Using ACR 5.6 to interpolate/upres crazy high

ziggy53

Per an older thread at the Adobe support forums:

http://forums.adobe.com/thread/311713?tstart=0

I determined that you can use ACR ver 5.6, part of my PS CS4 installation and upgrade, and upres to astonishingly large image sizes in a single operation (so that you don't demosaic and generate an RGB image in ACR and then later use PhotoShop to upres as a separate operation.)

Note that I am not saying this is useful or practical, only interesting. This may actually bring down your computer and possibly even cause a loss of data so proceed at your own risk.

The procedure:

Load a RAW image into ACR. (It makes no sense to load a JPG or TIFF even though ACR can be enabled to do so.)

In ACR choose the Crop tool

Select the entire image with the Crop tool (or as much as you can).

Right-mouse-click on the crop selection (Windows system).

Choose "Inches" as the Crop measure and then type in the inches, 8" and 12" for instance.

The inches you choose have to be an equivalent aspect ratio to the image aspect ratio in order to choose the entire image.

In the "Workflow Options", located under the image (clickable link), type in a large enough resolution number to generate the required image size.

(This can be up to 999 pixels/in)

When you open the image in PhotoShop it will be the size that you designated.

---

I just tested this on my system and I was able to (slowly) generate an image of 24" x 16" at 999 pixels/in and those settings resulted in an image size of 23976 x 15984 pixels.

No, the image quality was not perfect (awfully soft and visible aliasing of straight but angled lines especially obvious with areas of strong contrast.) This was from a Canon 5D MKII RAW file, but I do believe almost any compatible RAW file will do for input.

Since there is no control over the upres algorithm I am not saying this is better than using PS to upres, this is just an interesting capability, apparently only available since ACR 5.2.

Ric Grupe

Interesting.

Why wouldn't it make sense to load an edited TIFF and follow your procedure? It seems to me that I wouldn't want to up-rez until I was sending it to print. Why have all those pixels to edit on the up-rezzed image?

pathfinder

Ziggy, I am not sure I understand the logic for this entirely.

Modest uprezzing of the RAW to jpg in ARC has been possible for along time, and I use it from time to time.

But for really large prints I tend to go to Genuine Fractals, because the pixels I want to "enlarge", complete with sharpening halos, if needed, are the pixels in my image AFTER my editing in ARC, AND in Photoshop. They may have passed through other plug ins and various manipulations as well. My large prints are rarely only globally edited prints. That might not be as true for a studio portrait study.

Why would I be better off to enlarge the basic unedited RAW file in ACR? I appreciate having more pixels, but my image editing in CS4 is done in ProPhoto 16 bit files, so they are pretty good files also, aren't they?

BinaryFx

Somebody with intimate knowledge of raw conversion/rendering and the proprietary "behind the scene processing steps and order" would probably be best to reply...

Is the ACR/ALR resizing being done on "mosaic data" or "demosaic data"? And what are the pros/cons of interpolation of each? My guess is the resize is done after demosaicking (however I could also be wrong!

Is the resizing being done on linear or gamma corrected data? Again there are pros and cons for each.

Also worthy of keeping in mind, ACR/ALR do not use the same method as Photoshop for resizing, however with each major release Adobe may tweak or change something under the hood. Last that I heard ACR/ALR use Lanczos rather than Bicubic (variants).

Have a look at this discussion, an Adobe engineer takes part in the thread:

http://www.luminous-landscape.com/forum/lofiversion/index.php/t28857.html

Anyway, the technical details are all very interesting, however at some point it comes down to comparing the same image** resized to the same size using different methods - then printing the results and making evaluations (rather than just looking on the monitor).

**If starting with raw, one may try the raw file vs. a rendered TIFF file - does ACR resize the raw and the TIFF the same way - and how does this compare to say Photoshop or other 3rd party methods?


Regards,

Stephen Marsh

http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

BinaryFx

pathfinder wrote:

But for really large prints I tend to go to Genuine Fractals, because the pixels I want to "enlarge", complete with sharpening halos, if needed, are the pixels in my image AFTER my editing in ARC, AND in Photoshop.

Pathfinder, I personally would prefer to resize a larger image with *no* sharpening halos in the original smaller size and then apply the required size sharpening halos to the final print size image.


Stephen Marsh

http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

ziggy53

The "potential" logic behind doing a single upres, and behind doing it from the original RAW file in ACR, is that the process of demosaicing is a type of interpolation itself. The pixels in an original RAW (from a Bayer imager) lack complete color and complete luminance information. Those pixel qualities have to be derived from the surrounding pixels using a "Bayer interpolation" algorithm.

