Epson Native Resolution (360)

[Bart_van_der_Wolf]

Understood Bart, but you made a good point and a useful contribution, because on the basis of your intellectual curiosity I went back to "source" and got the following response:

<It is not Quartz “Composer” but Quartz “2D” - rasterizes drawing objects from the application and provides to printer driver. http://en.wikipedia.org/wiki/Quartz_2D>

Hope that helps clear-up the matter.

Hi Mark,

Yup, now it makes sense. Thanks for verifying with your 'source'.

Here is a more detailed description of the formal print process through the various components. It explains with a schematic in a bit more detail how the sending application (e.g. Photoshop or LR ) sends the print job to the printer, and that the printer driver is simplified because it can borrow some functionality from "Quartz 2D" and now only needs to generate the printer specific raw commands for the already 'cooked' raster data / pixels.

Thus the print job is an interaction between Application (supplies the image), Job manager (queries the destination printer's associated Printer Module to find out what data format it requires and, if necessary, the Job Manager uses a converter (Quartz 2D) to transform the incoming data into a format that the destination printer module can accept), "simplified" Printer driver/module (converts the incoming data into the raw commands for the printer), and OS (handles interfacing between modules and outside world).

So the modern manufacturer specific printer driver (now called printer module in OS X) is intertwined with some generally available OS functionality, and it utilizes Quartz 2D for pixel resampling as instructed by the Job manager which queries the printer module for the required PPI. Which makes sense, because why reinvent the resampling wheel that's already available.

In a Windows OS environment the printer driver functionality is currently less intertwined (the formal term is "interoperability" ), and presumably resampling to the printer requested resolution is done by the printer driver itself (although e.g. Direct-X (Direct2D) functionality could be utilized).

Unfortunately an inferior resampling filter seems to be used in the various drivers/modules (Nearest Neighbor, Bi-linear, or Bi-cubic), and without the possibility to influence output sharpening after resampling, so the general conclusion to do it ourselves (resample to exactly 300/360 PPI or 600/720 PPI by application or plugin, and sharpen) before sending the data to the printer remains.

Printing border-less will trigger another resampling step in addition to the already resampled/sharpened data, which will reduce print quality marginally (possibly less at 600/720 PPI because more precise raster data is available).

Cheers,
Bart

[Ernst Dinkla]

In a Windows OS environment the printer driver functionality is currently less intertwined, and presumably resampling to the printer requested resolution is done by the printer driver itself (although e.g. Direct-X functionality could be utilized)

Unfortunately an inferior resampling filter seems to be used in the various drivers/modules (Nearest Neighbor, Bi-linear, or Bi-cubic), and without the possibility to influence output sharpening after resampling, so the general conclusion to do it ourselves (resample to exactly 300/360 PPI or 600/720 PPI by application or plugin, and sharpen) before sending the data to the printer remains.

Printing border-less will trigger another resampling step in addition to the already resampled/sharpened data, which will reduce print quality marginally (possibly less at 600/720 PPI because more precise raster data is available).

Cheers,
Bart

Bart,

If I recall it correctly the Epson and more drivers improved after 2006, Windows versions. I would be surprised if any recent Epson driver still uses "Nearest Neighbour" for upsampling. For downsampling anti-aliasing is absent on the HP drivers I tested and my gut feeling is that a primitive resampling algorithm is used. It would not surprise me if these flaws are more common in today's drivers.  Qimage Ultimate does anti-aliasing in default on downsampling.

In case of printing from Qimage Ultimate the conclusion that we have to do the resampling/sharpening ourselves is a bit overstated, the application does a good job auto-magically, on the fly and without any changes to the original file. Beyond the defaults the print related choices are flexible enough to do any print job properly. I am not a big fan of Qimage's image editing but if it gets a proper Tiff to start from there is not much in manual steps needed for a good print.

I am not using borderless printing at all but Qimage can also interfere on the driver's borderless "expansion" on the input resolution with a preference setting that will result in a 720 PPI resolution + the exact sheet size to the driver. It will then be impossible to avoid some white at one or two edges where the normal borderless expansion gives some overspray, sheet/roll register is not that good on our printers.

--
Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
April 2014, 600+ inkjet media white spectral plots.

[Mark D Segal]

.........................
Printing border-less will trigger another resampling step in addition to the already resampled/sharpened data, which will reduce print quality marginally (possibly less at 600/720 PPI because more precise raster data is available).

Cheers,
Bart

Hi Bart,

On this one, my source advises as follows:

<use the “Retain Size” option in Epson LFP driver. It was requested/prepared just to avoid such an unexpected resampling.>

Cheers,

Mark

[Bart_van_der_Wolf]

If I recall it correctly the Epson and more drivers improved after 2006, Windows versions. I would be surprised if any recent Epson driver still uses "Nearest Neighbour" for upsampling.

