Wednesday, October 6, 2010

Photoshop and Image Resolution Part 2 -- Print

Please contact me via email:
paul@baldassinifineart.com

“People rush in where angels fear to tread”— Alexander Pope

Vector, raster, grids & dots
Scanning, display and output (even the final printed pieces) all have something in common: they are made up of -- simplistically -- dots. So, the ubiquitous use of “dpi” (dots per inch) seems to be a perfectly correct way of describing all of our resolution needs. Unfortunately its not -- there’s more to it. Lets start with the “dots” we stare at most of the day -- the ones on the monitor.

Scanned images or images created in a digital imaging editing program like Adobe Photoshop are often referred to as raster images, and are distinctly different from the images created in a drawing program, like Adobe Illustrator for example, which are called vector images. Although both are described mathematically as points on a grid, vector images are files whose mathematical basis is predicated on formulas describing lines and curves, and can be enlarged or reduced without affecting their output quality. Thus, vector images are resolution independent,  meaning detail will always appear sharp, scaled to any size and printed to any output device.

The grid points of a raster images, on the other hand, are resolution dependent, meaning that the quality of the image is compromised if enlarged or reduced on any output device. The video display monitor, for example, is an output device made of points (tiny square dots) in a fixed resolution grid. Enlarge the image and you enlarge the grid. Enlarge the grid enough,and you will quickly see the stair-step “jaggies” characteristic of raster images.

The grid points of both raster and vector images are described by numbers, and their mathematics need to be stored. Vector image files, however complex, are smaller than raster image files because the formulas used to create them can be stored compactly. For example, a line can be described mathematically on the grid by identifying its two endpoints only, and those two endpoints will always be in the same place, defining the same detail, relative to the resolution of the grid.

The large file size of a raster image, however, requires that each piece of information on the grid (all of those numbers) have to be stored. More detail requires a finer grid, which means higher resolution and the more points on the grid. That means more numbers, and therefore a larger file.

It’s getting a bit too deep
Zoom out to the maximum in Illustrator, and everything pretty much still looks the same, just bigger. Zoom out to the maximum in Photoshop and you’ll see those grid field points that your image is actually composed of -- lots of small square “dots” called pixels. Each pixel is represented by a particular formula of zeros and ones, commonly referred to as bit depth.

Bit depth as it applies to scanner (input) resolution is easier to understand if you think of it as three-dimensional, whereas imagesetter (output) resolution is only two-dimensional — dependent on the number of dots in the horizontal and vertical directions. Scanners not only have horizontal and vertical directions, but a third one — depth. Even though both scanners and imagesetters describe horizontal and vertical resolution in terms of spi/ppi (samples/pixels per inch) for the former and dpi (dots per inch)  for the latter — depth applies only to scanners or input. It describes the number of bits of grayscale or color information per pixel. Horizontal and vertical resolutions (i.e., scanning at 1,200ppi) are sometimes referred to as spatial resolution, while bit depth is often referred to as tonal resolution. What most people refer to when they use the term resolution, is spatial resolution.

Of bits and bytes
What exactly is in a raster or (bit-mapped) image file? Its those numbers again, and its measured in bits, bytes, kilobytes, megabytes and gigabytes. The smallest piece of information, or data, that can be stored digitally is a bit. It can distinguish only two things (called states): yes or no, black or white, on or off. Just like a light switch. A line art scan would be one bit “deep”and a line art image would be called a 1-bit image, (21) because only one bit is needed to distinguish black from white, like this:

1-bit image
This is why line art file sizes are so small. Grayscale images require more information, hence more bits. Eight bits actually, known as a byte. A 1-bit line art image, measuring 3 x 4 inches , scanned at a spatial resolution of 800ppi (the minimum for line art,) would consume just under one kilobyte (Kb) of disk space (there are 1,024 bytes in a kilobyte.) 

Most grayscale images are eight bits deep; enough to describe 256 levels of gray, considered to be the upper limit of tonal human vision — more than adequate to create the illusion of continuous tone in print. An 8-bit byte can accommodate 256 numeric values, from 0 to 255 because 2 to the 8th power = 256.

