Going Digital
The Resolution Conundrum:
How much is enough?
What dpi should I scan my photo at? How big should a digital file be for an 8X10 transparency? What is the best dpi for my desktop printer? The answers to these questions are often given without explanation, dismissing the inquirer with a “because that’s the way it’s always done” sort of response. I am going to attempt to illuminate some of the mysteries surrounding the issue of resolution in digital images before making specific “real world” recommendations for dpi settings. You might find some surprises along the way so bare with me as we dive into the arcane world of pixels, lines, screens and dots.
True story: A designer had created some artwork for a movie poster and sent the digital file to two separate output services. The first place made a continuous tone “LVT” transparency output on 8”x 10” Ektachrome film. The second place was a printer who was supposed to use the digital file to create the plates for printing the poster. The printer called up the designer to say that they couldn’t use the digital file for the poster because it wasn’t high enough resolution for poster size 27”x 40” output. The first place made a “res 24” 8”x 10” transparency from the 75 meg file that was sharper than camera original 8 x10 film so the designer sent the transparency to the printer who then scanned it to a 350 meg file for use in the poster. Why couldn’t the printer use the same digital file that created the transparency? It seems that, even among professionals, there is considerable confusion over resolution issues in photo-digital imaging.
Resolution = pixel density per inch
When someone talks about the resolution of a digital image they are generally referring to the density of pixel information per inch at a particular image size. This is most frequently expressed as the number of pixels per inch or dots per inch, i.e. 72 dpi . You almost never see 72 ppi even though pixels per inch is not exactly the same as dots per inch (more on this later). The size at which an image is displayed is critical to our understanding of resolution and it is often misunderstood. To say that an image is 150 dpi is meaningless without knowing the dimensions in inches of the image, the “i” part of dpi. Photographers often equate resolution with the “graininess” of an image and a low-res image is thought to be grainy. Resolution is also equated with focus and low-res images are thought of as soft or blurry. Grain and focus are issues that are completely separate from resolution in digital images.
Pixels have no inherent size
Pixels are generally thought of as being very small, however pixels can be represented at any size; pixels themselves have no dimension at all. A pixel is a numeric expression of a square field of color. The pixels in an image pack together without seams in a ordered grid. The red square below could be a representation of one pixel,100 or 1000 pixels that have the same red value.
By randomly altering the values of every other pixel by 20% we can see that in fact we have a grid of 10 x 10 pixels in a 1 inch square and this image would have a resolution of 10 pixels per inch.
We can say that this image looks pixelated because we can see every individual pixel in this image. Now step back and look at this image from 10 feet. Can you see the pixels? How about at 20 feet?
Viewing Distance Influences the Appearance of Pixels
Viewing distance has a great influence on the appearance of pixels in an image and an image that has a pixel density of 100 pixels per inch may look course when viewed at a distance of 12 inches but at 20 feet we will not be aware of pixels at all. This effect is the magic at work in the new photo-mosaic images that have become popular. When viewed at a distance the primary image is quite apparent. When you get close you can see the many tiny little images that actually make up the pixels of the primary image.
If you are aware of the pixels in an image at the intended viewing distance then that image can be said to be a low resolution image for that distance. Viewing distance is not often fixed and so we want to be sure that our images look good at the closest distance that they are likely to be viewed at. In our story about the designer and the movie poster, the transparency was printed at a pixel density of res 24. This is a term borrowed from Scitex, the dominant company in the pre-press industry; it means pixels per millimeter, in this case, 24 pixels per millimeter which is equivalent to 609.6 pixels per inch. At this pixel density you could put a loupe on the transparency and be hard pressed to find any pixels in the image. At a normal viewing distance of 12” to 20” you see only a smooth, detailed image. The movie poster would make the 5600 pixels of the long dimension in the image fit the 40” height of the poster and that would give us a pixel density of 140 pixels per inch. Now at a normal viewing distance of 3 to 6 feet you are not going to be aware of pixels in the image at all. However, if you close in to 1 foot or less ( lets say you wanted to read the billing block at the bottom of the poster) you could begin to see some pixelation in the image especially in diagonal lines or edges. This is the reason behind the knee jerk reaction of the printer.
