Digital printing: pixels, resolution and resampling

[RichC]

Advertisement (gone after registration)

Admin's note:

This great post was originally posted as reply to a member's question,

 

Definitions

 

First, we need to define a few terms.

 

Halftone. Most printers (inkjets, laser, printing presses) have inks of only a few colours, so a process known as halftoning is used to produce new shades: tiny dots of ink are printed very close together or overlapping, and, as these dots are so small, the human eye blends them together to give the appearance of new colours. If you look at a photograph in a book using a magnifying glass, you can see these individual dots of ink.

 

Colour halftoning. Dots of cyan, magenta, yellow and black (CMYK), respectively, which, combined, create a halftone pattern (greatly magnified). The eye merges theses dots, to give a continuous tone, the colour depending on the CMYK proportions

 

Resolution. For our purposes, resolution is the number of pixels or printed dots per linear inch comprising a digital or printed photograph, respectively. Resolution is measured in ppi or dpi (see below). Note the word ‘linear’: a resolution of 300 dpi means 300 dots vertically and 300 dots horizontally.

 

Dots per inch (dpi). The resolution of a printer is the number of printed dots of ink per inch, typically 300 dpi for a photograph printed on an offset printing press (for books, etc.). Despite metrication, the UK still measures print resolution in dots per inch, not per centimetre.

Pixels per inch (ppi). A pixel is the smallest element of a single colour that can exist in a digital image. The resolution of digital images is measured in pixels per inch. Often, ppi and dpi are used interchangeably, but this leads to confusion: use the former for the resolution of computer images on screen and the latter when discussing the output of printers (e.g. inkjet prints).

 

 

Relationship between ppi and print size

 

Resolution by itself tells us nothing about how large a particular image can be printed – it just defines the quality requirement, e.g. 72 ppi for a web page or 300 ppi for a printed magazine. The pixel dimensions of a digital image is the other factor. As an example, a 3000 × 2000 pixel image can be printed in a magazine at its optimum quality at a size of 3000/300 × 2000/300 = 10 × 6.6 inches.

 

 

300 ppi: the ‘standard’ resolution. Or is it?

 

The human eye cannot resolve much more than about 500 dots per linear inch when looking very closely at complex patterns of dots with indistinct borders between abstract shapes of different tints/shades (e.g. photographs). In practice, most people will probably not notice any difference if the dot spacing is reduced to as low as 300 dpi – and definitely not at a viewing distance of 30 cm or more. For this reason, 300 dpi has been the resolution used for decades in the print industry when printing halftoned photographs in high-quality books and magazines. It is pointless having a resolution above this limit, as the typical offset printing press is unable to use the greater detail: all that you’re doing is creating a large file that is unwieldy and wastes space. (Incidentally, shapes with more defined structure, e.g. text or line drawings, are printed in the way you expect, using solid, unbroken areas of ink – but require a higher resolution of at least 1200 dpi to ensure that the edges look crisp.)

 

In a traditional offset printing press, the dots in a halftone are regularly spaced, and we can assume a direct numerical correlation between pixel density (ppi) and the printer output resolution (dpi), i.e. each ink dot represents a pixel. So, when a photograph is to be printed in a book or magazine using an offset printing press, we need a digital photograph with a resolution of 300 ppi, to meet the 300 dpi requirement.

 

Before continuing, it’s helpful to understand how an inkjet printer works, as it’s not the same as an offset press. A pixel is represented using several overlapping ink droplets. In addition, the size of these ink droplets varies, and the printer uses a regular grid pattern when printing these droplets. There is thus no obvious direct correlation between dots of ink and image pixels, as there is with the offset printing press. So, what resolution should a digital image have?

 

The aforementioned 300 ppi figure is widely, but erroneously, assumed to hold true for inkjet printers: although offset presses are designed to print photographs at 300 dpi, many inkjet printers aren’t, and require a different resolution to achieve the best-quality print. Printing at a non-ideal resolution can cause a loss in sharpness and produce artefacts such as moiré.

