The Basics of Digital Imaging

The Basics of Digital Imaging Dr. Roger K. Moore ARPS Many people who are starting out on the road towards the photographic lightroom have little or no previous experience of computers and even less understanding of the fundamental principles of digital imaging. This note is intended to provide a simple grounding in the absolute basics in order to help those who have yet to embark upon the purchase of the necessary equipment as well as those who have already begun experimenting yet are still unsure about some of the key concepts involved. The main items covered are: the difference between a digital image and a conventional photograph, how a digital image is stored inside a computer, how file size is calculated, the difference between input and output resolution, how to achieve photographic quality results, and finally some (hopefully) useful rules of thumb. The Digital Image The key difference between a conventional photograph and a digital image is the way in which the colours and tones of a scene are recorded. In the case of a normal photograph, light reacting with a sensitised emulsion causes minute chemical changes, which form an invisible latent image. These changes are then amplified, using a chemical developing agent, to produce an effect visible to the human eye. In the case of a digital image, light reacting with an electronic sensor causes minute electrical changes, which are then amplified electronically to produce an effect visible to the human eye. Perhaps surprisingly, the same basic principles are being applied in both cases. The only real difference is that a conventional photograph records an analogue or continuous image, whereas a digital photograph records a digital or discrete image. This simply means that, whereas a conventional photograph is capable of recording a virtually infinite number of different hues and tones and can preserve very fine detail in a scene, a digital image is limited to a finite number of colours and tones and to a finite resolution of detail. However, this does not mean that a conventional photograph is inherently superior to a digital photograph quite the reverse. Analogue recording processes are prone to suffering from noise, thus a digital image with a sufficiently high resolution is quite capable of reproducing a scene which, when judged by the human eye, could have a higher fidelity than a conventional photograph. This is precisely what has happened in audio recording (with the emergence of the digital Compact Disc), and what is about to happen in digital radio broadcasting. The only issue is one of cost versus quality. 1

Bits and Bytes The basic concept for storing information inside a computer is to use electrical switches which are either on or off. This is what is meant by a binary representation, and it consists of just ones and zeros each of which is known as a binary digit or bit. This seems very different to the decimal zero to nine scheme that we use in everyday life. However it is actually very similar indeed. Using the decimal scheme, it is possible to represent any number using an appropriate sequence of decimal digits. For example, the number 1238 denotes one thousand, two hundreds, three tens and eight units that is, one-thousand two-hundred and thirty-eight. Notice that each position represent multiples of ten from the right-hand end ones, tens, hundreds and so on. 1 2 3 8 1000 100 10 1 Exactly the same scheme is used in the binary system except that each position represents multiples of two from the right ones, twos, fours, eights, sixteens and so on. So, using binary, it is also possible to represent any number using an appropriate sequence of bits. For example, the number 10110101 denotes one one-hundred-and-twenty-eight, no sixty-fours, one thirty-two, one sixteen, no eights, one four, no twos and one unit which, in decimal digits, is 181. 1 0 1 1 0 1 0 1 128 64 32 16 8 4 2 1 In everyday life we can use the decimal system to count up to any number we can think of. Even if the number is huge, we just use enough digits to describe it. For example, the number 3,562,727,637,268,049,484,393,836,383,974,674,839 is perfectly acceptable (although rather difficult to put into words!). Of course the same huge number could be expressed in binary digits but, since binary sequences are always longer than their decimal counterparts, the binary code would be very long indeed and so would take up a lot of space (i.e. memory) inside a computer. In practice, it is very unlikely that anyone would want a computer to be able to count up to such a high number. So, the maximum number of binary digits is normally fixed and this is referred to as a word. The latest personal computers typically use 32 bit words; older ones tended to use 16 bit words. In either case, each computer word is usually built up from multiples of 8 bits, and an 8 bit binary sequence is known as a byte. Therefore, the words in a 32 bit computer are each made up from four bytes, and the words in a 16 bit computer are each made up from two bytes. In one byte it is possible to count from 00000000 to 11111111 or, in decimal, from 0 to 255. That is, a single byte can hold one out of 256 different values. An 8 bit byte, therefore, is a fundamental unit in modern-day computers, and that is why the size of the main storage devices the random access memory (RAM) and the hard disc (HD) - is given in Mega (millions of) or Giga (thousands of millions of) bytes. So, a 32 Mbyte RAM can store 32x8 or 256,000,000 bits of information, and a 4 Gbyte HD can store 32,000,000,000 bits of information. In a typical computer application such as word processing, bytes are used to store characters. This means that a given font can contain up to 256 different characters. If you look in a computer (or printer) manual, you can often find a table showing, for each character in a font, the corresponding numerical value between 0 and 255. The values associated with the most common characters are standardised and these are known as the ASCII character set. For example, the character A is always represented by the number 65, B is 66, space is 32, a is 97, b is 98 and so on. This means that a text containing 1000 printed words with an average of five characters and one space per word would need at least 6000 bytes six kilobytes, 6 Kbytes of storage. 2

