The management of digital images through an imaging chain is a complex operation. The huge range of options available in terms of hardware, software, file formats and image-processing techniques mean that there are many decisions to be made during the passage of the image from acquisition to output. These decisions affect the way the image is processed, which will ultimately define the number and visibility of artefacts produced, and therefore the overall image quality.
A generic digital imaging chain is illustrated in Figure 25.1. Input and output devices vary widely in technology and characteristics; therefore, the image must be manipulated in many ways as it passes through the chain. A hybrid imaging chain combines silver-halide-based processes with digital processes, for example scanning a photographic original which is then processed and output digitally.
Consideration of image workflow is an important aspect in the management of a digital imaging project. The workflow defines the sequence of operations to be performed and the order in which they occur. The main aim of a good workflow is to provide a framework which simplifies the decision-making process, allowing the user to work efficiently while producing optimum results for the particular imaging application.
The definition of a suitable workflow for a photographic project is dependent on various factors. Ideally, particularly for a large-scale digital project, workflow will be reflected upon prior to the purchase of equipment. More commonly, however, the workflow needs to be optimized within the constraints of the available equipment, software and facilities. Obtaining and maintaining optimum image quality should be a primary concern, but must be balanced against the need for efficient processing, storage and cost. The requirements of the imaging application will ultimately determine the relative importance of these issues. If the image is to be processed for a specific output, then the workflow may be simplified. However, the future usage of the image is often more difficult to define and may be limited by this approach; this should also be a consideration in terms of decisions about both quality at capture and image archiving.
If the digital workflow is to adequately satisfy the needs of the application, then a thorough approach must be used, with thought given to the hardware settings to be used and the various image processes that may be applied in software. Technical aspects to be considered should include:
• the image resolution throughout the imaging chain;
• bit depth of the image;
• the necessity of interpolation (in resampling processes) and at what stage this should be performed;
• colour management and colour spaces to be used;
• where in the chain image adjustments or enhancements should be performed;
• whether automatic hardware settings may be an appropriate approach to speed up workflow;
• whether automatic hardware settings may be an appropriate approach to speed up workflow;
• file formats to be used and whether lossy compression is possible or desirable.
This chapter aims to look at the implications of some of these issues. It is not meant to be a prescription for an ideal imaging workflow, rather a discussion serving to provide some background to the subject, enabling the user to consider technical aspects of the imaging chain in a coherent way when approaching their own digital imaging projects.
The widely varying requirements of different imaging applications mean that it is impossible to determine a generic workflow suitable in all cases. It is possible, however, to establish a few general guidelines which may be applicable to multiple applications, as follows:
1. Image quality at capture. An overriding concern is image quality and key to this is optimal image quality at capture. One of the aims in the workflow should be to capture the image at the highest quality possible, within the limitations of the equipment and taking into account requirements for speed of processing.
2. Image quality through the imaging chain. Further to this, the workflow should endeavour to maintain image quality throughout the imaging chain, by minimizing the errors in the image introduced by various processes as it moves through the chain.
3. Workflow approach based on image output. Both the preceding guidelines must be balanced against the need for the image to be ‘suitable for application’. This means that in some cases speed of processing and small output file size may be more important than maximum image quality.
4. An efficient approach. The workflow should be efficient, meaning that at all stages in the chain the aim should be for a simplified approach if possible, with unnecessary processes being eliminated. The workflow must be flexible enough to allow for extra processing of individual images as required. Having a simple sequence of steps which will work for the majority of images will help to increase speed and minimize unnecessary quality loss.
5. Repeatability. An efficient approach will be repeatable, meaning that image processing is applied in an ordered and deliberate fashion where possible, rather than using an ad hoc approach which is much more difficult to replicate. It can also help to speed up image editing, as the processing steps become second nature to the user, and may facilitate automated (batch) processing.
6. Interpolation. Unnecessary interpolation should be avoided. If resampling processes are to be used, they should be kept to a minimum. Consideration should be given to the interpolation methods used and the point in the workflow where they will be applied.
7. File formats. The use of standard file formats ensures that results are predictable across different platforms, applications and imaging chains. File formats should be selected carefully in terms of their properties and suitability at a particular stage in the imaging chain (see Chapter 17 for more information on file formats). Thought should be given to the issue of image compression; lossy compression should only be applied if necessary (see Chapter 29).
8. Colour management. The aim of colour management is to ensure that tone and colour reproduction are controlled and produce predictable (and desired) results. This involves decisions about the colour space encodings to be used throughout the imaging chain and how and where colour conversions should occur. Unnecessary conversions between colour spaces should be avoided to minimize artefacts such as colour banding (a form of posterization, or colour contouring – see Chapter 19). In general, an ‘open-loop’ approach to colour management is desirable, allowing flexibility if further devices are added to the imaging chain. For the majority of projects, International Color Consortium (ICC; see Chapter 26) colour management provides an elegant solution and may be implemented with a greater or lesser degree of user input. It should be noted that an open-loop approach is not always required, for example in research applications, but these are a special case, in which colour is controlled by characterization and construction of transforms specific to the devices involved. In such cases, image workflow is defined entirely by and will be specific to the requirements of the project.
9. Archiving data. Image archiving requires decisions to be made in terms of suitable file formats, storage media and the way in which archived images are organized, managed and the image data will be migrated in the future. This subject is dealt with in detail in Chapter 18. Ideally, image data should be archived at several stages, before and after optimization.
10. Data organization and retrieval. The workflow should include a strategy for the cataloguing and retrieval of images, which requires a labelling system and keywords. Image database software and imaging workflow applications can assist in this process.
It is important to consider image output and image quality requirements prior to workflow design. If the output is known and speed and efficiency are paramount, then the best approach may be to capture images specifically for that output, i.e. of sufficient resolution but no more. This may be termed a production for output workflow where the stages in the imaging chain are considered in terms of the requirements at output. The advantages of this approach are simplicity and speed. With size and required quality at output specified, the preceding steps are easy to define. Colour spaces that are optimum for mapping to the gamut of the output device may be selected during or after capture. It may be feasible to apply some image adjustments using batch processing. This type of workflow is suitable when dealing with high volumes of images in which the images need to be processed and delivered quickly, which are typical requirements of photojournalism and sports photography.
