Upsampled Conversion - An Effective, Practical Technique

[Jerry Skrocki]

Eric Chan improved the up-sampling capability in Camera Raw 5.2 for Photoshop CS4. Output sharpening was also added to this version but is not recommended for an image that requires further processing.

This is what Jeff Schewe has to say about it:

"If you are going to be doing substantial imaging work in Photoshop including resizing and other things that will impact size and compositing, you really would _NOT_ want to use output sharpening in the workflow settings...the output sharpening is designed to be done ONLY when the final image size/resolution can be determined and set in Camera Raw. Resizing or resampling of an output sharpened image is not encouraged nor part of the post-processing in Photoshop.

However, if you DO accomplish everything in Camera Raw and simply want to print it at the PPI and size that comes out of Camera Raw, this is a workflow enhancement. I would hope that in the future, Camera Raw's output size/resolution usability is substantially improved. In ACR 5.2, the resampling algorithms have been improved to be equal to or better than Photoshop's own resampling option. By default, any downsample will use a flavor similar to bicubic sharper (don't know if it's exactly the same) but the upsample is an adaptive blend of regular bicubic (for small increases) and a bicubuc smoother flavor for larger resamples. Eric can tell you more about this improvement–it was one of his babies this time around (as well as fixing the output sharpening–thanks Eric).

But the fact is that the usability is still primitive. For example, you are limited to selecting one of only a few resample options and there's no readout for the resulting image dimension at a given PPI output resolution. Camera Raw knows all that stuff, but just doesn't have the UI to display it for useful purposes...also, the fact that Camera Raw may no be the optimal place to resample, you're still limited to a pitiful fews size options. That needs to be addressed...but not in the 5.x timeframe (I don't think)."

This is what Eric Chan (Camera Raw Engineer) has to say about the issue:

"I don't recall there being a thread specifically on the issue, but yes, image resampling was adjusted in CR 5.2 (and upcoming LR 2.2) both for downsampling and upsampling.

p.s. You will not get equivalent results to PS for several reasons, one being that PS largely operates in a gamma-encoded space (e.g., if you're using sRGB, Adobe RGB, or ProPhoto RGB, then you're either using gamma ~2.2, 2.2, or 1. as opposed to RIMM space in Camera Raw. "

Jerry

[MathewLodge]

Originally Posted by Doug_Kerr

Perhaps it is the still-mosaiced ("raw") data set that is upsampled. That would increase the amount of information in the "bit-count" sense, but not the amount in the true information theory sense. It is not as if, for example, we are actually sampling the image on the sensor at a finer pitch.

Well, for this to actually improve things in a way that there's more information in the up-rez'd image, the larger image has to be generated from the mosaiced data set. Otherwise it's equivalent to a raw conversion followed by an uprez and there's no net information gain. So I'm agreeing with you, Doug -- has to be that way.

That said, I still get surprised in an "information-theory" sense by image processing stuff, without causing Claude Shannon to spin in his grave. The last time was with Focus Magic, which uses de-convolution to extract (new) information about a pixel from its neighbors, thus improving image sharpness. The information was there all along, but a new technique puts it to work. Focus Magic is still my favorite for capture sharpening, and why I can never fully process an image in Adobe Lightroom.

[Noel_Carboni]

I have to admit I was a bit scared to update from Camera Raw 5.1 to 5.2, as I had a very high opinion of the detail that 5.1 was delivering at the upsampled resolutions, and I kind of felt any change could be for the worse, but I finally said what the heck and went for it.

If anything the upsampled conversions are even better in 5.2.

This is a crop from a well-exposed 40D image (native 2592 x 3888 = 10.1 megapixels) that's been converted to the upsampled 4096 x 6144 setting (25.2 megapixels!), then sharpened. I'd say the sharpness is simply excellent at this size!



I'm a happy camper with Camera Raw 5.2, and I simply can't see a need for a camera with more megapixels with this kind of performance from my current model.

-Noel

[Doug_Kerr]

Hi, Noel,

Originally Posted by Noel_Carboni

This is a crop from a well-exposed 40D image (native 2592 x 3888 = 10.1 megapixels) that's been converted to the upsampled 4096 x 6144 setting (25.2 megapixels!), then sharpened. I'd say the sharpness is simply excellent at this size!

Certainly a fine result.

Would it be possible for you to, say, do the conversion of the same negative file at "native" resolution (one pixel per sensel), save that to a TIFF file, then freshly open that file and upsize it (say, in PS proper) to the resolution used in the example, and let us see a comparable crop from it? It would be nice to in A-B comparison see how your scenario process improves on the more canonical one. I would expect you to sharpen in the process as you would see fit.

Thanks.

Best regards,

Doug

[Jerry Skrocki]

While the up-sampling is very good in ACR 5.2, it is limited to the few choices given in the interface. It also does not list the resulting file size prior to opening in Photoshop, nor does it give you control of the algorithm used in the conversion.

[Noel_Carboni]

Good idea, Doug, I'll try it.

Jerry, I am starting to think it's not an upsampling operation at all, per se, but rather a scan through the x and y values that will make up the pixels in the final image, and the algorithm choosing the input photosite readings from which to retrieve data.

-Noel

[Jerry Skrocki]

Originally Posted by Noel_Carboni

Jerry, I am starting to think it's not an upsampling operation at all, per se, but rather a scan through the x and y values that will make up the pixels in the final image, and the algorithm choosing the input photosite readings from which to retrieve data.
-Noel

If the resulting file size is greater than the native resolution of the RAW file, it is upsampled. The technique used in achieving the final resolution is what is in question.

Jeff Schewe (from Adobe User Forums) states:

"Camera Raw uses a Lancoz upsampling that is not the same as any of the Bicubic algorithms. In general one would want to use Bicubuc Smoother then apply more sharpening carfeully than using Camera Raw's upsampling or Bicubic Sharper. Bicubic Sharper is really a lazy way of upsampling while avoiding the whole sharpening issue. Bicubic Sharper was designed for downsampling because it needs a higher level of sharpening."

Most RAW files are 12 bit or larger, so it makes sense to use 16 bit in ACR. Instead of throwing away pixels with 8 bit color, we can get smoother color transitions using 16 bit and ProPhoto RGB.

According to one of the ACR developers, Eric Chan, both Camera Raw and Lightroom use RIMM colorspace which is most closely related to ProPhoto RGB.

"From a user's point of view, given the way CR/LR is designed, you may as well stick with ProPhoto RGB as a working space because CR/LR internally uses RIMM (which has PP RGB primaries)."

Jerry