Pixels Whereas a given text will contain a certain number of characters, a given visual scene may contain virtually an infinite amount of information. This is because the amount of detail that could be extracted from a visual scene is potentially immense imagine zooming in from a landscape, to a field, to a tuft of grass, to individual leaves, to the pattern of veins on a single leaf and so on. A camera using conventional film is able to capture this vast amount of information but, or course, only up to a certain limit that limit being defined by the quality of the optical elements used to focus an image on the film and the grain structure in the film itself. In other words, it is physically impossible to record detail beyond some finite limit, where that limit is defined by the physical properties of the capturing mechanism. A scene is thus said to be sampled at a certain level of detail. The same is true of a digital camera. An image of a visual scene is sampled, not by film grain, but by the individual light sensitive elements of a charge coupled device or CCD. And whereas the resolution of conventional film is very high (because of it s fine grain structure), the resolution of a CCD array is currently rather crude (due to the technological limitations in constructing such small arrays). Also, whilst the grain pattern in a film is fairly random in structure, the sampling elements of a CCD array form a regular matrix of individual picture elements or pixels. For example, a common sampling resolution among the cheaper digital cameras is a CCD matrix measuring 640 by 480 pixels. This means that the resulting image is stored in 640x480 or 307,200 pixels. film-based image digital image A digital image thus consists of a matrix of individual pixels, where each pixel captures information from a small area of the scene being pictured the more pixels, the higher quality the digital image. As in conventional film, the information that needs to be captured is the colour and brightness at each point in the scene. Again the information in the scene itself is vast; an infinite variety of colour and brightness is possible, even at an infinitesimally small point. This means that, in principle, the information that could be captured at each pixel is huge. Luckily the human visual system is itself limited in what it can perceive. So, whilst a scene contains a potentially infinite number of levels of brightness, the human eye can only discriminate up to a certain limit. Also, although a scene contains a potentially infinite variety of colours, the human eye will accept an approximation using appropriate combinations of just three primary colours - red, green and blue - RGB. It turns out that 256 levels is quite enough to record the brightness (of any colour) with a reasonable degree of fidelity. So, this means that a monochrome image (i.e. just brightness) can be recorded using one byte of information per pixel. This is what is meant by the term 8-bit greyscale. On the other hand, an RGB colour image uses one byte per colour which, for three colours, means three bytes per pixel. This is what is meant by the term 24-bit colour. 3

File Size An image which is n pixels wide and m pixels high contains a total of nxm pixels. A greyscale image has one byte per pixel, so it therefore requires nxm bytes to store it. For example, a 640x480 pixel monochrome image contains 307,200 pixels which is a file size of 307,200 bytes or about 307 Kbytes A colour image has three bytes per pixel, so it therefore requires nxmx3 bytes to store it. For example, a 640x480 pixel colour image also contains 307,200 pixels but this requires a file size of 307,200x3 bytes which is 921,600 bytes or about 922 Kbytes (i.e. three times bigger than the monochrome version). Because the relationship between width and height in pixels and the total pixel area is multiplicative, doubling the width and height quadruples the file size. For example, doubling the 640x480 colour image to 1280x960 pixels gives a new total of 1,228,800 pixels and this requires a file size of 1,228,800x3 bytes which is 3,686,400 bytes or about 3.7 Mbytes. Before getting into the confusing area of input and output resolution (see below), it is very important to appreciate that the dimensions of an image in pixels are absolute - the more pixels, the better the quality. Also, since file size is directly related to the number of pixels (and whether the image is monochrome or colour), it too is absolute the bigger the file, the better the quality. The only issue is how large is large enough? Data Compression Images recorded in a digital camera are often compressed in order to reduce the amount of information (the file size) and therefore increase the number of pictures that can be stored in the onboard memory. There are two types of compression: lossless and lossy. The lossless methods avoid storing multiple pixels which happen to have exactly the same value. These methods do not affect the quality of the stored information. Lossy methods, on the other hand, compress an image by throwing away some of the fine detail in places where it might not be noticed. These methods do affect the quality of the stored information. The most common compression technique for digital images is JPEG which is lossy. This means that it is very important not to keep saving and re-opening an image in JPEG format; each time more information is lost, and the quality of the image will gradually deteriorate even if no other changes are made. An example of a lossless compression technique is TIFF-LZW. Input Resolution Whereas a digital camera delivers an image of a certain size according to the size of it s CCD array (e.g. 640x480 pixels), images which are input through a film or flatbed scanner are converted to pixels at a user-defined number of samples per inch (spi). This is the input resolution and it is more often quoted as pixels per inch (ppi) or dots per inch (dpi). In this context, spi, ppi and dpi mean exactly the same thing. The input scanning resolution relates the physical size of the input image to the number of pixels. For example, a 6 x4 print scanned at 100 ppi will produce a digital image consisting of 600x400 pixels, whereas the same size print scanned at 300 ppi will result in a higher quality image of 1800x1200 pixels. Similarly, a 35mm slide scanned at 100 ppi would produce an image of only 140x100 pixels, so it would need to be scanned at 1200 dpi to match the quality of the 6 x4 print scanned at 300 dpi. 4

