Difference between AP raw conversion and Mac OS system raw conversion

[Peter W Gallagher]

I've owned AP since it was first published. I buy stuff by independent developers these days in the hope that it will become great. But I started using AP only a few days ago when I discovered the direction of the workflow (from 'developer' to 'photo') and watched a few video tutorials (well-made & relatively information-dense, but... video! ugh!).

 

I'm surprised to say that, now I "get it", I'm impressed by AP. This will be a killer suite when it's a bit more mature. As a user of specialist focus-stacking software, I was delighted by the built-in AP routines, for example.

 

I'm curious, however, about the choice between AP raw conversion and Apple raw conversion. Is there a difference?

 

I have a prejudice that raw-conversion is in fact critical although it's one of those things that gets swept under the carpet by most "photo" suites. I use Adobe Camera Raw by default but I can't decide whether its de-mosaicing etc is much different from the de-mosaicing in e.g. Photos. What is different is the non-linear adjustments to exposure and contrast/tone-curves (especially) that ACR applies by default (in Process 2012 and earlier) without telling the user (Adobe is a little more transparent about input-sharpening).

 

I'm sure Photos also applies it's own (hidden) post de-mosaic adjustments. They're not as clever as Adobe's in my view.

 

What does AP do? Does it apply some initial adjustments OTHER than those mentioned in the "Develop Assistant"? Are AP de-mosaicing routines developed in-house? Do they have some advantages we should know about?

 

PS: The only raw converter I own that does NOT make initial adjustments unless you tell it to is Raw Photo Processor (http://www.raw-photo-processor.com); a great program with many clever touches, if a little eccentric.

[James Ritson]

Hi Peter, you're correct that raw handling is important, and raw converters will handle demoisaicing differently with varying results.

 

There are several more steps involved - which are again handled differently between raw converters. Demosaiced raw data starts as "scene referred", which is a measurement of light in the scene. You will very rarely (if at all) see a raw file in its linear "scene referred" form. It then goes through gamma curve correction, gets mapped to a colour space and has a tone curve applied to produce a result more in line with the user's expectations (similar to an in-camera JPEG).

 

 

So to answer your question - yes, there's a difference between SerifLabs and Core Image RAW.

 

SerifLabs will demosaic, gamma correct and tone map, but as you've found, the additional tone curve is optional. If you turn this off, you'll only see the image with a gamma curve correction. No additional sharpening is added by default - this is left entirely up to the user. Colour noise reduction is added by default; previously it wasn't, and we faced a lot of criticism over raw development quality because users are so used to having raw processing software apply it automatically. The harsh reality is that yes, your camera really is that noisy ;). You can turn this off if you wish on the Details panel, I just wouldn't recommend it.

 

I've done some analysis, trying to make a SerifLabs-decoded image match a Core Image RAW-decoded image, and I've come to the conclusion that Core Image takes some additional steps. It adds sharpening whether you like it or not, there's no doubt about that. It certainly performs colour noise reduction, and I also believe it does some luminance denoising and then dithers slightly by adding in some fine noise to retain texture. This approach wouldn't be entirely out of step; for quite a while, Apple's H264 decoder added some fine noise to reduce blocking and banding (this is back in 2007/2008 when its hardware-based H264 support was less comprehensive). I'm unsure of Apple's modern approach to H264 but I expect it's more refined now.

 

 

At this stage of Photo's development, the raw handling could still use some improvements, and over time it will be improved: namely a better approach to demosaicing and some more effective noise reduction. Demosaic implementations are continually being researched and written about, and there is always scope to do better here.

 

 

As far as advantages of SerifLabs go, there is one that I can strongly point out: because betas are made available in-between major releases (either for bug fixes or to introduce new features), new raw camera support is added frequently - so if you invest in a new camera, chances are you could grab a beta and be able to open raw images sooner rather than having to wait for an official update. For example, 1.5 was released in December and supported the new Olympus E-M1 mk2 camera which also shipped that month. It also opened images shot with the camera's high res mode (sensor shifting to produce 80MP raw files) - a feature that's still yet to be supported in some raw converters.

 

At the end of the day, the best advice is to experiment and find which raw converter's results you like the most based on what you shoot; e.g. if you do a lot of high ISO urban photography you'll want some fairly robust noise reduction and, perhaps more importantly, good colour handling. If you're into landscape photography with difficult dynamic ranges perhaps you'd find the ability to remove the tone curve more useful. And so on and so forth...

 

Hope that helps!

[Peter W Gallagher]

James, thank you so much for a full and informative reply to a new user. It helps a lot.  Once more, I am impressed by Serif. 

 

I'm very glad to learn of Serif's choice to manage your own RAW processing; for the reasons you give but also because it shows your interest in image quality. 

 

I had seen the color noise reduction option (and the luminance noise reduction option) in the Develop Assistant. I have really no qualms about the former since I imagine it is more or less impossible to un-twist the Bayer sensor puzzle without some color noise. As for luminance noise: I prefer to make my own luminance corrections, so I'm grateful for the option to separate it from the color-noise treatment (I use a Nikon D810 normally on the lowest ISO I can manage, so I usually choose to accept luminance noise, if any, to keep overall IQ high).

 

Your observations on Apple Core-Image RAW routines as implemented in Photos seem exactly right to me. Adobe ACR does not seem to add anything to retain texture but a little digging with RawDigger shows that their tonal curve manipulation and added exposure can be quite large. The result is still, usually, a good place to start user-development. But I've seen a few cases in my own images where the (roughly 1.3ev) added exposure and (increased) contrast has been misleading (they also move typically push the black-point a bit, but I don't argue with that).

 

If you don't mind I have a follow-up question about input sharpening in AP. But I'll make a separate topic in this forum about that.

 

Peter