 

Epson inkjet desktop printers (such as my 2100 model) typically have a resolution of 2880 × 1440 dpi. This refers to the total number of ink droplets printed, and is thus not the printer resolution: as discussed, these droplets are printed in a grid pattern, and it is this grid that actually defines the printer resolution – so, the huge 2880 dpi ‘resolution’ is just marketing hype! Inkjet manufacturers don’t tell us is what the true resolution is, presumably because it will seem disappointingly low. Also, manufacturers give the resolution horizontally and vertically, but only the lower of the two figures (1440 dpi) is relevant - once again, marketing hype to bamboozle us consumers.

 

Using files with a resolution (e.g. 300 ppi) that does not match an Epson inkjet printer, degrades print quality, because the print driver resamples the image during printing, affecting sharpness and tonal range, as there is no longer a 1:1 correspondence between pixels and ink dots. For most prints, this degradation is minimal and may not be discernible – but why accept lower print quality?

 

We thus need to find out the resolution of this aforementioned grid of grouped ink droplets. It makes sense that it would be close to the 300 dpi ‘standard’. Also, it has to be divisible into 1440 dpi. That gives us a choice of 240, 288 and 360 ppi. Tests (e.g. see Inkjet Resolution) show that 288 ppi consistently gives the best results, regardless of printer settings. So, the native resolution of Epson printers is 288 ppi. But, as I said, although there's no reason not to use 288 ppi, choosing a less ideal resolution such as 240 ppi may not make a discernible difference (unless you're into examining prints with a loupe!).

 

Canon inkjet printers have a resolution divisible by 300, and use the traditional 300 ppi resolution.

 

For the rest of these posts, I’m going to use resolutions suitable for Epson printers, since that's what I use.

 

 

Maximum print sizes from digital images

 

A 6 MP camera produces an image of roughly 3000 x 2000 pixels, so an image resolution of 288 ppi will give you a print 2000/288 x 3000/288 = 7 x 10.5 inches in size – about A4.

 

The Leica M8, a 10 MP camera, produces an image of 2600 x 3900 pixels, sufficient for a 9 x 13.5 inch print at 288 ppi, and good enough for an A3 print.

 

For non-critical prints to be viewed at a typical reading distance, the 288 ppi criterion can be lowered to about 180 ppi. This quality corresponds, roughly, to that of photos in expensive coffee table books. If you think about photographs that you've seen reproduced in high-quality publications, it’s apparent that this resolution is capable of producing what, too most people, is excellent print quality.

 

At 180 ppi resolution, a 10 MP image can be printed at about A2.

 

If printing at 180 ppi, the photographs must be perfect, i.e. with minimal noise, without sharpening or JPEG artefacts (‘jaggies’) and produced using a high-quality camera or scanner.

 

For viewing at greater than reading distance (i.e. for prints larger than A2), the resolution requirements are less stringent, as viewers will be further away: fine-art prints for wall display are usually viewed from a distance of at least 2–3 feet). A resolution as low as 120 ppi can still produce arresting images. This means that most people would be impressed by a good 6 MP image as large as 20 x 30 inches, as long as they were viewing it from a few feet away.

 

Viewing a photo on-screen at 50% magnification will give an approximation of the appearance (sharpness, etc.) of the printed image.

 

[End of Part 1]

Edited February 15, 2010 by admin
typo

[RichC]

Maximising image quality for printing

 

I’m only going to mention sharpening and noise reduction in this section, but there are a host of other things that need to be considered.

 

 

Noise reduction

 

Noise reduction should be done sparingly and selectively, and is the first stage in post-production of a digital photograph. A common mistake is to remove all noise, to the detriment of fine detail and image quality. Consider whether the noise will be visible in the printed image – which it may not be if the image contains a lot of texture (foliage, sand) or is to be printed A3 or smaller.

 

The raw converter may not be the best tool for removing noise: dedicated tools such as the NoiseNinja Photoshop plug-in, allow finer, more selective control.

 

 

Sharpening

 

All digital photographs need sharpening. Sharpening is best done manually, and so automatic sharpening should be turned off in the raw converter. That said, sharpening can affect the overall contrast, and you may prefer the look of images with a tiny amount of sharpening added in the raw converter (most of the sharpening should be done later). Similarly, noise reduction should be left until later. Both noise reduction and sharpening are done on the exported TIFF file, in image-editing software such as Photoshop.