Producing images for a particular output may limit alternative usage, particularly if the images are of limited resolution, or if lossy compression has been applied. To maintain the widest possible range of options for image output in the future, a production for optimum quality approach may be used. In this case, images are captured at the optimum resolution of the capture device and remain at that resolution without resizing. Colour encodings are selected so that colour gamuts large enough to encompass the gamuts of a range of output devices and master copies are archived in an uncompressed file format. This approach is less efficient in terms of the necessary levels of image processing and storage. The same workflow requirements are typical of imaging applications dealing with low volumes of very-high-quality images, for example in high-end advertising, museum archives or editorial photography.
Resolution is an image attribute that has a significant effect on image quality, both in terms of the ability of the system to adequately represent scene content and also upon the subjective quality of the image as perceived by the viewer (see Chapters 19 and 24). Changes in pixel count in digital systems are predominantly achieved by interpolation (described below), which may introduce digital artefacts into the image. The impact of the image’s pixel resolution and any changes in it must therefore be considered as a fundamental part of the image workflow.
The term resolution when discussed in relation to digital imaging systems can be confusing. As described in Chapters 19 and 24, the ‘true’ definition of resolution is as a measure of the detail-recording capability of a system. In digital imaging the term is applied in different ways. It is often used by camera manufacturers to describe the pixel count of a sensor. This is more correctly termed the pixel resolution and may be defined as the number of pixels horizontally and vertically, or as a combined figure in terms of megapixels, which describes the total number of pixels in millions of pixels. The spatial resolution of image sensors may alternatively be defined as described in Chapter 19 in terms of the number of available pixels per unit distance, commonly pixels per inch (ppi).
Pixel resolution is often quoted as a figure of merit when comparing different sensors. However, the actual detail-recording ability of the system, although related to this, is not defined by it. In comparing two sensors, the sensor with a larger number of pixels will not automatically produce better image quality as many factors, including the sensor dimensions, will influence the size and shape of the point spread function.
The ability of the imaging chain to represent detail is not only dependent upon the capture resolution, but is also influenced by the resolution of any output device in the chain. As the image moves through the imaging chain, the number of pixels will remain the same unless cropping or resampling processes take place. However, these pixel numbers only become meaningful when considered in terms of device resolution, particularly that of the output device. Most devices in the chain will have a native pixel resolution, which may be defined as the maximum number of pixels that may be captured or reproduced without interpolation and is an inherent property of the system; this will frequently be described in pixels per inch (ppi). For output devices this will define the total number of pixels, and the maximum number of pixels per inch that can be reproduced without interpolation. Again, the effective spatial resolution of the output device will be defined by the size and shape of the point spread function (PSF) and various other factors (see Chapters 19 and 24). It should also be noted that a pixel displayed on a monitor may be represented by various different shapes and configurations of RGB triads (see Chapter 15). Furthermore, a printed pixel may actually consist of various configurations of groups of printed dots rather than a single dot if digital half-toning is being used to represent continuous tone (see Chapter 16).
If the image has been captured at the native pixel resolution of the sensor and is to be displayed or printed at the native resolution of the output device, then each pixel from the input device will correspond exactly to a pixel in the output device. In this situation the spatial dimensions of the output image will be fixed by the number of pixels in the image. In many cases, however, the image will need resizing either up or down to achieve the required dimensions. The resizing of an image is a resampling process (sampling intervals are redefined), i.e. because there is no longer a 1:1 ratio between pixels at input and output, pixel values will be interpolated.
Because resampling to a lower resolution can produce aliasing of high frequencies within the image, better results may be produced by preceding the interpolation with a low-pass filtering stage, to remove the high frequencies. This process may be built in to specialized down-sampling algorithms and software.
Interpolation in the form of resampling may occur at multiple stages in the imaging chain. If the image is captured either by a scanner or a digital camera at a resolution other than the sensor’s native resolution, then an interpolation algorithm is applied in the digital signal processor of the capturing device and, in the case of a digital camera, may be implemented prior to, in conjunction with or after colour demosaicing. Similarly, interpolation may be applied by the RAW conversion software (see Chapter 17). The default resolution identified for the image when it is opened in the RAW application interface is the actual number of pixels captured by the sensor and so corresponds to the native input resolution; anything else will be achieved by interpolation. Alternatively, resampling may be applied in an image-processing application. At output, images at resolutions other than the native resolution of the output device (if a printer) will usually be interpolated to that resolution by the device’s software; hence, more predictable and theoretically optimum results will be achieved by sending the image to that device at that resolution to start with.
The resizing of an image is not the only operation involving interpolation. Spatial or geometric transformations of the image plane (such as rotation, translation, correction of geometric distortions and ‘free transforms’) involve the repositioning of all or some of the pixels within the image. These operations often result in pixels being mapped to positions between pixel coordinates (Figure 25.2), in which case the pixel values must be interpolated. They therefore represent a further potential compromise of image quality.
The process of interpolation involves the estimation of new values from known values. In general, interpolation is an averaging process and therefore cannot provide an increase in the resolution of fine detail, i.e. new pixel values are generated, but new levels of detail are not resolved. Nonadaptive interpolation algorithms use the same interpolation method across the entire image plane. There are various approaches, which differ in terms of the size of the neighbourhood of pixels used in the calculation and the method of estimation used. The choice of a particular method is a balance between speed of processing, accuracy and resulting artefacts. More details on the bilinear algorithm mentioned below are given in Chapter 23.
The simplest approach to estimate new pixel values is nearest neighbour interpolation, which simply replicates image pixels, giving the new pixel the same value as its nearest neighbour (Figure 25.3a). This is the fastest method, but produces the greatest loss in quality, due to the artefacts incurred, and is used mainly for images containing illustrations with hard edges. Because pixel values are not averaged, but are copied, jagged artefacts will become particularly apparent on diagonal edges (see Figure 25.4b).
Bilinear interpolation estimates missing values by combining the values of the four closest pixels, fitting linear functions between adjacent points and then calculating the average value from these functions (Figure 25.3b). The averaging process results in a slight blurring of the image (see Figure 25.4c). Bilinear interpolation is slower than nearest neighbour interpolation, but will produce fewer artefacts in continuous-tone images.
Bicubic interpolation is even more complex, using the values of 16 pixels and fitting a cubic surface function between them to provide an estimate of the new pixel value. The larger number of pixels used in the calculation and the complexity of the function mean that the estimation is more accurate compared to the previous two methods. This provides a significant improvement in image quality, with smoother results on diagonal edges but less overall image blurring (see Figure 25.4d).