The choice of input scanning resolution depends on the intended use of the resulting image. In particular, the scanning resolution is chosen according to the required output size and quality. The larger the required physical output, the higher the scanning resolution needs to be in order to maintain a given level of quality (i.e. to maintain a constant amount of information per unit area). On the other hand, increasing the scanning resolution whilst keeping the physical output size the same increases the quality of the output print (i.e. increases the amount of information per unit area). As stated earlier, the number of pixels in an image (and whether it is monochrome or colour) defines the file size. A higher scanning resolution captures more detail from the image and results in a larger file size. However, doubling the scanning resolution quadruples the number of pixels (and the file size), and this is why high input sampling resolutions can lead to very large file sizes. Clearly very large file sizes take up a lot of room on the computer and are much slower to work with. Therefore it is important to know what target physical size and quality is required on output, so that a sensible decision can be made about the required input scanning resolution (in order to keep file sizes as low as possible). For example, the input resolution required for an image to be reproduced on a web page would be very different to that required for an image which is to be reproduced as a 16 x12 exhibition-quality print. Output Resolution As indicated earlier, once stored inside a computer a digital image no longer has any inherent physical size. For example, a 600x400 image could have come from a 6 x4 print scanned at 100 ppi, or from a 35mm slide scanned at 400 ppi. This means that an image can be printed out (or displayed) at any size depending upon the output resolution which, confusingly, is also expressed in ppi or dpi. Of course, if an image were to be printed out at the same resolution that it was scanned, then the physical output size would be identical to the original input dimensions. For example, a 6 x4 print scanned at 100 ppi would give 600x400 pixels, and 600x400 pixels printed out at 100 ppi would produce a 6 x4 print. However, changing the output printing resolution would change the physical size of the output image. For example, 600x400 pixels printed at 200 ppi would produce a 3 x2 print, and the same image printed at 50 ppi would produce a 12 x8 print. So for a given image, output resolution is traded against output size the higher the resolution the smaller the output, the lower the resolution the larger the output. Hence an image which is printed out at a lower output resolution than it s input scanning resolution would be magnified, and an image which is printed out at a higher output resolution than it s input scanning resolution would be reduced. These principles are directly analogous to the process of enlargement in conventional photographic printing. Clearly to print a 35mm slide at a reasonable size requires a significant degree of enlargement. For example, the ratio between the dimensions of a piece of A4 paper and the dimensions of a 35mm slide is about eight-to-one. So, in order to print a 35mm slide at A4 size, the input scanning resolution for the slide would have to be 8 times greater than the output printing resolution. In other words, if the slide were to be scanned at 1000 ppi, then it would have to be printed at 125 ppi (i.e. 1000/8) to fill a piece of A4. Similarly, the ratio between the dimensions of a piece of A3 paper and a 35mm slide is about eleven-to-one. So, if it was desired to print a 35mm slide at A3 size using an output resolution of 300 ppi, then the slide would have to be scanned at 3300 ppi (i.e. 300x11). 5