Thus, the brightest value of eight binary bits would be the number 255. An 8-bit grayscale image, measuring 3 x 4 inches, scanned at a spatial resolution of 300ppi would weigh in at just over 1,024 kilobytes, which is equal to one megabyte (Mb).

Color images require even more information, and are usually 24 bits deep. A 24-bit color pixel is actually described by the combination of three 8-bit values, one each representing Red, Green, and Blue, and each one having a maximum color value of 255. In Photoshop, the information stored in the Red, Green, and Blue Channels combine together to give us the illusion of color reality that we see onscreen. A 24-bit color image contains enough information to distinguish over 16.7 million different colors (2563 or 224). The file size of 24-bit color images can be quite large — a 24-bit 8 x 10 inch image destined for print media and scanned at a spatial resolution of 300ppi would equal 20.6 megabytes.

There are 1,024 megabytes in a gigabyte (Gb). It would take forty-eight, 24-bit, 8 x 10 inch color images scanned at a spatial resolution of 300ppi to consume that much disk space.

Color by numbers
The data in the image file is organized in columns and rows, like a spreadsheet, or a checkerboard (which always has eight columns and eight rows for a total of 64 squares). The grid of pixels that define the screen of a video monitor might measure 1,280 x 1,024 for a total of 1,236,992 squares.  Thats a lot of squares. These squares or “dots” of a video monitor is simply a data field; our digital image file is an abstract thing -- nothing more than a collection of specific numerical data composed of complex strings of numbers representing the  RGB (Red, Green, Blue) color of every pixel. So then, 255-255-0, or (Red=255, Green=255, Blue=0), would denote one yellow pixel in a screen image, while 45-0-106 would denote a purple pixel, and so on with each image pixel. Nothing more.

Having no real physical size, digital image files become realized ONLY by physical reproduction in some form or another. In other words, the image must be printed (output) to a specified size. Until then, they are nothing more than numbers -- an informational record.  Digital images are usually  printed (output) to a laser or inkjet printer, or an imagesetter. For digital images requiring no physical output, such as those created for use  on world wide web, for example, the output device which your images will be “printed” to is your display monitor.

Display resolution -- no dots, no spots
When scanning or creating images for video monitor display only, resolution merely determines image size. Assuming that greater resolution reveals more detail, it also makes the image larger — much larger that the ability of the output device to display it at 100%. Display screens are just not large enough. Since video monitors are relatively low-resolution devices (typically 70 - 100ppi), to view an image at 100% size and and still fit on the screen means that the scanning resolution or spi (samples per inch) needs to be the same as the monitor resolution. And its NOT 72dpi, a magic number that many seem to think is the set resolution for all video monitors. Better to forget about it.

Video monitors display a fixed area of pixels, such as 1,280 x 1,024, the screen display area of most 17-inch video monitors. It is this “X by Y” number of pixels, like 380 x 260, that matters when describing video image size. And for video, it doesn’t matter that the 380 x 260 image was scanned at 72 or 300ppi — either way, the display area on screen is the same -- but for an image that will be printed, it matters a lot. The same image will appear larger on a 640 x 480 screen than on an 832 x 624 screen, and will look tiny on the 1,600 x 1,200 screen of a high-resolution monitor.

For video (or any other) images, figuring out both the image size and the file size requires some simple math. For example, my Sony Artisan GDM-C520K is set to resolve at 1600 x 1200ppi (pixels per inch). It’s screen display measures 15.5 inches horizontally. So, 1600 ÷ 15.5 = 103ppi (not the usually and wrongly assumed 72ppi of all display monitors). Knowing this then, let’s say we had original color art measuring 4 x 5 inches that required viewing on screen at 100%. Scanning at 103ppi (spatial resolution), with color depth set to 24-bit (8 bits per channel; tonal resolution of “millions of colors”), would create a file whose dimensions (image size) was 412 x 515 (4 x 103 = 412; 5 x 103 = 515) and consume 621Kb (file size) of disk space.