Adaptive interpolation algorithms change their approach pixel by pixel for different types of image content to minimize the visibility of artefacts (see Chapter 14 for an example of an adaptive algorithm employed in digital cameras). Such algorithms may be implemented behind the scenes by a hardware signal processor to produce optimal results for that particular device. They may also be applied in proprietary software developed to manage specific imaging processes, such as image enlargement or printing. Many adaptive methods will apply a different version of the algorithm at edges, as these are the image areas where interpolation artefacts are most visible. There are a number of adaptive methods now available in more recent versions of image-processing software which are optimized for specific resampling operations.
Image ‘Anzac Bridge at Night’ © iStockphoto/TSKB/Tim Barrett
Figure 25.4 illustrates the artefacts that may be introduced by interpolation algorithms used when resizing an image. Therefore, if these are to be minimized, interpolation should only be applied when absolutely necessary. As already described, the point at which interpolation is applied will be dependent upon the workflow approach.
For example, if the images are only to be used for a web page, then the maximum or native resolution of the sensor may be unnecessary. In this case, it may be perfectly adequate to capture images at a lower resolution, or to down-sample quite early on in the imaging chain. If the images need resizing, optimum results may be achieved using the adaptive interpolation algorithms automatically applied by the capture device. However, it may be that the results are better for the particular output using one of the algorithms available in proprietary RAW conversion software or image-processing software. It is important to note that the optimal approach can only be established by comparing images side by side using both methods.
Clearly, if the aim of the workflow is to provide optimum quality and the flexibility to output images to different media, then the image should not be resized until much later in the imaging chain. In particular, if the image is likely to undergo a significant amount of editing, then higher quality will be maintained by performing image adjustments on an image of higher resolution and then downsizing. The most common approach is therefore to resize the image at the end of the image-processing pipeline when preparing it for output. The image may then be sharpened if necessary to counteract the smoothing effect of the interpolation.
When images are to be prepared for a specific output, it is important to calculate the required pixel resolution based upon the resolution of the output device and the required physical dimensions of the output image. Such a calculation is needed to ensure that captured images have a high enough pixel resolution for the output requirements and to establish whether interpolation will be necessary. As described in Chapter 14, scanners usually have flexibility in terms of the input resolution that may be set by the user. Equation 14.1 describes the relationship between input and output resolution in terms of the dimensions of the scanned original and the output image. When capturing images using a digital camera, the choice of input resolution will be limited and is unlikely to exactly match that of the output device (Table 25.1).
In general, if images are to be prepared for display on a monitor, they may be specified in terms of the number of pixels required. The required output image size can be determined by establishing how large the image is to be displayed on screen and basing the calculation on one of the screen dimensions given in Table 15.1. If the image is to be placed on a web page, the screen dimensions should be selected carefully to represent an ‘average’ monitor. Monitors of higher resolutions will simply display the image at a size covering a smaller proportion of the screen.
SYSTEM |
PIXEL DIMENSIONS (H × V) |
XGA monitor resolution |
1024 × 768 |
Required resolution of an image to be displayed at three-quarters of screen size |
768 × 576 |
Minimum capture resolution for a typical digital compact camera |
640 × 480 |
Minimum capture resolution for a typical professional digital SLR |
1936 × 1288 |
Images to be prepared for print must be considered in terms of the required output print size. Printer resolutions are variable. As already described, the highest quality will be achieved if the image is sent to the printer at its native resolution, maintaining a one-to-one correspondence between input and output pixels. However, not all printers have a native resolution. In some modern inkjet printers in particular, it is possible to vary both numbers and sizes of printed dots, producing a range of printed resolutions for different media. For ‘photographic quality’ images, inkjet printers commonly work with resolutions from 180 to 360 ppi. It is important to note that technical specifications for inkjet printer resolutions commonly refer to the maximum number of dots per inch. These values can be slightly misleading, as they do not correspond to printed pixels per inch, which are produced from groups of dots of different colours. The correct value for maximum ppi should therefore be established from the manufacturer. Continuous-tone printers may vary between 240 and 400 ppi (see Chapter 16 for further information on typical resolutions of different printing technologies). The standard resolution quoted for offset printing is 300 ppi, but the required resolution may be lower than this. Required pixel dimensions for the images can be calculated by multiplying the required resolution in ppi by the image dimensions in inches. For example, for an image printed at 300 dpi and 10 × 8 inches, 3000 × 2400 pixels are necessary.
As described in Chapter 9, analogue-to-digital conversion applied to the signal output from the image sensor may result in much higher bit depths than those in the captured image. This is prior to the many other image-processing stages that happen in the capturing device and allows a reduction in the rounding errors produced as a result of working with integer mathematics (see also Chapter 21). The image, once saved, will be in most cases either 8 or 16 bits per channel. The choice of bit depth is a balance between requirements in convenience of processing, file sizes and image quality.
The convention of working with 8-bit images is based on the need to provide enough individual levels to visually represent images as continuous tone. However, as images are processed through the imaging chain, particularly as tonal corrections are applied in either levels or curves adjustments (see Chapters 21 and 27), or as a gamma correction, many intensity levels (grey levels) may be quantized to the same integer value, resulting in a tonal range in which some grey levels are missing (termed posterization). Bit-depth requirements for gamma correction are discussed in Chapter 21 and the effects on the image output levels illustrated in Table 21.3. Figure 25.5 shows the effects of posterization on an 8-bit image and its histogram (note that this image has been very heavily processed to emphasize both effects).
Working with 16-bit images means that there are many more levels available (65,536 instead of 256) per channel, reducing the effects of posterization, as illustrated in Figure 25.6. Thus, 16-bit quantization helps to maintain image quality but produces files twice the size of the equivalent 8-bit image. Sixteen-bit files are not supported by all file formats (currently, they are supported by RAW, TIFF, PSD, PNG and JPEG 2000 formats). Until recently, the level of 16-bit support in image-processing applications was limited, meaning that certain processes available for 8-bit images could not be performed on 16-bit images, although in current software versions this is less of an issue.