 

There are three types:

 

1. capture sharpening

2. creative sharpening

3. output (or print) sharpening.

 

Capture sharpening and creative sharpening are done simultaneously, and is the next stage in post-production after noise reduction. Capture sharpening removes the blur created by the camera sensor, notably by a filter in front of the sensor (the anti-alias filter - OK, I know the Leica M8 lacks this, but it still benefits from some sharpening). The sharpening required will be the same for every photograph taken by a particular camera. Capture sharpening should maximise image sharpness without creating artefacts. Creative sharpening is selective sharpening of an image: for example, in a portrait you may wish to maximise the sharpness of the eyes, have the rest of the face less sharp, and not sharpen the background.

 

Other adjustments to the image are now done, such as conversion to black and white, and cloning out dust.

 

The final stage is output sharpening, which applied to images ready for printing, to compensate for the sharpness that will be lost during printing (caused by the spread of ink droplets).

 

 

Increasing the maximum size

 

If you want to print a photograph much larger than its maximum size (i.e. pixel dimensions divided by ppi), the print quality can be improved by 'resampling', i.e. the photo is enlarged by adding extra pixels, padding out the photo and making it look smoother – thus avoiding that jagged pixellated look seen when you simply magnify an image. Resampling uses mathematical techniques called interpolation to add the extra pixels as seamlessly as possible.

 

Professional print-makers, such as David Adamson, regularly interpolate 10 MP photos from high-end cameras such as the Leica M8 to create 30 x 40 inch inkjet prints.

 

 

When to resample

 

If you can avoid resampling, you should, as resampling can unnecessarily degrade image quality. A cut-off of about 180 ppi for a minimum print size of A3 is reasonable, taking into account that larger prints will be viewed from further away. It's worthwhile making test prints of part of your photo, unsampled and resampled, to see which you prefer.

 

If you know you will be creating a large print from an image, increase the size of the photo by resampling as the first step in your workflow (second, if cropping), and definitely before sharpening. Resampling later can lead to artefacts and reduced image quality.

 

 

Resampling methods

 

There are many resampling techniques available for use on image files. If you are enlarging up to 200%, all methods provide similar-looking prints, so Photoshop’s bicubic interpolation algorithm should suffice. Later versions of Photoshop have an alternative method, bicubic smoother, producing, as the name suggests, a smoother-looking image, which may give better-looking results, depending on the image.

 

However, more sophisticated tools are available, such as PhotoZoom and Alien skin Blow Up. Various tools are compared here (Digital Photo Interpolation Review).

 

If you use a PC, Qimage (DDI Software) is a program dedicated to printing digital photographs, and has a choice of several interpolation methods, and gives excellent results: choose your print size, and Qimage will automatically resample the image, apply output sharpening and print the image. I always use this. Yes, you can get the same quality using Photoshop, etc., but it's a lot less effort - not only does Qimage automate printing but it saves settings and entire print jobs.

 

Raw converters can output interpolate files, but this is not recommended as the resampling techniques aren’t usually as sophisticated as those in Photoshop or other image-editing programs, and can create artefacts.

 

 

Sharpening resampled images

Sharpening resampled images needs care, as it can create artefacts: do not sharpen the whole image – only those regions that need it (e.g. eyes).

 

Some images benefit from being resampled to about 20% larger than the target size, using a gentle interpolation technique (e.g. Photoshop’s bicubic smoother), sharpened, then resampled to the target size, followed by a tiny bit more sharpening.

 

 

Adding grain

 

Sometimes noise is a good thing!

 

Some people add ‘grain’ to digital images, after resampling and sharpening – this grain is much finer than the native grain of film, and should be at the limit of perception. The addition of this noise, though barely visible, adds visual texture and enhances the perception of detail, and is an extremely effective technique for large digital prints. If you’re using Photoshop, try 2% monochromatic Gaussian noise on a duplicate layer with 70% transparency.

 

I like the look of large prints treated this way, and often add noise to prints larger than A3 after resampling the image.