Hi Ernst,

I have also not found any evidence for that yet (neither for Canon or HP drivers), but also not what they actually do use instead (and how that is affected by driver settings). It would indeed surprise me, given the capabilities of the various APIs (Quartz 2D, Direct2D, Open GL, etc.). I know that even for Display rendering (many frames/second) Nvidia CUDA allows a choice including Mitchell Netravali filtered scaling (which works very well for both up-/ and down-sampling).

For downsampling anti-aliasing is absent on the HP drivers I tested and my gut feeling is that a primitive resampling algorithm is used. It would not surprise me if these flaws are more common in today's drivers.  Qimage Ultimate does anti-aliasing in default on downsampling.

Yes, Mike added that functionality at my request, and Qimage allows the user to even adjust it in a few steps if the default setting doesn't satisfy.

In case of printing from Qimage Ultimate the conclusion that we have to do the resampling/sharpening ourselves is a bit overstated, the application does a good job auto-magically, on the fly and without any changes to the original file. Beyond the defaults the print related choices are flexible enough to do any print job properly. I am not a big fan of Qimage's image editing but if it gets a proper Tiff to start from there is not much in manual steps needed for a good print.

I agree, most defaults are very sensible and never need adjusting, but there is still the possibility to override in very specific scenarios. The on-the-fly conversions, even multiple sizes from the same source on the same page, of course with subsequent automatic output sharpening, make it a productivity enhancer with super quality output results.

I am not using borderless printing at all but Qimage can also interfere on the driver's borderless "expansion" on the input resolution with a preference setting that will result in a 720 PPI resolution + the exact sheet size to the driver. It will then be impossible to avoid some white at one or two edges where the normal borderless expansion gives some overspray, sheet/roll register is not that good on our printers.

Indeed, it again attempts to do better than the straight forward print instructions. It's explained in this tutorial video quite clearly. That's also where the feedback of the actual PPI from the printer driver in Qimage's interface is very useful. It has warned me at occasions about a printer driver setting that changed the PPI that I intended to sent. One of many useful features.

Cheers,
Bart

[Bart_van_der_Wolf]

On this one, my source advises as follows:

<use the “Retain Size” option in Epson LFP driver. It was requested/prepared just to avoid such an unexpected resampling.>

Hi Mark,

Thanks for that addition. Since I don't print on an Epson, I'm not sure if all (desktop and LF) Epson Drivers offer that feature. Anyway, Qimage did already allow to override the standard printer driver response for some years, so now everybody can be happy.

Cheers,
Bart

[Mark D Segal]

Not everybody can be happy on account of QImage - it's Windows only and "some of us" print using OSX. The LFP solution is available to those who use Epson printers. I have no idea what happens for Mac users of Canons or HPs.

[Bart_van_der_Wolf]

Not everybody can be happy on account of QImage - it's Windows only and "some of us" print using OSX.

I meant that even those unfortunate enough to not use Qimage (which BTW runs fine on Macs under Parallels or similar) can apparently use the switch in the driver your source mentioned.
 
Cheers,
Bart

[Paul2660]

And the question I have ask, is:  Do you feel that the output from Qimage at 360, is that much better?  I personally don't see it.  Qimage (which I  have used for over 6 years on and off, so I am pretty familiar with it), can do a great job, however it's still another step.  Ion't work on MAC's, every often, and print only from Windows.

The LR 5.4 print module for me, is just so much more simple in both the interface (Qimage is not, never has been).  I say this as I have spent about 1/3 of my life either selling, supporting or working PC's and most software I can grasp.  Qimage I can't.  It might just be me, however the interface is not at all straight forward.   

LR still (and most likely never will) have layers, but it has variants.  But for me, keeping the file at 300dpi, no matter if it's Fuji, Canon, Nikon or Phase One, and then letting LR make the final size for the print to 360.  I have also tried the separate uprez software solutions, mentioned earlier, and even on larger output, I personally don't see the difference.  And I am confident the average viewer won't either.

Not saying one method is better than the other in final output, but to me the workflow within LR is superior. 

Paul

[Bart_van_der_Wolf]

And the question I have ask, is:  Do you feel that the output from Qimage at 360, is that much better?  I personally don't see it.  Qimage (which I  have used for over 6 years on and off, so I am pretty familiar with it), can do a great job, however it's still another step.  Ion't work on MAC's, every often, and print only from Windows.

Hi Paul,

I guess it depends on image content, pre-processing, and actual print medium (glossy / matte / canvas XYZ) and brand. All I know is that Qimage will use 720 PPI if allowed by the 'finest detail' setting checkbox in the Epson printer driver, and the Qimage Ultimate version will use the new Deep Focus Sharpening (DFS) method for halo-free smart sharpening after resampling. This alone, combined with very good resampling algorithms, should be able to get the best quality out of a given file, perhaps only bested by using a specific resampling software and a feature Detail output enhancement plugin.