At the time of writing, a limited number of printers support the printing of 16-bit images with the aim of providing an extended printer gamut. This is useful when working with RAW files captured at 16 bits, using a large-gamut colour space such as ProPhoto RGB (see Chapter 23) and maintaining 16-bit workflow throughout. Whether the improvement in output gamut compared to the gamuts of 8-bit printers using eight or more inks is significant enough for 16-bit printing capability to become the standard is as yet unclear.
The Universal Photographic Digital Imaging Guidelines (UPDIG), produced by the UPDIG coalition (see Bibliography), describe one of the aims of digital workflow as the preservation of image appearance as the image moves through and between imaging chains. This requires colour management of the imaging chain, with information necessary for the transformation of image colours between colour spaces to be included with, or embedded in, the image file. Chapter 23 describes digital colour spaces and provides information on the characterization of digital imaging devices. The implementation of these processes in ICC colour management architecture is the subject of Chapter 26, where colour management workflow is covered in more detail. Thus, here the subject is introduced in the context of more general imaging workflow.
In ICC colour management, the processes of calibration and characterization result in device profiles. These are data files containing information about the tone and colour reproduction of each device. Profiles are used by the colour management system (CMS) for the colour conversions between devices. They may be sent as a separate file with the image file or embedded in it. If an image with an embedded profile is transmitted between imaging chains, the colour appearance of the image should be preserved. This is assuming that the profiles are accurate, that the conditions to which the device was calibrated prior to profiling remain the same and that the destination imaging chain is also colour managed using ICC colour management. Ideally, all devices in the chain should be profiled.
Camera profiles are only useful if applied to image data in a scene-referred image state (see Chapter 23). The majority of cameras do not, however, render scene-referred image data, but instead use one of two alternative colour rendering options. Intermediate reproduction description rendering encodes the image directly into a standard output-referred RGB colour space such as sRGB or Adobe RGB 98 and is the most widely used approach. The standard colour space is chosen by the user in the camera settings. Alternatively, manually deferred colour rendering may be used, in which the data are output as a RAW file. In this case the image after RAW conversion is also (usually) encoded in a standard output-referred RGB colour space, again selected by the user. Therefore, the profile used with an image captured using a digital camera will usually be the colour space profile for the colour space in which the image is encoded. This subject is covered in more detail in Chapters 14 and 26.
The display profile is extremely important, as the display is used for image editing and soft-proofing of images. The display profile should ideally be a custom profile produced by taking measurements from the surface of the display. Ambient viewing conditions must be carefully controlled by all image users to ensure that the display profile remains accurate. Scanners and printers are usually supplied with generic profiles from the manufacturer, but better results for both devices will be obtained from a custom profile. Generic scanner profiles do not generally compensate for the range of different settings that may be applied, although scanned images may be saved to a standard colour space to avoid the need for a custom profile. Custom profiles are also optimal for printers to ensure accurate soft-proofing and output colour rendition.
Unless an input profile is being used, a typical aim in setting the colour space at acquisition is to select the standard RGB colour space encoding to most closely match the gamut for a particular output. If the output is unknown, then it is advisable to select a colour space which is large enough to encompass most or all of the gamuts of possible output devices. There are a range of standard RGB space encodings, each covering different gamuts, which most closely match different ranges of input and output devices; no one colour space will be ideal for all imaging situations. There are four currently specified in generic RAW conversion software, the details of which are covered in Chapters 23 and 26; therefore, they are described only briefly here.
sRGB (Standard RGB colour space) was originally designed as a default colour space for multimedia images; therefore, its gamut is optimized to most closely match reference cathode ray tube (CRT) display gamuts. sRGB has a relatively narrow gamut and it is also very mismatched with CMYK spaces, being unable to reproduce a large part of the cyan region which is available to printers (see Figure 25.7). These two factors make sRGB less suitable than some of the colour spaces described below, if images are to be output to print. However, if the workflow output is known to be for web or multimedia display, sRGB is the optimum choice; it is also commonly specified as the colour space to be embedded in images to be submitted to professional digital colour labs or consumer digital print vendors for the production of display prints. If images to be output in sRGB do not require colour or tonal correction, then the simplest approach is to capture into sRGB. However, in many cases better results may be obtained by capturing into a colour space with a wider gamut, performing image editing and then converting to sRGB.
Colormatch RGB was introduced by Radius™ for its Pressview range of professional graphics monitors. Its advantage was a tonal response that closely matched press output; therefore, it was used particularly for high-end pre-press work. It has a larger gamut than sRGB, although still clipping some cyan areas when compared to CMYK colour spaces, but is relatively small compared to the majority of other gamuts now available. Adobe RGB 98 has mainly superseded it as the RGB colour space of choice for printed output.
© LOGO, a Gretag Macbeth company
Adobe RGB 98 has a wider gamut than sRGB and Colormatch RGB. It encompasses most of the CMYK gamuts used in offset printing; therefore, it was initially used as a default working colour space for pre-press work. Currently, it is commonly used in two situations: (1) if future image usage is undefined – its wide gamut is most likely to encompass the gamuts of potential devices, both display and hard copy; images should be captured into and maintained in Adobe RGB 98; (2) when the image output is known to be via inkjet or dye-sublimation printing, therefore a wide gamut is also required. In these printers, the conversion from RGB to CMYK is performed internally by the device driver (rather than prior to the image being sent to print) and they require an RGB file as an input; therefore, the RGB colour space needs to match the CMYK space as closely as possible, and Adobe RGB 98 is a good choice.
Although sRGB and Adobe RGB 98 can fairly adequately match the gamuts of the majority of output devices currently available, neither colour space comes close to using the full extent of the colour space of the majority of image sensors. Capturing into these colour spaces fundamentally limits the range of saturated colours that can be reproduced by the sensor. ProPhoto RGB (developed by Kodak ProPhoto RGB as a version of ROMM RGB; see Chapter 23) is a colour space with an extremely wide gamut; indeed, part of its gamut falls outside the gamut of the human visual system, illustrated in the CIE x, y chromaticity diagram in Figure 25.8.
The extent of ProPhoto RGB means that it is a much closer match to the gamuts of the majority of image sensors. Acquiring images into ProPhoto RGB therefore allows more of the gamut of the image sensor to be utilized. However, this approach should be used with caution; some of the colours available in ProPhoto RGB cannot be displayed on a monitor, meaning that there is no way of checking them during editing. The size of the gamut means that ProPhoto is only really practical when using a 16-bit workflow from input to output; if the gamut is represented using only the 256 levels per channel available in an 8-bit image, then posterization will be severe, as a result of the large colour differences between the different levels. In a 16-bit workflow, however, the wide gamut allows space for the extra levels that are available. With advances in both display and printing technologies, gamut sizes will increase in the future and images captured currently in a wide gamut may be output in the future to take advantage of this progression.