[RichC]

Summary and postscript

 

(1) The image resolution should be about 300 ppi, to match the resolving power of the human eye. The ideal image resolution should match that of the printer for optimum print quality - but the reduction in quality by not using the ideal resolution is minimal and possibly not noticeable, provided it's close to 300 ppi.

 

Note that when setting the ppi in Photoshop, the dialog box tells you the print dimensions in inches (or centimetres, etc.) that matches this ppi - this ppi is correct only for that size, because if you print at a larger size, the ppi will reduce since you're stretching the image (the number of pixels can't change). This leads onto the next point.

 

(2) The dimensions of your photo in pixels is the important factor when considering resolution - divide the image dimensions in pixels by the desired print size in inches. As mentioned in point (1), this should be about 300 ppi (minimum of about 200 ppi, if printing at A3 or larger.

 

Examples:

Consider an A3 (16 inch) print. Using an 8 MP or higher camera (and not cropping the image much), the image dimensions in pixels are adequate to give a ppi above 180 ppi, so you don't need to check anything: just set the ppi in Photoshop (or whatever), then print. However, if you're cropping heavily or making a large print, do check that the ppi isn't too low. For example, the Epson R-D1 I used to own produces 6 MP images of 3008 x 2000 pixels: if I wanted to make an A3 print, then 3008/16 = 190 ppi, which is starting to push the envelope somewhat - A3 is still small enough for people to comfortably view close up at the same distance as smaller prints, so although 190 ppi doesn't look bad, smaller prints can look better, as do A3 prints from cameras with more megapixels (and hence a higher ppi). This is one reason why I upgraded to the 10 MP Leica M8, which can handle A3 printing easily (3936 pixels wide, so 3936/16 = 250 ppi), as well as, just about, A2 (24 inch) prints - 3936/24 = 164 ppi (to view A2 prints properly, you need to stand back, so 164 ppi is adequate as viewing from further away reduces the required resolution).* The Leica M9 with its 18 MP (5212 x 3472 pixel) images is almost overkill for A3 prints, as 5212/16 = 325 ppi (differences between M8 and M9 A3-sized prints of the same subject will be inconsequential as the ppi for both cameras at this print size is excellent - possibly slightly more detail may be seen in some A3-sized M9 images if you look very carefully with a magnifying glass), but A2 prints will be improved (5212/24 = 220 ppi for the M9 compared with 164 ppi for the M8).

 

[ * I've noticed an annoying tendency for some people to stick their noses 2 inches from a very large print and mutter about sharpness - probably the same ones who worry about the sharpness of lenses! Just slap them across the back of the head and tell them to stand back a few feet! ]

 

(3) Avoid resampling, provided the calculated resolution remains above 180 ppi - resampling can affect image quality even if using dedicated software like Genuine Fractals. If the resolution works out close to this low limit, print resampled and non-resampled images to see which you prefer.

 

If you want to resample, I recommend SizeFixer (www.fixerlabs.com) - provided your camera is supported (which the Leica M8 is), so you can use "Super Resolution", its results are superior to other resampling programs (I spent a weekend testing them) - make sure the results don't have an overdone "oversharpened" look, by carefully adjusting the settings. The problem with most resampling programs is that they can produce geometric artifacts.

 

There's a Photoshop technique that can give better results than most resampling programs: see www.outbackphoto.com/workflow/wf_60/essay.html.

 

(4) Anything much higher than 300 ppi is waste of printing time and file space, as the extra detail can't be seen or printed. And most definitely don't use image resolutions that match the so-called printer resolutions of 5760 dpi, 1440 dpi, etc. - these are not true resolutions of the kind I'm talking about, and serve only to confuse - marketing departments quote them simply because they look impressive!

 

(5) This is really just reinforcing point (2)... Pay attention to your image size in pixels. I work in publishing, and a common problem is clients who ignore image size - I'm often sent images at the requested 300 ppi resolution (these are destined for a printing press, so 300 ppi is what I want) but which are too small to use: an image may well be 300 ppi, but if it is 600 pixels wide, say, it will be only 600/300 = 2 inches wide if printed at optimal quality!