But how could scanning the transparency of the original image in our story improve the resolution to a density of 300 pixels per inch and make the printer happy? The image doesn’t change does it? This is the most perplexing aspect of digital imaging and it is at the heart of the question: how much resolution is enough?
Interpolation Changes the Pixel Count
It turns out that there are many ways to increase the pixel count of an image to suit a particular viewing distance. Re-scanning the digital transparency to a higher pixel count than the original digital file is one method favored by pre-press services because by doing that they can also take advantage of their high-end scanners CMYK conversion and unsharp masking capabilities to improve the print reproduction of the file. The pixels rendered on the film of the digital transparency are slightly blurred and the edges blend together a bit in a kind of natural anti-aliasing. When the image on film is re-scanned to a new size (pixel count) additional sharp pixels are used to enlarge the original blurry pixels. In essence the scanner interpolates pixel information to add pixels to the original file, thus increasing its size. We can achieve a similar kind of image scaling in Photoshop using bicubic interpolation. This is a fairly sophisticated method of averaging the value of two adjacent pixels to put a pixel in between them and expand the size of an image. This does far less damage to the quality of an image than projecting an image through a lens on to photographic paper although the type of quality loss is similar: the image loses focus. When we have an image that has 5600 pixels in the long dimension we can easily interpolate that to 11200 pixels and make even the finicky printer of our story happy with the digital file.
How far can we take this? What is the minimum pixel count we can get away with and what is the ideal resolution for digital images printed at a given size? That depends on a number of factors and there is no hard and fast rule but, more often than not, we can get away with less than what is commonly recommended.
Reproduction Methods Influence the Appearance of Pixels
All our concerns about resolution in digital images center around the different methods of reproducing those images so that they are intelligible to the human eye. Continuous tone methods, like photographic prints are an ideal way to present any image. Usually a continuous tone printer, like a dye-sub printer, is capable of printing at 300 pixels per inch. A Fujix printer (which uses photographic material similar to type-R paper) can print up to 400 pixels per inch. Here more pixels per inch will always look better but in practice it is hard to tell the difference between 320 ppi and 400 ppi because 320 pixels per inch is close to the resolution threshold of the paper (put a loupe down on a Fujix print and check it out for yourself). While you can certainly make a case for an image that has a lot of fine straight diagonal lines in it, 200 pixels per inch is usually the minimum resolution for an image to appear smooth and unpixelated at a 12 inch viewing distance with a continuous tone print. Images printed with dithered tone methods are another story.
By dithered tone I mean any method that achieves the appearance of continuous tone by placing a limited number ( usually 4 to 6) of colored dots in close proximity and varying either the size or the density of the different colored dots to visually mix different colors. For reasons which will become clear later it is harder to define the discreet pixels of a digital image with clusters of different colored dots than continuous tone methods. Various clusters of different colored dots are used by all ink on paper methods to reproduce pixels of a digital image. Traditional offset lithography uses a method of dithered tone generally referred to as a line screen. This is a regular pattern of cyan, magenta, yellow and black dots arranged in clusters called rosettes where the distance between the centers of the dots is fixed and the size of individual dots varies in order to change the overall balance of color of the rosette. Line screens are identified by their resolution which is given by the number of lines per inch that the screen can reproduce. Thus a 150 line screen can reproduce 150 lines per square inch. It is easy to see that if we have black lines against white that are a single pixel in width we need to put a single pixel line of white between the black lines in order to see lines at all. So we double the lines to come up with 300 pixels per inch for a 150 line screen. This is a little misleading because, though it may be true for black lines on a white background it is not easy for a traditional rosette to resolve light brown lines against a dusty green background. This is where we discover that pixels are not dots.