Image files for offset printing are finally output as CMYK files, the particular colour space being dependent on the press and paper type. The process of converting to CMYK is not a simple case of changing image colour mode in an image-processing application but requires, as for all conversions, source and destination profiles and rendering intents. Often, RGB files with embedded profiles will be supplied to the printer, who will perform the conversion. Alternatively, the photographer performs the conversion using a profile supplied by the printer. In this case, the point at which the RGB-to-CMYK conversion is carried out must be considered in terms of workflow. In the early days of colour management, it was common to convert to the relevant CMYK colour space as soon as possible after capture: an ‘early-binding’ workflow. This approach is simple and means that the way the image is processed is constrained within the output gamut from the very beginning. However, it is now much more common to operate using an RGB workflow until the image is to be output, described as a ‘late-binding’ workflow, therefore only limiting the gamut at the end of the workflow. Adobe RGB 98 is a good option in this case, for the reasons described above. Using an RGB workflow results in smaller file sizes through the earlier stages of the imaging chain and more available image-processing operations, and facilitates the potential for alternative output gamuts in the future. The implications of early or late binding are also briefly discussed in Chapter 26.
An effective workflow relies on the use of file formats which are standardized, meaning that they can be recognized across imaging chains and platforms. The file formats that are now standards or de facto standards for imaging applications have various properties, meaning that they are most useful for different parts of image workflow. Important features are: (1) the use of lossless or lossy compression (see Chapter 29); (2) support for layers, paths and other image editing functionality; and (3) storage and processing requirements of the workflow. Often, different formats will be used at different stages in the workflow. Chapter 17 covers the properties of the more commonly used formats in detail. File formats for archiving are also discussed in Chapter 18.
The still image file formats available in digital cameras are often limited to JPEG, RAW and in a few cases TIFF and/or JPEG 2000 files. Usually JPEG captured files will have a range of options for quality settings and sometimes an additional range of resolution settings, whereas RAW and TIFF files will be captured at the resolution of the image sensor. The range of formats available for scanned images is much greater.
The workflow approach is the biggest factor influencing the selection of formats. In a ‘production for quality’ workflow, optimum quality will be achieved by capturing in a RAW format, which also has advantages for image archiving. In a ‘production for output’ workflow, particularly if the fast turnaround of a high volume of images is a requirement, then the extra processing involved in dealing with a RAW file (and the extra storage required on a memory card) may make it a less appropriate choice.
The lossy compression employed by JPEG will create visible artefacts at high compression rates, although the effects of this on perceived image quality will depend upon the imaging application, resolution of the image and the scene content (see Chapter 19 on image quality and Chapter 29 for examples of lossy compression artefacts). If the application requires high quality but a fast turnaround, then capturing to uncompressed TIFF files will allow the acquisition of images without compression artefacts, with an inevitable trade-off in terms of file size and versatility. TIFF files are larger than their RAW equivalents, without the level of control that RAW conversion affords the user.
For Internet publishing, JPEG files are generally the preferred option for continuous-tone images. Whether these have been captured as JPEG files or converted to JPEG for output will depend upon the type of work being done. JPEG files are often described as perceptually lossless at low compression rates, i.e. when captured at the high-quality JPEG setting in digital cameras (this equates to Q8 or above in the quality scale of 1 – 10 sometimes used) – see Figure 29.14. Although TIFF files are generally preferred as output files for printing, high-quality JPEGs may also be suitable for printed images for some applications, as long as they are at a high enough resolution.
Imaging applications such as sports photography and much of photojournalism commonly rely on JPEG capture. In these cases, large volumes of images are likely to be captured at one time; therefore, storage is limited to that available on the camera. Additionally the images need to be transmitted to news desks as quickly as possible, so image processing needs to be minimal and file sizes are limited by available bandwidths for transmission. In such cases, much of the image processing, such as tonal correction, white balance and sharpening, may be applied in the camera using auto or custom settings to generate images that are good enough for the required output.
Capturing RAW files represents a significant departure in imaging workflow. As described earlier in the chapter, RAW capture defers colour rendering and many other image processes to a later stage in the imaging chain, separate to acquisition. The properties of RAW file formats are described in detail in Chapter 17, and RAW image-processing workflow is discussed in Chapter 14, so are only summarized here. A RAW file is not a ‘finalized’ file in the same way as JPEG or TIFF files. RAW files consist of almost unprocessed data directly from the image sensor, prior to colour interpolation (‘demosaicing’). This bypasses much of the processing that is automatically applied by the acquisition device when saving to other formats, such as white balance, gamma/tonal correction, colour space and even resolution settings. RAW workflow contains an extra image-processing stage immediately after capture, that of RAW conversion, which is carried out using either a plug-in such as ‘Camera RAW’ (see Figure 25.9) in Adobe Photo-shop or an image management application such as Adobe Lightroom, or using proprietary RAW conversion software specific to the RAW format (as described in Chapter 17, RAW is a generic term for a group of proprietary formats based upon the same principle of manually deferred colour rendering; as yet there is no standard).
The acquisition stage in RAW workflow is therefore much simpler than when using other file formats. In general, the only settings in the camera that will influence the raw data captured are the ISO speed, determining the effective ‘sensitivity’ of the sensor (or rather the gain applied to the sensor signal), and the aperture/shutter speed combination selected from the resulting exposure reading. In many cases captured raw data will have linear tone reproduction, which allows gamma correction during processing. The method of ‘exposing to the right’ of the histogram (described in Chapter 12) may be used to expand the number of bits allocated to shadow detail within the image, which are reduced as a result of the linear output. In some devices a non-linear tone curve may also be optionally applied to the raw data at the point of capture. Although other variables such as bit depth, white balance and colour space may be set in the camera, they will not have an influence on the results, as they are adjustments that are applied after, or in conjunction with, demosaicing. For all other file formats, these settings define image processes which are performed by the capture device digital signal processor (DSP). In the case of RAW files the processes are instead applied during RAW conversion. The camera settings will be included with the data and may appear as the default settings in the RAW conversion software, but may be changed by the user.