Edited February 14, 2010 by RichC

[RichC]

A bit more on the theory and techniques of resampling, building on my posts above.

 

As I defined briefly under the "Increasing the maximum size" heading in my earlier post, interpolation - often called (less accurately) "resampling" or "uprezzing" - of a digital image is the process of increasing its dimensions by adding extra pixels.

 

As an example, my Leica M8 creates a 10 MP image file, which has dimensions of 3904 x 2620 pixels. However, my photo agency requires 18 MP files - an image of 5126 x 3440 pixels. I thus resample my M8 photos in software like Photoshop to increase their size to 5126 x 3440 pixels. The software is adding [(18 MP - 10 MP)/(10 MP)] x 100 = 80 per cent more pixels to my photo.

 

So, how does the software add these extra pixels? The answer is carefully, using a mathematical equation to examine the colour and arrangement of pixels in an image. The data from this equation is then used to add new pixels that resemble their original neighbours closely. It's not an easy task to guess which pixels to add where. Consider a hair - a thin, curved line: when an image of a hair is resampled, as well as retaining accurate colour gradients, the edge of the hair must keep its curve whilst still appearing sharp against the background- what you don't want is, for example, sudden jumps in colour ("blocking") nor a jagged edge ("stairstepping") or general blurriness.

 

There are many equations used for interpolation. A common method is bicubic, used in Photoshop - it's pretty good, but there are more complex methods that can give better results. I say "can" because all methods have pros and cons: for example, fractal interpolation (which concentrates on the patterns within an image), used by several resampling programs such as Genuine Fractals, gives the appearance of greater detail and sharpness compared with more traditional, basic methods such as bicubic, but can produce unsightly geometric artefacts - e.g. granular surfaces in an image such as sand or leaves can end up with pronounced unnatural-looking ridges and swirls. Bicubic resampling gives less sharp results than fractal resampling, but is far less prone to producing artefacts that would be visible in a print. Like most things in life, there is no "one size fits all" solution...

 

An example interpolation equation (the bicubic formula):

 

 

This website gives a brief overview of the main methods: www.dpreview.com/learn/?/Glossary/Digital_Imaging/Interpolation_01.htm

and this one compares various resampling methods and programs for digital images: www.americaswonderlands.com/digital_photo_interpolation.htm

 

If you're resampling to create an image up to about 50% larger, it really doesn't make much difference which resampling method you use: the print will look the same, even under a loupe.

 

If you're resampling to over 50% (like my Leica M8 photos - 80%, which I consider to be about the practical limit, beyond which prints start to look poor), the choice of resampling method does matter. I use two methods:

 

1. Photoshop (see www.outbackphoto.com/workflow/wf_60/essay.html):

(a) Optimise my photograph, including "capture sharpening"*

( Resample using bicubic smoother to 20 per cent larger than my target size

© Examine the image and make any changes I deem necessary (e.g. further capture sharpening - don't overdo this).

(d) Resample to your target size using bicubic sharper, and again make any changes I think necessary such as more sharpening (often not needed).

 

* See my earlier post, above.

 

2. The SizeFixer program (www.fixerlabs.com).

 

Often, both methods give similar results, and which is used doesn't matter. Sometimes, one method gives superior results. Rarely, the methods produce images that have areas that are noticeably both better and worse when compared - in which case I'll combine the best bits of both images to create the "perfect" image.

Edited January 24, 2011 by RichC

[RichC]

Epson R3000 ... prefers 360 PPI.

 

Original poster here...

 

Although I haven't updated this thread, I get slightly better results at 360 dpi with my Epson R2880 printer too compared with 288 ppi. My old Epson R2100 definitely preferred 288 dpi, as did my friend's Epson R2400.

 

Bidirectional ("fast") printing can give inferior results, with banding, etc.: for optimum quality, it's always best to turn fast printing off (despite your tests showing no difference at 360 ppi).

 

Thanks for doing these tests and posting the results - it seems that

the native printer resolution of current Epson printers like the R3000 is now 360 dpi

 

so, 360 ppi resolution images should be sent to them.

 

Apart from this resolution change, the info in my posts remains valid.