If the RAW data is first interpolated for size and then for color and luminance it may be possible to derive a more accurate Bayer interpretation of both luminance and color information. While this would not produce more detail, it would potentially produce finer edges and smoother lines.

I don't know for sure what internal method ACR is using for the combination of scaling and Bayer interpolation, so I'm not willing to say this is in fact an improvement. All I can say for sure is that this process is not well documented.

If you read the thread I posted the previous link to, you will see that it was Jeff Schewe who said, "... until ACR 5.2, Camera Raw was not an optimal place to do resampling. Now it is ...", which leads me to believe that the scaling of images in ACR starting with ver 5.2 is somehow different.

I'll be testing the capability myself to see if there is a noticeable difference, but for now I just felt it important to let our members know of the capability.

pathfinder

ziggy53 wrote:

The "potential" logic behind doing a single upres, and behind doing it from the original RAW file in ACR, is that the process of demosaicing is a type of interpolation itself. The pixels in an original RAW (from a Bayer imager) lack complete color and complete luminance information. Those pixel qualities have to be derived from the surrounding pixels using a "Bayer interpolation" algorithm.

If the RAW data is first interpolated for size and then for color and luminance it may be possible to derive a more accurate Bayer interpretation of both luminance and color information. While this would not produce more detail, it would potentially produce finer edges and smoother lines.

I don't know for sure what internal method ACR is using for the combination of scaling and Bayer interpolation, so I'm not willing to say this is in fact an improvement. All I can say for sure is that this process is not well documented.

If you read the thread I posted the previous link to, you will see that it was Jeff Schewe who said, "... until ACR 5.2, Camera Raw was not an optimal place to do resampling. Now it is ...", which leads me to believe that the scaling of images in ACR starting with ver 5.2 is somehow different.

I'll be testing the capability myself to see if there is a noticeable difference, but for now I just felt it important to let our members know of the capability.

I see that comment now, Ziggy. Yes, it does sound like he is alluding to a newer, better algorithm, doesn't it?

pathfinder

BinaryFx wrote:

Pathfinder, I personally would prefer to resize a larger image with *no* sharpening halos in the original smaller size and then apply the required size sharpening halos to the final print size image.


Stephen Marsh

http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

I don't think we really disagree. I prefer no visible halos whatsoever also.

I use uprezzing in ACR for modest increases in image size. If I really want substantial image enlargement - say 5-10x. then I go to Genuine Fractals which includes a sharpening routine for large enlargements. I would not omit the sharpening I do in RAW conversion just because I want to enlarge the image later in GF. My goal with sharpening in ACR is to have no halos visible, even though I did mention keeping them in my earlier post. One of the things I really like in ACR is the masking utility which allows great control of halos in capture sharpening.

More a matter of phrasing than actual disagreement I think.

BinaryFx

pathfinder wrote:

I don't think we really disagree. I prefer no visible halos whatsoever also.

More a matter of phrasing than actual disagreement I think.

We are both reasonable people, so no actual disagreement!


Cheers,

Stephen Marsh

http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

arodney

IF you know from the get go, you need to rez up from raw data, do it in ACR/LR instead of Photoshop. And yes, there's no reason to do this with existing rendered images. The source? Engineers at Adobe, testing this with Jeff Schewe etc.

Interpolation algorithms in the ACR engine are more refined than BiCubic (smoother, sharper). It uses a adaptive bicubic algorithm. Source, Eric Chan of Adobe.

pathfinder

So Andrew, you would uprezz 10X or 15x in ACR rather than later with an edited image?

Some plug ins used in editing in PS, will crash if presented with very large images, I suspect - I have experienced this from time to time.

I gather you are saying there is no need for Genuine Fractals or Blow Up or the other uprezzing engines on the market today then?

ziggy53

pathfinder wrote:

So Andrew, you would uprezz 10X or 15x in ACR rather than later with an edited image?

Some plug ins used in editing in PS, will crash if presented with very large images, I suspect - I have experienced this from time to time.

I gather you are saying there is no need for Genuine Fractals or Blow Up or the other uprezzing engines on the market today then?

With larger images i often have to clear some of the history in order to proceed and yes, I believe some filters do not return memory properly upon exit. Those filters will cause problems faster with bigger file sizes.

I would regard the ability to upres in ACR as a useful option. Obviously, if you have a proven workflow and you are happy with the results then that workflow and results will continue to satisfy.

I do think that using ACR to help crop images and retain as much quality as possible is a reason enough to develop a new strategy for those occasions when you need to crop and retain the best possible results.

I don't expect a level of magnitude improvement, but every little bit helps. I'll be testing some images in the next couple of weeks and I'll post the results either way, the good or the bad. I encourage everyone who wants and who has time to do the same.

I can dream, can't I?