Depending upon the software, RAW conversion generally includes options for limited resizing, fine-tuning of exposure and white balance, and setting of acquisition colour space as a minimum. Some versions allow complex tone and colour adjustments, for example to bring out shadow or highlight detail. These may be by the use of simple sliders or the more sophisticated editing afforded by levels or curves adjustments (see Chapter 27). Additionally, sharpening, noise removal and geometric corrections may be possible.
The range of RAW conversion software available and the differing degree and types of image processes available to the user can be problematic, especially if several cameras from different manufacturers are regularly used. Because RAW formats are proprietary, and non-standardized, the arrangement of the data, the information included with it and the RAW conversion software interface vary between manufacturers. Adobe has progressed some way towards providing a solution to this with the Camera RAW plug-in, which can perform RAW conversion using the same interface for the majority of versions of RAW files. However, because this is a ‘one-size-fits-all’ approach, in some cases better results may be obtained using proprietary converters. Further to this, Adobe have also developed their own RAW format, the Digital Negative (.dng), which is promoted by Adobe as a universal RAW format and is currently submitted for standardization. Some camera manufacturers have included support for .dng files as an option in their digital cameras.
The obvious advantage to working with RAW files is the degree of control afforded to the user. The image processing that is applied automatically by the capture device when rendering to an intermediate reproduction description is instead performed interactively.
Most importantly, the editing is non-destructive. The results are viewed on a large (and preferably profiled) display, prior to being finalized. Colour interpolation and other image processes are only applied to the data at the point at which the image is saved to another format – the user simply sees a preview of the results on screen prior to this.
The implications in terms of image quality are most apparent when applying corrections to the tonal range. If an image is saved as another format, then its tonal range and therefore its exposure level will be set across the range of values defined by the bit depth of the image file. As already discussed, applying image processes such as gamma correction to the range of values may result in posterization of the image, particularly in an 8-bit image. Changes in global brightness levels and image contrast are fundamentally limited – if the grey levels in the image fully cover the extent of the image histogram, then these changes may ultimately result in clipping of either highlights or shadows. With a RAW file, however, the full bit depth of the sensor is still available, meaning that the range of captured image values may be carefully adjusted to fit this range without clipping.
Figure 25.10a illustrates a slightly underexposed low-contrast image, which has been captured as an 8-bit JPEG file and a RAW file. Tonal adjustments to correct the exposure and contrast must be applied to the JPEG file in image-editing software (Figure 25.10b); for the RAW file the adjustments are applied during RAW conversion (Figure 25.10c). Although tonal adjustment has improved the exposure and contrast of the JPEG image in Figure 25.10b, there is some obvious clipping in the highlights. The adjusted RAW image in Figure 25.10c retains much more highlight detail, particularly noticeable in the hair and also in the pattern on the table. The histograms of all three images are displayed in Figures 25.10d – f. The last histogram, that of the adjusted RAW file, displays significantly less clipping than the one from the adjusted JPEG image, particularly in the highlights. It also displays less clipping than the original underexposed image, indicating that the actual exposure has effectively been changed, rather than the tonal range being redistributed between the same points. Furthermore, it is smoother, indicating a more even distribution of tones.
Using a RAW workflow has further advantages in terms of image archiving. The non-destructive nature of RAW conversion and the fact that the processed image is saved as a new format means that the RAW file is unchanged and can be repeatedly reprocessed.
The optimized RAW file can additionally be saved as a .dng file, maintaining the adjustments but allowing further editing. Assuming that the .dng format is standardized, it will become an ideal format for archiving images.
Image editing is a fluid process, in which the image may be processed in a rather ‘non-linear’ manner. The user may apply processes and then return to previous edited image states, combine edited elements of the image, process only selected image areas, or fine-tune the degree to which a particular image process affects the image. Various functionalities in modern editing applications, such as an ‘undoable’ and optionally non-linear history palette, the use of layers, image masking, sophisticated selection tools, and the ability to convert selections to vector paths, facilitate this approach to image processing.
Image file formats used at the editing stage are most useful if they are able to support some or all of these functions, allowing the user to save and reopen the image at a particular stage in the editing and pick up where they left off. Adobe’s Photoshop Document (PSD) format was developed for this purpose and most fully support these functions. TIFF and JPEG 2000 support unmerged layers. The PNG format does not support layers but allows for an extra alpha channel with variable transparency, of the type used in image masks. All of these formats are able to support 16 bits per pixel or per channel.
Additionally and importantly, all of these formats employ lossless compression (although JPEG 2000 also optionally employs lossy compression). Editing a lossy compressed image and then resaving it with lossy compression will inflict further errors on the image; each edit and save will cause further deterioration. Therefore, if an image is likely to require complex editing using the tools described above, then it should ideally be converted from a lossy format such as JPEG to a PSD or TIFF file, as an intermediate.
File formats for output are determined by the requirements of the particular imaging application. If the user is sending files directly to an output device then they are usually optimized, and the same considerations, described in the previous section on image acquisition, apply. If the images are for the Internet, then saving them as JPEG files in the sRGB colour space is a common solution, providing small file sizes and optimal colour reproduction on screen. Images to be sent to desktop printers should generally be either high-quality JPEG files or TIFF files, although PSD files may be proof printed. Generally RAW files require conversion to another format before proof printing.
When files are being delivered for reproduction elsewhere, the required format will usually be specified. For imaging applications such as photojournalism, high-quality JPEGs with the appropriate ICC colour profile are often adequate for both web and printed outputs. International Press Telecommunications Council (IPCT) metadata describing the source of the image, photographer’s details, copyright, captioning and keywords should ideally be embedded in the file, to ensure that all the required information is transmitted with the image.
In workflows producing higher quality printed media, images may be output as optimized TIFFs or high-quality JPEGs. Alternatively, proprietary RAW files may be delivered, allowing clients to perform the optimization themselves, although this is much less common and means that the photographer has no real influence over how the image is processed. A better option may be to deliver DNG files which have been adjusted, allowing the client to see the photographer’s optimized version, but with the option to edit it themselves.