So then, the formula for determining file size is:

(w x h) x bits-per-sample ÷ 8,192 (÷ 1,024 if more than 1Mb) = file size
in pixels       bit-depth            bits/Kb               bits/Mb

412 x 515 = 212,180 x 24 = 5,092,320 ÷ 8,192 = 621Kb

Photoshop Image Size dialog box -- Resolution test



To see this in action just open a new Photoshop document and set the numbers yourself just like in the screen grab above. This file would only be good for posting online. For print media you would need resolution of 300ppi; for inkjet printing you would need resolution of 360ppi. Do the math or plug 360 into the resolution field of the previous test and you get a much larger file weighing in at 7.42Mb.

That’s all there is to it — it really is all in the numbers. And the numbers, at least as far as the scanner and monitor are concerned, are issues of pixels and samples and there are no problems in making an RGB color of any pixel. But other output devices, like laser or inkjet printers and imagesetters, cannot do that and have very sophisticated ways of producing an image. And it’s also about numbers.

The black and white of grayscale
What exactly is grayscale, anyway? One look at a black & white photograph shows that there is little in it that is actually black or white. In fact, most a of photograph’s subject matter contains gray only, of varying shades, which provide both detail and contrast. In general, the more shades of gray in a photograph, the more “realistic” it appears. The “black & white” photography of Edward Weston, Ansel Adams, and Herb Ritts are outstanding examples of continuous-tone grayscale images. A photograph is commonly referred to as a continuous-tone image because of an (almost) unbroken, nearly infinite range of gray tones between black and white.

A printing press, however, can only print one color at a time, so it is unable to directly reproduce a continuous-tone image. Instead, a photograph, or any continuous-tone image, must first be converted into a halftone. Making a halftone is a technique that has been used for centuries by printers to produce the illusion of many shades or tones in an image with only one color. In the case of a grayscale image, that color would be normally be black.

To accomplish this, basically what happens -- or rather what used to happen until sophisticated digital imaging equipment caught up -- is that the photograph is itself “photographed” -- shot through a fine screen, which breaks up the image into a grid of evenly spaced dots (sound familiar?). Varying shades of gray then, are simulated by the variations in the size of these halftone dots. Under magnification, a photographic halftone will reveal dark areas as having many large white dots and lighter areas as having many smaller dark dots. At normal reading distance, however, the dots all blend into the illusion of continuous shades of gray, looking much like the original photograph. The sharpness of a halftone — its resolution — is determined by how closely spaced together those dots are. The spacing of the dots is often referred to as frequency. For example, if the screen through which the photograph was shot consisted of lines of dots spaced at 75 per inch, then the resulting halftone would have a line frequency of 75 lines per inch or 75lpi. In a “traditionally” produced halftone, the nearly continuous gray tones were achieved by varying the sizes of the component dots. But laser printers and imagesetters have fixed resolutions, and cannot produce variable-sized dots. Instead they have a set or fixed grid of equal-sized dots, corresponding to the device’s “resolution”. These dots are, in fact, called spots and the measure of how many of these spots can be made in a linear inch define the addressability of an imagesetter device, and not its resolution, often used interchangeably.

Grids with lots of dots made out of spots
To better understand exactly what you must know and do to create quality images for print requires an explanation of the following terms and concepts:

Imagesetter resolution is a measure of the ability of an imagesetter to render fine detail. Imagesetter resolution is a function of addressability, laser spot size, film transport, film and film processing. The common usage of the term is usually limited to just one portion of this: addressability.

• Addressability is a measure of how many marks (spots) an imagesetter can make in an inch.

• Dot per inch (dpi) is a measure of resolution (addressability). It refers to the number of laser spots in an inch. It has NOTHING to do with scanning or caputure resolution.

• Laser spot is the smallest mark that an imagesetter can make on paper or film. Referred to as spi (spots per inch).