Printer/Display Resolution The output resolutions described above are a (variable) property of an image and should not be confused with the printer/display resolution which, confusingly, is also expressed in dots per inch or dpi. The printer/display resolution is a property of the output device, and it might be fixed or controllable. The larger the resolution, the greater the detail that can be re-produced (and, for a printer, the slower the printing). A computer display screen typically has a display resolution of 72 dpi, and a photographic-quality ink jet printer typically has a printing resolution of 720 dpi 1. A printer with a dual specification such as 720x1440 dpi is capable of delivering twice as many dots across a piece of paper that it can down the paper. In an ideal world, each pixel in an image would be output using a single (multicolour) dot on a screen or printer (this is exactly what happens when an image is displayed at 100% on a computer screen i.e. one pixel is displayed per RGB screen dot). However, for a printer, a single dot is only capable of being present or absent; it is not able to represent 256 different levels of brightness. Printers overcome this limitation by using patterns of dots (halftoning) to simulate the different levels, so the effective resolution known as the line screen frequency - is much lower than the quoted dpi, e.g. a 720 dpi printer might have a corresponding line screen frequency of 180 lines per inch (lpi). The printer or display resolution therefore sets an upper limit for the output resolution i.e. the maximum quality that can be obtained from a given output device. If an output device is sent information at a higher resolution than it can reproduce, then two things can happen: first, the extra information is simply thrown away (it is wasted) and second, moire-type artifacts can be introduced. (This latter point may be quite important if it is intended to re-photograph a large image off of a screen. The solution is to re-size the image such that it can be displayed at 100% - see below.) Requirements for Photographic Quality Output For a printer, the best quality is usually achieved with about one-and-a-half to two image pixels for each output pixel. For example, the best quality for a 180 lpi printer would be achieved with an image output resolution of 360 ppi. At lower output resolutions, a printer would deliver lower quality results simply because less information is being printed. Half the output resolution would be one quarter of the information per unit area. Nevertheless, reducing the output resolution (e.g. to increase the size of the output print) may not produce a very noticeable drop in print quality. Some fine detail would be lost, but areas of gradually changing tone or colour would be reproduced almost as well. However, there comes a point when the output resolution becomes so low that it is possible to see the individual pixels (particularly along diagonal edges); this is known as pixelation. Pixelation usually becomes quite noticeable for output resolutions below 120 ppi. So, for a 720x1440 dpi printer it is advisable to aim for output resolutions between 360 and 120 ppi, preferably nearer the higher end for the highest quality. Another consideration is the level of detail that the human viewer can perceive at any given viewing distance. Clearly, a 6 x4 print would normally be viewed much closer than a 16 x12 print. So, surprisingly, higher quality output may be required for smaller prints. 1 Note that a printer with a specification of 720 dpi is capable of printing 720 dots per inch from each of its ink cartridges, so its resolution is not 720 divided by the number of inks (as is often assumed). 6

It turns out that the human eye can discriminate 5 cycles (i.e. 10 dots) per mm at a viewing distance of 250mm 2. This would suggest an optimum output resolution of 254 ppi. For closer viewing, a higher output resolution would be needed, and for viewing at a distance, a lower output resolution could be used. Unfortunately it s not easy to stop people peering closely at even an exhibitionsized print! All this means that, in order to produce an A4 print (i.e. 11 by 8 ) of the highest possible quality on a 720 dpi 180 lpi printer, it would be necessary to use using an output resolution of 360 ppi and an image size of 3960 (i.e. 11x360) pixels by 2880 (i.e.8x360) pixels. For a colour image, this would correspond to a file size of 34,214,400 (i.e. 3960x2880x3) bytes or about 34 Mbytes. For a monochrome image, this would correspond to a file size of 11,404,800 (i.e. 3960x2880) bytes or about 11 Mbytes. Also, to produce such an image from a 35mm original would require an input scanning resolution of around 2880 ppi. Unfortunately a 34 Mbyte file is pretty large and might be difficult to work with. Also many low cost 35mm scanners do not reach 2880 ppi input resolution. So, at the other end of the scale, the absolute bare minimum requirements to produce an acceptable A4 colour print would be to use an output resolution of 120 ppi. This would mean an image of 1320 (i.e. 11x120) pixels by 960 (i.e. 8x120) pixels and a corresponding file size of 3,801,600 (i.e. 1320x960x3) bytes or about 4 Mbytes. For a 35mm original, this would require an input scanning resolution of 960 ppi, and this is well within reach of many lowend film scanners. The same arguments applied to an A3 print lead to an absolute bare minimum image size of 2280 pixels by 1320 pixels and an input scanning resolution of 1320 ppi for a 35mm original. This corresponds to a file size of about 8 Mbytes for a colour image and 2.5 Mbytes for monochrome. Note that the highest resolution available from a Kodak Photo-CD - so called 16 base - is 3072 pixels by 2048 pixels. For A3 this would give an output resolution of 186 ppi which is middle quality. For A4 it would give an output resolution of 256 ppi which is towards the high quality end. The highest resolution available on the Nikon Coolscan range of scanners is 2700 ppi which is getting close to the maximum quality possible for producing an A4 print from a 720 dpi printer (and towards the high quality end for A3). Although there is considerable discussion about the file sizes needed for the highest quality output, it is also worth bearing in mind that high resolution is only really important for the finest detail in an image. So, if an image consists mainly of smooth continuous tones, and any fine texture details are not vital to the overall effect, then there is no need to be concerned about achieving the highest possible output resolutions. It is, however, still important to steer clear of excessively low output resolutions, since pixelation will become obvious even in a continuously toned image. 2 According to Patrick Forsyth in the Royal Photographic Society DIGIT Magazine, Issue 5, January 1998. 7