It would surprise me somewhat if there would be no improvement over 360 PPI prints, although it might turn out to be subtle for some subject matter and small input file sizes. I can certainly see the improvement on my 300 -> 600 PPI prints (architecture/nature/macro subjects) on Canon printers, and that makes it worth the effort to me. YMMV. Of course, Qimage also allows to do complex print jobs (multiple output sizes of the same file with the highest quality), and it remembers settings used before, and allows to save and recall specific settings and templates from earlier print-jobs, very useful for re-prints.

The LR 5.4 print module for me, is just so much more simple in both the interface (Qimage is not, never has been).  I say this as I have spent about 1/3 of my life either selling, supporting or working PC's and most software I can grasp.  Qimage I can't.  It might just be me, however the interface is not at all straight forward.  

LR still (and most likely never will) have layers, but it has variants.  But for me, keeping the file at 300dpi, no matter if it's Fuji, Canon, Nikon or Phase One, and then letting LR make the final size for the print to 360.  I have also tried the separate uprez software solutions, mentioned earlier, and even on larger output, I personally don't see the difference.  And I am confident the average viewer won't either.

I agree on the convenience part, and sure LR makes things easy and it's not all that poor at upsampling either. But it is also not giving the best quality, nor does it claim to (neither does Mike Chaney, but he does try to push the boundaries to a higher level). LR can use a sort of layered approach with OnOne's Perfect Layers, but that result will be based on a rendered version of the original.

Not saying one method is better than the other in final output, but to me the workflow within LR is superior.

 

Correct, convenience is always welcome. Sometimes good enough is just that, good enough. But I do think that there is more quality to be had than many people are achieving right now. For some, like me, that quality is very high on the list of priorities.

Besides, raising the bar will also force LR to improve its output quality ..., so everybody gets to benefit.

Cheers,
Bart

[Dave Gurtcheff]

Greeting all:
I have read this entire thread, and it is eye opening, as it recommends just the opposite of what I have been doing. I like and make big prints, typically 13"x19", 16"x24" and 20"x30" (24"x32" for 645D files). I am 77 years of age, and not a software expert, so I rely on those that are. I am a BIG fan of Photokit Sharpener, and have used it since day one, and still do. Based on the PK instructions, my work flow has been:
1. Optimize the raw file, add retouching, levels, etc, and save it as a " Master".
2. When I get a print order, say 13" x 19", I rename the file, flatten it, apply PK Capture Sharpen, then size to 13x19. If the file is les than 240 dpi, I upres to 240 using either Blow Up, or Photoshop. With my newer cameras, the files are usually much bigger than 240 ( usually an odd number like, say, 421 dpi). When this is the case, I leave the file as is, apply PK output sharpen, and print.
From what I have read here, I should upres to 360, and if my file is bigger than this, I should upres to 720, then output sharpen, and print. Is this correct?
Thanks
Dave

[digitaldog]

From what I have read here, I should upres to 360, and if my file is bigger than this, I should upres to 720, then output sharpen, and print. Is this correct?
Thanks
Dave

Try one image as suggested and one the way you normally do it. Resize as suggested then run PKS. Then just look at them, you'll know what's now best for you.

[Paul2660]

Greeting all:
I have read this entire thread, and it is eye opening, as it recommends just the opposite of what I have been doing. I like and make big prints, typically 13"x19", 16"x24" and 20"x30" (24"x32" for 645D files). I am 77 years of age, and not a software expert, so I rely on those that are. I am a BIG fan of Photokit Sharpener, and have used it since day one, and still do. Based on the PK instructions, my work flow has been:
1. Optimize the raw file, add retouching, levels, etc, and save it as a " Master".
2. When I get a print order, say 13" x 19", I rename the file, flatten it, apply PK Capture Sharpen, then size to 13x19. If the file is les than 240 dpi, I upres to 240 using either Blow Up, or Photoshop. With my newer cameras, the files are usually much bigger than 240 ( usually an odd number like, say, 421 dpi). When this is the case, I leave the file as is, apply PK output sharpen, and print.
From what I have read here, I should upres to 360, and if my file is bigger than this, I should upres to 720, then output sharpen, and print. Is this correct?
Thanks
Dave

Yes,

And if you print from LR, you can just let LR do the final uprez to 360 or 720.   This is what I am doing for prints up to 36 x 72 and they all look great.  The key for me is just letting LR do the final uprez.  I am printing images from various cameras, MFD, and 35mm digital.   As I mentioned in earlier posts, I have tried everything but Qimage.   I tried Blowup a few years ago, and did not like the results, but this is something that is going to to unique to each person's eye.  I only use 720ppi on smaller images, say 18 x 25 or down.