• Spot size is the width of the laser beam in microns, usually measured under a specific set of conditions.

• Screen ruling is a measure of the fineness of a halftone screen. The higher the number, the finer the screen. A common screen ruling used for most sheetfed and web offset printing is 150 lines per inch or 150lpi. This means that there are 150 lines of halftone dots in an inch. The distance from the center of one of these halftone dots to the next would be 1/150 of an inch. Other commonly used screen rulings for offset printing, depending on the characteristics of press and paper, are: 133 or 150lpi, and for newsprint, 85 or 65lpi. A higher screen ruling demands higher resolution.

• Lines per inch (lpi) or line frequency refers to the number of lines of halftone dots in an inch.

• Halftoning is the process used to convert a continuous tone image into a pattern of tiny dots of varying sizes.  The resultant image is referred to as a halftone. These halftone dots create the impression of many shades of gray, but can be reproduced using only one color of ink.

• Halftone dot is used in halftoning to give the impression of gray. Halftone dots of different sizes are used to represent different shades of gray. A digital halftone dot is made up of many laser spots, like this:

Dots, spots, cells and numbers
A fundamental limitation of laser printers stands in the way of producing a halftone in the traditional way: i.e., evenly spaced dots of varying sizes to create the effect of continuous tones. Both laser printers and high-end imagesetters can only make dots of a single size (the size of which varies according to the device’s resolution). These devices “fake” halftone dots of varying sizes by clustering together many laser spots. A laser spot is the smallest mark that an imagemaker can make in an inch, often termed resolution, but more accurately called addressability. The boundary that encloses many of these potential these laser spots represents one halftone dot and is referred to as a cell. Counting all of the potential laser spots within the cell boundary is equal to the number of gray levels that a particular halftone cell is capable of producing.

16 x 16 digital halftone cell showing single laser spot
The 16 x 16 grid shown here, for example, can produce 256 levels of gray (16 x 16 = 256), which is the number of gray levels required to accurately simulate a continuous tone image in print. As we have already seen, the data contained in a digital image is ephemeral, awaiting for us to give it the property of physical size, Thus, an image can always be output as having 256 gray levels and appear tonally correct as any given size on the display screen. But this does not mean that the image is suitable for reproduction at any size.

Actual output image size (dimensions) are inversely proportional to its resolution. For example, a grayscale 4 x 5 inch image to be output to my laser printer at a frequency of 75lpi would require a resolution 150ppi to assure 256 levels of gray. It’s dimensions would be 750 x 600 pixels and  weigh 440Kb. If this image is rescaled in Photoshop to 8 x 10 inches, its resolution would change to 75ppi, but the file size would remain the same (750 x 600 pixels) and still be 440Kb in size. It would look EXACTLY the same on the screen, only the physical dimensions would be different. Thus to produce a full tonal range on my laser printer with a spatial resolution of only 75ppi, the line frequency would have to be an unusually coarse 37.5lpi, useful for silk-screen printing on fabric, perhaps, since detail will be obscured by dot size.

Sharpness and detail (hence quality) of the image is dependent not only on the ability of the input device to sample detail, but also on the ability of the output device to resolve that detail.

Understanding the relationship between both input and output “resolution”, then, is extremely important. It is the first step in the process of acquiring a digital image and can make or break the success of a printed image. If you understand the concepts outlined so far, then you will understand digital halftones and avoid the potential problems in creating them. Posterization, or “banding” of an image, for example is a function of screen ruling, resolution, and the number of grays, and is usually undesirable.

As stated already, the relationship between the screen ruling of an image and the resolution at which the image will be output plays an integral role in image quality.  Fortunately, any value between 100 and 256 levels of gray will usually be enough to produce an acceptable digital image for print, with paper quality often being the overriding determining factor.

The rule of sixteen
I mentioned earlier that the number 16 would show up again and again. 16 x 16 = 256, for example -- the other magic number, and if you multiply 256 x 4, you get one kilobyte. That said, the easiest way to calculate the highest screen ruling to use on any particular output device is simply divide its addressability (resolution) by 16.