If capturing RAW files it makes sense to archive the original RAW file prior to any processing. This facilitates the possibility of optimizing the images several times for different types of output and means that a master copy will always be retained. Ideally, the adjusted file should also be archived. The format for this will depend on the workflow, but using a format that supports layers means that image adjustments can be changed in the future, as the image will not be finalized until the layers are merged or flattened. Finally, if images are being prepared for a specific output, then it is good practice to archive a copy of the finalized file, in case there are any problems with file transmission, or a client loses a file and requests a copy. The file format in this case will depend upon the output and may include lossy compression.
RGB master files are optimized files with a wide gamut of RGB colour space which are kept at the native resolution of the sensor, unsharpened and saved as PSD or TIFF files. These represent the highest quality version of the image after it has been optimized and finalized. Archiving RGB master files with RAW files means that a high-quality and optimized file can be repeatedly used, for example for different outputs, without having to go through the RAW conversion procedure. For very-high-quality output, such as that required in high-end advertising, the RGB master file may be used to prepare several versions of the image for the client, each for a different output.
The production of an optimized image involves the application of a number of image adjustments, which are necessary to compensate for sensor characteristics and capture conditions. Optimization may involve cropping and resizing, tone and colour corrections, noise and dust removal, and a limited amount of sharpening. Corrections may be applied globally, but often there is the necessity for some local adjustments as well. Global adjustments are applied first, which may negate the need for further enhancements to specific areas of the image.
A high-volume, lower-quality workflow will be faster and more efficient if some of the adjustments can be automated at image capture. Examples include the application of an optimized-for-output custom tone curve in the capture device, or applying in-camera sharpening and noise suppression. In examples where JPEG images are being captured at the native resolution of the sensor, minimal global adjustments to tone and colour may be adequate for many images. Other processes such as resizing, geometric correction and further localized noise removal and sharpening may be necessary, but can be applied on an image-by-image basis in an image-editing application.
For high-end workflows more image adjustments may be required to optimize the quality of the final image. Fundamental in defining the order in which image adjustments are performed are the file formats and hence the types of software being used. If RAW files are being captured, then higher quality will be maintained by performing global adjustments to white balance, tone reproduction and colour in the RAW processing software. Global sharpening and noise removal may also be applied at this point. This is analogous to the in-camera processes mentioned above. Once these fundamental corrections have been applied, the image may be saved and opened in an image-editing application for further processing.
If RAW conversion is not part of the processing pipeline, then the majority of adjustments will be performed in a single editing application. There are exceptions to this, such as the use of third-party software and plug-ins developed to perform particular image-editing processes. Examples include applications developed specifically for resizing, noise removal or the sharpening of images, which use more sophisticated adaptive algorithms than those commonly available in image-editing applications.
When all image enhancements are being carried out in an editing application, geometric operations such as rotation, perspective correction and cropping should be performed at the beginning of the adjustment process. Cropping unwanted image areas is particularly important, as the extra redundant pixels will otherwise be included in the image histogram and may affect tone and colour corrections using levels and curves. It is logical to perform global tone and colour adjustments as a next step. Levels adjustments to individual colour channels can improve tone reproduction, enhance overall contrast and correct neutrals. Curves adjustments have the added advantage of altering one part of the tonal range (or colour range, if applied to separate channels) without significantly altering other areas, but this requires some skill (see Chapter 27 for more details on these types of spatial processes). Both types of adjustments are best applied as adjustment layers rather than directly to the image values. Editing in this nondestructive manner allows results to be fine-tuned and helps to minimize clipping, colour banding and posterization.
Noise removal and sharpening may be applied at various stages in the imaging chain. It is important to note that the application of one operation may result in the need for the other, as global sharpening can enhance noise and noise removal may result in unwanted softening of edges. Under certain capture conditions, such as those using high ISO speed settings, or long low-light-level exposures, noise removal will be necessary and may be best performed in the capture device or the RAW processor.
Sharpening may be applied at several different stages. Although in theory sharpening should not be applied until the final stage, due to the possible enhancement of noise, in practice it is often useful to sharpen earlier in the imaging chain to ensure that image quality is satisfactory for the requirements of the workflow. Capture sharpening may be performed in-camera or in scanner software to compensate for the digitization process, or alternatively during RAW conversion. Sharpening may also be performed if the image is being edited, as a final processing stage to compensate for blurring introduced by other processes such as resizing. Finally, the image may also be sharpened for a specific output. Whether sharpening is applied at only one or all the different stages will be dependent on the workflow and output.
It is important not to over-sharpen, as this will result in obvious ‘halo’ artefacts (see Figure 27.22). Improved results may be achieved using adaptive algorithms, or by applying luminance sharpening only. Results can be fine-tuned by applying the sharpening using a layer. Additionally, if a master file is being sharpened, then the sharpening should be saved on a separate layer, so that the degree and method of sharpening may be altered for different outputs.
Having discussed various aspects of image workflow, we complete this chapter with some case studies illustrating typical workflows currently used for different types of imaging. These are genuine examples, based on information provided by three different photographers.
The images produced in this example cover a range of different types of high-end commercial photography, such as portraiture, fashion and architectural work. Important objectives of this workflow are the acquisition and maintenance of optimum image quality and flexibility, where the output media is not defined, or where several output images are required on different types of media (at different qualities).
Images are captured at the native resolution of the sensor (e.g. 29 megapixels), using a medium-format digital back. The camera is controlled using remote capture software and a laptop. RAW capture is the only option and, in this case, scene-referred image data are output, to be converted to an output-referred image state using an embedded camera profile. A large range of profiles for the sensor under different illumination conditions are supplied. Using a camera profile provides a much larger gamut than any of the standard RGB working spaces commonly available at RAW conversion. Images are captured at the full 16-bit depth of the sensor.
The majority of image adjustments are applied during RAW conversion, using a workflow management application such as Adobe Lightroom. These include cropping, tone and colour corrections, all performed within the colour space of the sensor. If the image is being prepared directly for printed output, then it will be sharpened at the last stage.
If the image requires more localized adjustment, or is to be combined with other images, it is saved as a PSD file and imported into Adobe Photoshop. In this case sharpening is not performed during RAW conversion, but is applied using an unsharp mask (see Chapter 27) as a final editing stage in Photoshop.
Image quality is maintained by avoiding resampling processes where possible, applying the majority of image adjustments in a non-destructive manner (until the file is finalized), either during RAW conversion or using layers in Photoshop and working at 16 bits within the large gamut of the sensor colour space until the output stage.