Ink Dot Clusters Define Pixel Colors
Actually, we need a certain minimum of different colored ink dots in order to define the color of a pixel. The dusty green color of a single pixel will need a mix of cyan, magenta, yellow and possibly black dots to fully define it’s color. Subtle differences between adjacent pixels are averaged together in the pattern of dots and a higher pixel density for a line screen will result in a narrower range of tones reproduced because more pixels are averaged together for a given dot cluster. In practice there is a trade off between image detail and tonal range. A digital image can easily contain a smooth gradation that has a full 300 different tones per inch, a 150 line screen is never going to be able to distinguish more than 150 or so of those tones because single pixel values that close together will be averaged together in the rosette pattern. If we are after maximum tonal range we can see that “double the line screen” is actually wasted resolution. Scitex recommends a pixel density of 1.6 times the line screen; for a 150 line screen that would be 240 pixels per inch. Again, Scitex CT files are generally figured at pixels per millimeter and 240 ppi falls between res 8 (232 ppi) and res 10 (254 ppi) both considered acceptable for a 150 line screen. Black line art, on the other hand, will seem a bit course unless it is imaged at at least double the line screen. For some types of imagery 150 pixels per inch can actually look better on a 150 line screen than 300 ppi
Lets look at resolution as it applies to stochastic screens. A stochastic screen is a method of dithered tone that uses varying density of same size dots to achieve the appearance of continuous tone. A good example of this occurs in the Epson ink jet printers. The original Epson Stylus Photo printer was an instant hit with photographers because it was a very inexpensive printer that made very impressive photo-real color prints with six colors of ink dots. The resolution of the printer was advertised as 720 dots per inch and many photographers, including at least one reviewer, immediately tested the printer with huge 100 megabyte 720 pixel per inch files. 720 dpi for the Epson printer means that the printer is capable of putting 720 individual dots close together in 1 inch. However, if we divide that by the number of different colored dots (6) we get 120 which is the maximum number of 6 color clusters in 1 inch. Unless the color of a pixel matches one of those 6 ink colors we are going to need at least 3 different dots to fully define the color of that pixel. So lets divide 720 by 3; now we have 240 which turns out to be the same resolution that Scitex recommends for 150 line screens. In fact, using any more than 240 pixels per inch in an image printed by this printer is a waste unless we have extremely fine detail with limited tonal/color range (like B+W linework).
To review:
Digital images are made up of discreet squares of color called pixels. Pixels have no inherent size and so the resolution of digital images is always expressed by the number of pixels per inch (or millimeter) at a given image size. Images can be considered to be low res when the observer is aware of individual pixels in the image at a viewing distance considered normal for that image use. Continuous tone methods for printing pixel images can define each pixel with a solid color and can easily handle pixel counts up to the maximum addressable by the device without loss in quality. Dithered tone methods (like traditional 4-color printing) deliver the appearance of continuous tone by placing individual dots of color close together to visually mix colors. Tonal range in these methods is limited by the density of dots the printer is capable of placing on a page and how many dots are used to define the color of a pixel. The quality of an image reproduced by dithered tone methods is a trade off between image detail and range of tones.
