC decreases as you apply higher ISOs - that's the reason why you're using higher & higher ISOs.)
The under-filling of the sensels at higher ISOs is shown in the table in this message:
https://www.pentaxforums.com/forums/115-pentax-k-5/162940-k5-vs-k7-dynamic-ra...ml#post1689551
Another reason why analogue gain is used, is to boost the weakest parts of an under-exposed image above the noise floor of an ADC. However since Exmor sensors have very low-noise ADCs, the need to due this is reduced. This introduces the possibility of shooting high-contrast scenes in raw at either base ISO or lower ISOs than you normally would, and then boosting the image brightness afterwards in PP. While the review image will be quite dim in the back LCD screen, the advantage is that the extreme highlights are much less likely to have been clipped,
when taking the shot. Afterwards, If you were just to boost them directly up to same rendering level as normal ISO boosting would, you would still face the same likelihood of DN max clipping occurring, as before. But since you are free to fiddle with the TRC during PP, you can adjust it to accommodate the rendition of the highlights
which have not yet been clipped. Looked at this way, an "ISOless" (exposure-centric rather than ISO-centric) approach to shooting with a suitable camera offers a superior form of highlight recovery.
Dan.