I have worked with Qimage, Photozoom Pro, Perfect Resize mainly.  I have tried up rezing the file in each of these to 360dpi, and printing the same file from LR at 360ppi, to me the differences are not enough to warrant the extra time and work since LR just seems to do as good a job.   

Jeff's book, "The Digital Print" goes into more detail on this issue and is well worth the read. 

The Print Module in LR will take a bit of getting used to, but to me it's a great overall solution.

Paul

[Dave Gurtcheff]

Thanks digitaldog and Paul. I have LR and never use it. Old habits die hard. I use Bridge to review a new shoot, then convert from RAW using DXO if the camera/lens is supported, then post process in PS and use that for printing. I need to expand my horizons, thanks
Dave

[digitaldog]

Just printed another Roman 16, 7x10@300 is the resolution. Printed from Lightroom:
1. Native rez (check box off, send 300ppi)
2. Resample to 360 (Print Resolution check box on)
3. Resample to 360 and Finest Detail on in driver.

Epson 3880, Luster paper, custom profile, OS X 10.9.3 latest version of LR.

Photographer's viewing distance, they look identical to me.

[Manoli]

3. Resample to 360 and Finest Detail on in driver.

Andrew,
Shouldn't no.3  be resample to 720 AND Finest Detail on in driver. ?
As you sent it, we're back to the OS / print driver presumably doing the interpolation.

But I do agree with you, that you'll need a loupe to possibly see the difference, other than in large expanses of almost monochromatic colour or exceedingly fine detail.

[digitaldog]

Andrew,
Shouldn't no.3  be resample to 720 AND Finest Detail on in driver. ?

Yes and maybe no <g>. When I do set it to 720, the LR warning triangle appears because the native resolution is too low. I could do that but that now means a new rule (if less than 360, set 720 and Finest Detail). And considering I see nothing different from all three, I'd be shocked if upsizing from 300 to 720 would do anything useful.

[Jim Kasson]

I think we'd all be better off if the printer driver and/or OS didn't resample the image to some power-of-two ratio of that of the marking engine prior to doing diffusion/dither halftoning. Oh, I see how if makes the driver easier to write, and probably makes it faster to boot, but with today's computers the driver should be able to do some more work and still keep up with the print head. If the drivers worked that way, we photographers could think about more important things than printer resampling.

But what do I know? The last halftoner I wrote was twenty years ago, to check out some ideas about tradeoffs between marking engine resolution and number of inks, and it never was anywhere near production quality. Since the printer was simulated, it could be molasses-slow, too.

If the driver is going to continue to halftone from resampled images, can we at least put the blame where it belongs and talk about the "native resolution" of the printer driver not the native resolution of the printer?

More thoughts on the subject here.

Jim

[Bart_van_der_Wolf]

And considering I see nothing different from all three, I'd be shocked if upsizing from 300 to 720 would do anything useful.

Hi Andrew,

Get ready to be shocked ...

When using dedicated upsampling software, such as Perfect Resize or PhotoZoom Pro, actual resolution is added on edge detail. That means that the edges do not get upscaled the same as other image areas, but they remain sharp and lines stay thin, and thus at a higher resolution. Lightroom apparently does not do that, it uses a more classical upsampling approach.

This is not new, but you may have missed the discussion which took place more than two years ago ...

Recent versions of Perfect Resize do a reasonably good job at this, but PhotoZoom Pro does it a bit cleaner IMHO. The differences in actual print between the two will not be that noticeable unless one does really huge output. Also Alienskin's Blowup adds this type of edge resolution, but in my experience it tends to round/fill sharp corners more than PhotoZoom Pro.

Cheers,
Bart

[Mark D Segal]

Bart, if the main difference is at the edges, would a slight difference in sharpening technique bridge the performance you perceive between the various approaches?

[Bart_van_der_Wolf]

Bart, if the main difference is at the edges, would a slight difference in sharpening technique bridge the performance you perceive between the various approaches?

Hi Mark,

Yes, an adaptive sharpening technique might achieve similar results, but it will have to work on a much larger data-set so it won't be lightning fast. One thing it certainly can do is remove some of the upsampling blur by using deconvolution. Another thing is vectorizing the edge-like detail, and replace/blend that sharp and smoothed edge in (with a generic edge mask or by painting it in). Again, basing this on the smaller image is faster.

In fact, I've been experimenting with selective edge enhancement with a Photoshop layer that was treated with Topaz Labs Clean and or Simplify. However, that does base its processing on the already upsampled data, and its quality. It can turn aliased edge detail into jaggies. So it is probably more effective to do it correctly right from the start, with the upsampling routine.

The three mentioned resampling applications, Perfect Resize, Blow-up, and Photozoom Pro, do some mix between vetorizing and resampling the original data. Obviously, this will benefit some image detail more than other detail.

Cheers,
Bart