Recomendations
So what is the minimum pixel density for good quality in a digital image and what is the ideal, or best pixel density? The answer depends on the device used to reproduce the pixel image and the size and distance at which we view the image. Small images reproduced with continuous tone methods will generally require higher pixel densities because they invite closer scrutiny. Larger dithered tone images require the lowest pixel density because they are viewed at greater distances (billboards are produced with a 25 line screen - they are viewed at 50 feet or more distance). Sometimes the pixel density is fixed by the output device and digital files will be interpolated to size for the device. Postscript printers will happily rip (rastorize) files to the requested size regardless of the pixel count of the file so you have to be more carefull about what pixel density you use for the size of the output. Practical considerations will often force us to pick a minimum pixel density for very large images and interpolate to a more ideal pixel count. Generally an image created at a pixel count near 4,000 x 5,000 pixels can be interpolated successfully to almost any size and still look good. Remember that large images are viewed at greater distances and so will tolerate a lower pixel resolution. Some suggestions for various output devices follow:
Transparency output, like an LVT, generally requires the highest pixel densities because they are almost always used for larger sized reproductions, whether by projecting an image on a big screen or by re-scanning for a large size print. A minimum resolution of res 20 (508 pixels per inch) or 3810 x 4699 pixels for a 7.5” x 9.25” image with a maximum of res 30 (762 ppi) or 5715 x 7048 pixels for a 7.5”x 9.25” image works well on transparency film.* Smaller image sizes, like a 35mm slide may be better at higher resolution, especially if you are planning on projecting it on a large screen. For images at res 24 and lower ,imaged on LVT, always have the service bureau double the res at output, this forces the LVT to use a smaller spot for the imaging light source and will result in a sharper image. If you are going to make a 4x5 transparency to use for a large Cibachrome print or a Duratrans enlargement aim for at least res 48 (1219 ppi) - the resulting image will tolerate enlargement better.
Fujix prints look best at 320 ppi but can often tolerate 200 ppi images well.
Large Light-Jet prints can be prepared at 150 ppi and interpolated up to the maximum device resolution of 305 ppi at output (have the service bureau do it) I have seen good results with 100 ppi files ( scaled 300%) For practical reasons it becomes difficult to deal with 50”x 50” images at 300 ppi so do yourself a favor and work on a smaller file - interpolating even 400% can work with some images.
Work larger Iris ink jet prints at 150 ppi and have the service bureau scale 200% to 300 ppi. for smaller sizes on watercolor paper there’s no need to work higher than 200 ppi, with glossy stock stick with 300 if you can - all files will have to be interpolated to 300 ppi at size regardless.
Standard SWOP 4-color printing @ 150 line screen is generally best at between 225 to 300 ppi but not higher; interpolate down if your file has a higher pixel density at the size you want. Again, for poster size images you can work at 150 ppi or even 100 ppi and interpolate up for output.
Encad and other ink jet plotter prints will not require higher than 100 ppi because the coarse dither of their stochastic screen will obscure any pixilation and higher pixel densities will not improve the tonal range of the prints.
A good rule of thumb for desktop inkjet printers like the Epson is to devide the output resolution by 3 or 4 (720/3=240 ppi. 1440/4=360 ppi) Larger 17”x 22” prints on something like the Epson 3000 can look fine at lower pixel counts depending on the paper used ( fine art watercolor paper requires less resolution than glossy coated paper)
In conclusion I would like to point out that the perception of quality in digital images has less to do with the resolution of that image at a given size than the tonal range and color subtlety of the image. The type of image also greatly influences the appearance of pixelation in the image; a shot of highly detailed metalwork with fine engraved lines will require a higher pixel density than a beautiful nude with moody lighting. Photographers in general tend to be resolution snobs for no good reason, often wasting money on ultra high res scans and res 40 - 8”x10” transparencies when a 15 meg file is more than enough for a full 8.5” x 11” print. Many people will turn their nose up at the typical 18 meg digital camera file because they claim that it isn’t “like a hi-res scan from real film”; this is ludicrous, a 60 meg scan from a 35mm original only serves to resolve the dye couplers in the emulsion of the film (you get greater grain detail) without adding any image detail. That same 12 meg digital camera file can be scaled 200%, has virtually no grain and can look better at 11”x14” on a continuous tone print than the grainy 35mm slide printed on type-R paper. You have to be aware of the final use of the image and it’s intended viewing distance before you can make blanket judgements on resolution requirements — most commercial uses just don’t require such high resolution. This is not to say that real film is not superior in a fine art setting— its just different. Photo-digital imagery really is a different medium and because of the nature of digital output, lower resolution images can work just as effectively to communicate a full range of visual imformation as higher-res analog ouput.
In digital images sometimes less is more, but more often than not, it is more than enough.