For example, the addressability or resolution of my laser printer is 1,200 laser spots per inch -- usually referred to as 1,200dpi. If you divide 1,200 by 16 you get 75. This means that there are seventy-five halftone dot cells per inch, measuring 16 x 16, each cell containing 256 potential laser spots, thus allowing 256 levels of gray. 75lpi is the highest screen frequency available with the greatest levels of gray. My laser printer is physically not able to output at a higher resolution and maintain 256 levels of gray. The device can produce continuous gray tones and high resolution but it can’t do both at the same time.

If, for example, I output my image at 150lpi, the grid would change to 8 x 8 cells, (1,200 ÷ 150 = 8). However, the number of gray levels decreases to only 64. Why? Because a halftone cell dot measuring 8 x 8 contains only 64 potential laser spots, yielding only 64 levels of gray which would result in banding and abrupt transitions in the image.

To output the image at 150lpi and contain 256 gray levels, would require a device capable of producing 150 halftone cells measuring 16 x 16 per inch. That means an imagesetting addressability (resolution) of 2,400dpi (150 x 16 = 2,400).  F.Y.I., laser spots of this addressability are measured in microns. One micron equals .0003937 inch. One point equals 352.78 microns or .0139 in.

The last piece of the puzzle
Now that all the technical stuff is out of the way, you can go ahead and scan with the confidence that your image will contain enough numbers to be the size you want it, at the proper line screen (if your image to be printed), and containing the right amount of tones. There is, however, one last piece to the puzzle.

Exactly how much “resolution” is enough to scan at and/or how much is enough to create original art directly in your favorite painting or imaging program?

There is much debate on this issue, but its safe to say that input resolution (spi) should be at least 1.5x and no more than 2.5x the desired line frequency (lpi). A factor of 2x is the commonly used ratio of spi to lpi; the ratio most used since the advent of “desktop” publishing.

Thus, using the math learned so far, an original piece of art measuring 4 x 5 inches to be printed in color at 100% size at a 150 lpi should be scanned at 300ppi (1,200 x 1,500ppi) with the color depth set to “millions of colors” (24-bit). The math to figure it all out goes like this:

1,200 x 1,500 = 1,800,000 x 24 = 43,200,000 ÷ 8,192 = 5,273.43Kb ÷ 1,024 = 5.15Mb

Which is fine if you plan on only using that image one time at that size and at that line frequency. I find that most images almost always end up being repurposed for something else down the road, and would rather have too much resolution than not enough. So, if you have the disk space, or some archiving hardware, I recommend always scanning at the maximum optical resolution of your scanner.  These days its typically 1,200spi/ppi or higher. And of course, this produces quite a large image file because it contains a lot of numbers.


The same 4 x 5 inch original scanned at 1,200spi/ppi (a typical optical scanning resolution of many of todays scanners) weighs in at just under 83Mb (see screen grab below). But, it would look exactly the same onscreen. Why? Because scanning with all that extra resolution has no effect on the quality of the image at all — it just takes up more space on your hard drive. It doesn’t increase the quality of the image any more than pouring a 16-ounce glass of water into a five-gallon drum will make the water taste better. It would just take up a lot of room in your kitchen.

Photoshop Image Size dialog box -- Resolution test 2
But, since images for print require more resolution to look good than screen images, that 4 x 5 inch image just scanned at 1,200ppi now contains enough data to be output at 300ppi (proper resolution for 150lpi printing), at a size of roughly 14 x 16 inches. To repurpose this overweight image for use at the original size of 4 x 5 inches, however, a lot of data would need to be jettisoned. This is called resampling, in this case, downsampling, covered in depth in a previous post. Whenever an image is resampled, its integrity is severely compromised — if the original was sharp and detailed to begin with, after resampling it will now appear blurred when viewed onscreen at 100%. Sharpness, hence detail, needs to be restored.

That’s a story for another day.

Thanks again for listening.

P.



No comments:

Post a Comment