Re-Sizing an Image Changing the output resolution of an image in order to produce a larger or a smaller output is often termed re-sizing. However, it is very important to appreciate that this simply involves the association of a different output resolution with an image, the actual image size in pixels (and thus the file size) does not change at all. Re-sizing which does involve changing the file size is known as re-sampling (or interpolation ), and this actually changes the information in the image. Reducing the image size by re-sampling (or down-sampling ) reduces the amount of information stored and thus reduces the quality. The only times you would be likely to do this would be: when constructing a montage and one component is too large 3, for reducing the file size in order to fit it onto a floppy disc, for producing an image to be displayed at 100% on a screen, or for producing a lower resolution image for a web page. Increasing the image size by re-sampling (or up-sampling ) requires new information to be guessed from the information which already exists in the image and thus also reduces the quality. The only times you would be likely to do this would be: when constructing a montage and one component is too small, or if the image needs to be printed larger than you anticipated and you want to avoid the appearance of pixelation by lowering the output resolution too much. Up-sampling leads to a greater reduction in quality than down-sampling (because throwing information away is less of a problem than inventing information). This is why scanning software uses a higher sampling resolution than is needed when an image of an intermediate size is requested (in these circumstances the scanning software down-samples the image before sending it into the computer). It is clearly important to avoid unnecessary re-sampling of an image, and thus care must be taken to avoid doing it if one only wishes to re-size for producing a different sized output. For example, in Photoshop s Image, Image Size dialogue box, the re-sample option should normally be left unchecked. 3 Note that the transform tool is equivalent to re-sizing, so it is important not to use it repeatedly as information is lost each time. 8

Some Useful Rules of Thumb Some of the key recommendations arising from the information contained in these notes are summarised below: Don t store an image on which you are constantly working in JPEG format. Use the native file type for your imaging software or a lossless format such as TIFF. Try to perform all modifications to an image in Photoshop (or whatever imaging software you use) rather than in scanner or printer driver software. Think of image size in terms of pixels (because this gives an absolute indication of the relative sizes of different images monochrome or colour). For example, if you use rulers, set them to indicate pixels. Avoid using the interpolation option in your scanner software. If you need to increase or decrease the image file size, use Photoshop it is likely to be much more accurate. Whatever size you are printing, always make sure that the output resolution does not fall below 120 dpi (otherwise you are likely to see pixelation in the printed image). To produce reasonable quality results from a 35mm original use an input scanning resolution no lower than 960 ppi if printing to A4 and no lower than 1320 ppi if printing to A3. Remember that a small print will be viewed at a closer distance, so it may be inadvisable to use a lower input scanning resolution. In fact, scanning a 35mm original at 2700 dpi gives excellent quality at all output sizes when taking this into account. Never re-size an image with the re-sample box checked unless you have a really good reason. Don t apply the transform tool repeatedly it is equivalent to re-sampling and quality will deteriorate each time it is used. For photographic quality prints, always set the printer to the maximum number of dots per inch regardless of the paper type being used. Don t worry too much about obtaining the highest output resolution if your image mainly consists of continuous tones with no important fine detail. Don t worry too much about anything - have fun! For further information, you can email me at rkmoore@compuserve.com Also, to see some more examples of my digital images, you might like to visit my web page at http://ourworld.compuserve.com/homepages/rkmoore 9