Following the adjustments applied during RAW conversion, the optimized RAW files are archived (note that the adjustments are still non-destructive at this stage and simply illustrate the editing intentions of the photographer). If the images are further edited in Photoshop, then a working copy in PSD format, with adjustment layers still intact, will also be archived.
The final stages in preparing output files for print involve a colour space conversion to Adobe RGB 98, upsizing if a large print is required and conversion to 8-bit TIFF files. Images for Internet output are converted to sRGB, down-sized and saved as 8-bit JPEG files.
A large area of forensic imaging is concerned with the capture and presentation of evidential images, described by the Home Office Scientific Development Branch (HOSDB UK): ‘Evidence, in terms of a still image or video footage, is the presentation of visual facts about the crime or an individual that the prosecution presents to the court in support of their case.’ Such images require an audit trail, which may be written, electronic or both, describing the image source and capture conditions. The audit trail also includes details of any image processing carried out, to prove the authenticity of the image.
This example is concerned with the capture of finger-print images. The fine detail contained within fingerprints requires high-quality and high-resolution images for analysis or comparison. The international standard for fingerprint capture, based on the HOSDB UK guidelines (see Bibliography), specifies that the output images should be at lifesize magnification with a resolution of 500 ppi.
Images are captured at the native resolution of the sensor using a digital single-lens reflex (SLR) camera. They are most commonly captured as TIFF files (as a non-proprietary format), although working practices vary. In some cases proprietary RAW files may be captured instead. Capture colour space is set to sRGB, either in the camera settings or during RAW conversion. Generally image sharpening (if an option in the camera) is turned off at the capture stage. Low ISO speeds are recommended to minimize noise, although some images may be shot in low light levels. Noise removal may be performed by filtering, or by averaging multiple images of the same subject (see arithmetic processes in Chapter 27).
This stage varies depending upon how the image is going to be used, the local agency guidelines and the amount of processing applied at image capture in the camera settings. As a minimum, tonal correction is applied using levels and curves. Initial enhancements to tone and colour may be applied to RAW files in the RAW conversion software. The image is saved as a TIFF, copied and the copy is opened in another application for further processing. Following initial adjustments, the TIFF image is converted from RGB to an 8-bit greyscale image (using a ‘channel mixer’ to fine-tune the conversion, rather than a simple mode change), in accordance with the standard guidelines. Subsequent processing may include noise reduction, and background pattern removal, followed by local enhancements to selected areas of the image. The image may need to be resized to conform to the output resolution guidelines. This is usually performed at the end of the processing pipeline to minimize quality loss and may be followed by sharpening or blurring depending upon the image and how it is to be used. Finally, the image is saved as a TIFF ‘working copy’.
The images may be output to a high-end printer such as the Fuji Pictrography, or they might be saved to an image database for further analysis, possibly as JPEG files. The optimized TIFF file, after initial adjustment, is archived. The output working copy may also be archived.
This example is based upon working practices for clinical photographers in UK hospitals. The emphasis in this workflow is on the need for speed in processing and delivery, with some sacrifice in quality, as relatively low-resolution images are required at output. The image types and uses may vary, but in general the majority of images will be output as 6 × 4 inch prints, to be appended to patients’ records, or to be stored in an electronic database, and viewed on screen.
Required magnification varies and will be specified for each clinical department. Images are captured at the native resolution of the sensor using a digital SLR. Current practice is moving towards the use of full-frame sensors and RAW capture, although at the time of writing it is still quite common to use smaller sensors in semi-professional SLRs and capture high-quality JPEGs. Ideally, white balance is performed by measurement from a grey card. Images are captured into the sRGB colour space. If the recommendation to capture RAW files becomes standard practice, then it is likely that wider gamut colour spaces may be specified during RAW conversion for image editing.
JPEG captured images are archived at capture prior to any adjustments being applied. If they are only being downsized for output they may remain as JPEG files, but if further editing is required, they will be changed to a lossless format such as PSD or TIFF. Adjustments may include global tone and colour correction, dust removal and local colour corrections. The image is then resized, sharpened if for print output and resaved as a JPEG.
RAW files are initially renamed using a unique patient identifier and then resaved as RAW files. Following this, RAW conversion is normally performed in a generic RAW converter (such as Camera RAW). The majority of image adjustments are performed in this application, due to the non-destructive nature of the editing process, with Adobe Photoshop only being used for locally selected adjustments. Again, the image is resized at the last stage. If it is to be opened again for editing it may be saved as a TIFF, but in general output files are high-quality JPEGs.
Output images may be printed to inkjet printers or delivered as electronic files to an image database. This is a central server-based database, which allows access across the UK National Health Service to clinicians for their particular patient group. All output files are archived, along with the original captured files (JPEG or RAW files prior to optimization).
Special thanks to Andrew Schonfelder, John Smith and Simon Brown for their input in this chapter.
Anderson, R. (Ed.). Universal Photographic Digital Imaging Guidelines version 4.0. Published by the UPDIG coalition, available for download from www.updig.org
Cohen, N., Maclennan-Brown, K., 2007. Digital Imaging Procedure version 2.1. Home Office Scientific Development Branch.
Fraser, B., Murphy, C., Bunting, F., 2005. Real World Color Management, second ed. Peachpit Press, Berkeley, CA, USA.
Green, P., MacDonald, L., 2002. Colour Engineering: Achieving Device Independent Colour. Wiley, New York, USA.
Hunt, R.W.G. 1997. Bits, bytes and square meals in digital imaging. Proceedings of the IS & T/SID 5th Colour Imaging Conference.
International Standards Organization 2004. ISO 22028-1:2004. Photography and Graphic Technology – Extended Colour Encodings for Digital Image Storage, Manipulation and Interchange, Part 1: Architecture and Requirements. Available online from www.ISO.org
Lukac, R., 2008. Single-Sensor Imaging, Methods and Applications for Digital Cameras. CRC Press LLC, Boca Raton, FL, USA.
Sharma, A., 2004. Understanding Color Management. Delmar Learning, Thomson Learning, New York, USA.
Sharma, G. (Ed.), 2003. Color Imaging Handbook. CRC Press LLC, New York.
Süsstrunck, S., Buckley, R., Swen, S. 1999. Standard RGB color spaces. Proceedings of the IS & T/SID 7th Colour Imaging Conference.
3.135.192.141