Rambling on trends towards AAless and pixel-peeping.

Started Jul 29, 2014 | Discussions thread
Flat view
Timbukto Veteran Member • Posts: 4,988
Rambling on trends towards AAless and pixel-peeping.

It has gotten to the point that people keep parroting that AAless is an advantage. And not just people as in consumers, but people as in actual editorial reviewers or youtube reviewers. Not even Foveon AAless or X-trans AAless mind you being the advantage...but that plain old vanilla Bayer AAless is being parroted as an advantage, often listed in the 'Pros' section of many a camera review. A true pity that some Pentax/Ricoh engineer may have spent sleepless nights devising a novel microvibration based AA sensor only to realize that pixel peepers just want their pixels subjectively sharp at 100% no ifs, buts, about it. And the market exec can point and laugh, and really get rid of such engineers, because truly in today's market they can sell something for more $$ with less cost.

There is another area where I remember the removal of AA filters to be marketed as a 'good' thing, and that was 15+ years ago with the popularity of Redbook PCM audio and audiophoolism, many audiophools believed in non-oversampling non-AA sound even from 16-bit Redbook...there even existed some unfortunate ADC's that did this (i.e. these digital recordings will forever be marred by alias artifacts that creep into lower frequencies that can be heard).

You can see an example of such audiophool sentiments being bullied by sound engineers here - http://repforums.prosoundweb.com/index.php?topic=15092.0

But there is no such bullying in the photography playground, its the AAful sensors that get bullied...we just go with it even though the science and evidence strangely enough is more easily seen than it is heard.

However there is a fundamental difference between high frequency audio sampling and the application of AA filters there, and high frequency spatial sampling and the application of AA filters in that domain, and it is the following:

Humans outright accept that anything above 22khz is mostly meaningless information in the domain of music reproduction...sounds that clearly do happen in ultrasonic frequencies are not considered meaningful to us in any way, even if its quite possible they occur all the time in nature. They obviously have no need to harmonically occur from the human voice or human instruments, since again, there is no need to design instruments that reproduce frequencies our ears cannot hear. As such we can have equipment that handles high frequencies via oversampling...slap on an AA filter where necessary far enough outside the range of hearing to effectively create perfect brickwall filters (the problem with AA filters without oversampling is that they blur frequency information under the established nyquist frequency). With oversampling you can effectively apply gentler AA where it will blur the higher frequencies but they are well beyond nyquest frequency. In other words there may be bats or vampires that complain about AA filters, and there may be deaf middle-aged men that will complain about imaginary problems that do not cost them enough disposable income...but for the most part we can consider the application of AA filters in the audio domain to be rather *sound* science.

Imaging domain has similarities but also fundamental differences. In the imaging domain I can argue that the act of capturing higher spatial information (higher MP or dpi if you will), and typically downsampling to web/print sizes of 2-3 MP or 8 MP print is also very similar to oversampling in the audio realm. The difference however is...at 100% crop and analysis of high frequency information...the spatial data is still 100% meaningful to the human interpretation. High frequency spatial data is no different from just looking at something up close...it is not an alien concept to see something closer...it is an alien concept to start hearing ultrasonic frequencies. If however the desired output of your photography is to share web sized images or prints...you can truly adopt the same sound methodologies of oversampling and AA used in the audio world or all other industries it is effectively used (such as 3D gaming, etc).

*IF* however you are focused on the continual absorption of more and more data and the ideal 100% crop, you effectively are following the endless cycle which is conductive to GAS and pixel peeping. I would call it visual masturbation since really pixel peeping is currently really a non-shared activity in this day and age...only certain photographers delight in doing this to their *own* images but it is rare if not difficult to get others to pixel peep your own images unless to discuss technical merits. In fact I have images that to me I appreciate how they look...even at the 100% crop level...and even if I post the full-size image for C&C, I guarantee that 95% of the viewers will evaluate the image at a heavily downsized level and not pan and pixel peep the image.

Another trend this pixel peeping leads us down is that it is absolutely *clear* that modern OVFs and even EVFs quite simply do not give us a reasonable preview of what our 100% crop images look like...not even remotely close! Every tack sharp picture I take even wide open on film era primes on my 6D reveals more details in review than I ever anticipate during the capture at OVF time. In a way photography becomes more like a boy opening a box of cracker jacks for the mystery toy, when it should really be more about the composition and framing you achieved in the OVF during capture! Granted we all revel in these surprising details, but there comes a point where focusing on one might be at the detriment of the other (i.e. if you have a super high MP camera that allows you to crop to utterly ridiculous levels, it is not feasible that an OVF can preview this for you without an OVF that literally is designed with zooming optics...its possible that an EVF can however in the future via magnification off the sensor). It also presents a bigger burden to the hybrid or CDAF type AF systems which are more reliant on examining pixel data (i.e. if they are forced to subsample for speed they obviously will not be as accurate).

Granted I would rather we oversampled *more* than Canon typically does in their 20-22MP cameras...if anything the 24MP AA-ful APS-C sensors are the closest imaging sensors that mirror the ideals of oversampling and AA in audio engineering especially as its not as easy to find APS-C level lenses that can resolve high frequency spatial data, but apparently its not difficult to blow past the pixel pitch of the 36MP D810 even with zooms like the 24-70.

In any case I vastly prefer my audio AA filtered + oversampled, prefer my 3d games with the best AA filters applied, and currently still prefer AAful imaging sensors *especially* if its traditional Bayer. There are more than a few sample D810 images that I can greatly downsample to 12x8 print and see obvious moire and aliasing issues where the AAful counterpart would have downsampled just fine to produce the exact same sharpness without the moire (examination of either DPReview or ImagingResource studio shots reveal the very same thing). I still pixel peep my images, but within the context of knowing that at 100% crop, the slightly blurred image 100% crop may actually hold advantages after downsampling and decimation. It is funny how most photographers say not to pixel peep...but nearly *every* photographer judges technical quality of lenses and photos based on sharpness of 100% crops. This is really due to the ease of imaging software presenting information we can interpret, but in some ways the analogy of reviewing the 100% crop is like reviewing the ultrasonic frequencies of your music recording.

The only way to really adopt the conventions set in audio engineering however is for the professional imaging world to do away with pixel peeping, and stop moving the nyquist frequency up and up as MPs get higher and higher, but to establish a set frequency that is most effective based on the spatial information that can be provided by high quality optics on a given sensor size with modern stabilization + handheld, and anything higher is just 'oversampling'. Another nyquist frequency should be established based on perhaps tripod or even flash studio shots which should be substantially higher. Only then can you also establish the perfect brickwall filter for an established nyquist frequency. In other words there are really some practical engineering decisions that can be made based on how you plan to use your camera. But the photography world is going boutique just like high-end audio gear, so we know what that eventually means...pixelphiles, like audiophools, just love to pay more money, for less results.


 Timbukto's gear list:Timbukto's gear list
Canon EOS M Canon EF 200mm f/2.8L II USM Canon EF-M 22mm f/2 STM
Flat view
Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow