Microscopia electrónica de barrido de los nanocompositos

The goal of this chapter was to provide insights into the possibility of using the CIE based L∗_a∗_b∗ _{colour space, as frequently used analytically and in industrial quality sensitive en-} vironments, for the domain of colour constancy. For this purpose, two well-known colour constancy algorithms the White Patch Retinex and the Grey World Assumption have been implemented to work on an L∗a∗b∗ colour space. The implementation attempted to build on a 1:1 preservation of the dening ideas of these algorithms without introducing any distorting improvements for a comparison. These algorithms are based on an implicit white point estimation, which is used to correct the colour space. We have analysed which methods of chromatic adaptation transformations work for the L∗a∗b∗ colour spaces. The CIE prescribed von Kries type transformations specically the Bradford transformation consistently yielded good results with the L∗a∗b∗ colour space. We showed results for successfully applied L∗a∗b∗ based colour constancy on several image sets, as well as vali- dated the potential for usability ofL∗a∗b∗based colour constancy for the purpose of object

4.7. CONCLUSIONS ON COLOUR CONSTANCY 59 recognition through colour indexing. Finally, we have discussed current weaknesses of the current implementation with indications on how to overcome them. Furthermore, details for an implementation regarding the selection of histogram ranges, as well as discarding clipped pixels have been highlighted.

Overall, the results have shown that the application of colour correction through the modied colour constancy algorithms onL∗a∗b∗colour spaces does work. Results on whether the results are superior or inferior to the original RGB based implementations are still inconclusive. For a more quantitative comparison of colour constancy on L∗a∗b∗ vs. RGB colour spaces the colour indexing would have to be applied consistently to one colour space, regardless on what colour space the colour correction operates.

Future work in this area should include further comparative analyses with image sets that were not as under-exposed, or more normally exposed to conrm the theory, that the Bradford transformation preserves the saturation of under-exposed imagery. Furthermore, the behaviour of the colour constancy algorithms with respect to used parameter sets (for dierent colour spaces) needs to be analysed, to improve the eectiveness of the implemen- tation. Lastly, it should be possible to signicantly improve the colour indexing performance by selecting histogramming ranges to t more tightly to the encountered chromaticity gamut.

Chapter 5

Colour Management

If one says `red' (the name of a colour) and there are 50 people listening, it can be expected that there will be 50 reds in their minds. And one can be sure that all these reds will be very dierent.

Josef Albers, Interaction of Colour, (1963)

The purpose of colour management is to achieve most constant colour reproduction of images acquired from arbitrary input devices, onto other arbitrary output devices. In image reproduction workows, this is not a simple source to destination mapping, but includes various steps of intermediate image manipulation (editing) as well as output simulation (proong) on alternative output devices (e. g. printing a sample on a comparably cheap ink jet printer to simulate the appearance for the industrial production printing press). The industrial demand for these steps was quite large due to too many rule of thumb manip- ulations in the processing chain. The International Color Consortium (ICC) was funded in 1992 under the lead of FOGRA (German Graphic Technology Research Association). The ICC was to dene universal standards, tools and applications for characterising dierent devices through ICC proles and the transformations between the colour representations using these proles through Colour Management Modules (CMMs) [4548].

5.1 Introduction

A high similarity between the colours of images between input and output devices is called colour delity. To achieve colour delity, Colour Management Systems (CMS) are used, however it is almost impossible to yield ideally correct results.

Colour management systems use device independent descriptions (device proles) and device independent exchange colour spaces (prole connection spaces, PCS). The CMS uses the CMM to convert between these device dependent and device independent colour descrip- tions to enable colour delity between input and output devices. The numeric representation

through colour descriptors in the input/output device spaces (often an RGB or CMYK tu- ple) and the PCS (a CIE L∗a∗b∗ or XYZ tuple) is dierent, but should match up visually with good accuracy. The PCS colour representation can therefore be seen as a canonical representation of the colour itself, whereas the device dependent representations are only valid for a particular device under specic conditions. The reason for this is, that colour descriptions are device specic, and appearances can be very dierent. The CMS translates between the dierent interpretations.

In many cases, the colour workow is not short and linear just between an input and an output device. For those simple cases one single device link transformation could be used. However, with a variety of input as well as output devices the number of potential links increases dramatically. If one is dealing with the common case of having to manipulate imagery interactively on a work station, which is to be printed later, already two dierent types of output devices (the screen as well as the printer) have to be handled and managed to express the same colour [45]. If the printing is nally going to happen on a printing press, often a preview (or more accurately a proof) is printed on a smaller commodity printer (e. g. ink-jet printer) to simulate the nal product. In this case, the dierent printing characteristic of the press has to be simulated by the CMS for the smaller printer.

Several factors need to be considered in the ability to match colours between dierent devices. These all may induce shortcomings in accuracy of the colour translation, and will in many cases cause limitations in dierent regions of the colour space for dierent transformations.

• Concepts of colour image perception and generation: additive colour systems (for screens) vs. subtractive colour systems (for printing);

• Dierently primary colours;

• Dierent media properties (whiteness of the paper or media, blackness of the CRT or at screen, illuminant white point).

Ideally one would work only with a true spectral/physical representations of colour. This however, is in practice not possible, as it would involve too much eort, and for practi- cal purpose current devices are using mostly three (or four for printing) primaries. Devices with higher spectral resolution tend to be very specialised, expensive and not much com- monly available. Colour management tries to decouple the handling of colour from spectral distributions towards working on colour tuples only, as they are commonly used for input and output devices, but handle transitions between them in a scientically sound way.

ICC based colour management has been designed with the whole process in mind, as it is required by the graphic arts industry. Therefore it is laid out to aid in minimising the pro- gression of colour errors through an imaging workow, which covers capturing, modication, analysis, output, proong and potentially further steps.

5.2. HARDWARE 63 The ICC describes Colour Management in their section Terms and Denitions 4.8: Colour Management (digital imaging) of the ICC specication [49] like this:

communication of the associated data required for unambiguous interpretation of colour content data, and application of colour data conversions, as required, to produce the intended reproductions

NOTE 1: Colour content may consist of text, line art, graphics, and pictorial images, in raster or vector form, all of which may be colour managed.

NOTE 2: Colour management considers the characteristics of input and output devices in determining colour data conversions for these devices.

In the following of this chapter, we are discussing dierent aspects of colour management. Sect.5.2 starts o with ways to handle the characteristics of the hardware in the form of input and output devices. In Sect. 5.3, colour proles, as derived from the hardware, are covered, with the process of the prole creation in Sect. 5.4. These proles then can be used to transform colour representations as outlined in Sect.5.5. The device independent colour is dened relative to the standard illuminant, and a conversion to an alternative illuminant is described in Sect.5.6. Dierent freely available implementations for CMS' are listed in Sect. 5.7. Finally, in Sect.5.8, we are concluding of how colour management and its properties are useful for the scope of this research.

5.2 Hardware

As every human perceives colour individually dierent, also devices do so. Particularly dierent classes of devices understand colour dierently, as they might use dierent colour spaces to represent the colour, use dierent sensors, potentially with dierences in the pri- maries, due to variations in production over time or age of the device, etc. This can be seen both from the side of perceiving as well as reproducing colour.

To describe the behaviour of the dierent devices and device types, their colour response has got to be measured and compared to a reference that is known in a standard colour space. This reference can be established for input devices by measuring the test target's colour (see Sect.5.4) with a precise spectrophotometer prior to capturing it with the hardware (e. g. a scanner or camera). For output devices a test target pattern is transferred to the output device, and the output is measured afterwards using a spectrophotometer or colourimeter in a standard colour space.

To capture and control the behaviour of the hardware, two basic processes can be used: Characterisation and calibration. Most commonly one is dealing with characterisation, in which the device is used as it is, and the relationship of canonical colour to device colour is analysed. This describes the colour imaging characteristics of the device. Calibration

is often used synonymously to characterisation. But strictly speaking, a calibration may include adjustments to the device itself.

To circumvent the need of full characterisation or calibration many devices are produced to a certain common norm. One of these is for example the very common sRGB standard colour space created cooperatively by Hewlett-Packard and Microsoft in 1996 for use on monitors, printers and the Internet. It has been designed with the goal of being a common denominator to avoid colour mismatches in day-to-day image representation. As it is as- sumed to work on a wide variety of dierent classes of devices, it has been chosen to have a comparably small gamut of colours it can represent, which mostly work on a wide variety of imaging devices. But many devices' colour imaging capabilities go beyond the specication of sRGB, so it is not necessarily good to rely on it for all purposes. Furthermore, manufac- turers are inherently interested in selling their hardware. Therefore, it is quite common that factory settings are aiming to improve brightness and brilliance to be as good as or better compared to competing products. Manufacturer settings therefore are tuned for marketing, not colour delity.

In imaging, therefore, one can gain signicant improvements in colour accuracy by using colour management with characterised (and possibly calibrated) hardware. Even devices from a single manufacturer, or within the same product line, can vary quite signicantly [47].

5.3 Colour Proles

Colour proles in the form of ICC proles form the basis to convert colour representations through ICC standardised colour management. They describe a mapping of device colours (colour response or display colour) to canonical standard colours. All colour proles map some specic colour space to a prole connection space (PCS). This PCS is either CIE XYZ or CIE LAB, which are dierent but equivalent representations of the same colour space. The specic colour spaces are usually either device spaces or working spaces.

Device spaces are specic to a particular device's colour representation. These could be for a scanner or digital camera often in RGB or YUV, and for a printer for example CMYK or RGB. The proles are only valid for the specic conditions they were created for. These conditions on the one hand consist of device settings, and on the other hand further boundary conditions. For cameras, these are specic light conditions, for a printer the media (paper, lm, ink, toner, etc.) have to be set as constant for a given prole. The working conditions of hardware may also need to be kept within certain boundaries (e. g. by ensuring a sucient warm-up phase for devices).

Working spaces are colour spaces that are used for processing and editing of colour im- agery. These can be spaces such as sRGB, Adobe RGB or ProPhoto. As these working spaces are completely xed and dened, they will be equal in all cases, and predened, con- stant proles may be used. In working with colour imagery, one has to keep in mind that a

5.4. PROFILE CREATION 65 conversion to a working space may induce clipping (limiting intense colour values to mini- mum/maximum intensity), if the gamut of the working space is too small; or posterisation (banding, coarse colour sampling), if the working space gamut is too large.

Lastly, some other ICC proles exist: Some are used to represent the canonical CIE colours for conversions to the PCS; Others may for example describe a direct link between two distinct devices.

Proles can be represented separately to be used by the CMS, for example in the le system, a data base or directly in memory. Within certain image formats themselves they may also be embedded (such as TIFF, JPEG, PNG, EPS, PDF, and SVG). This may be particularly convenient for images to store colour information in their original format without degradation of quality, by embedding the capturing device's characterisation or the quantisation of the chosen working space [45,48].

5.4 Prole Creation

The prole creation is based on a colour measurement. Commonly, spectrophotometers or colourimeters are used for this purposes. For output devices (e. g. monitor, printer), the known device colours will be measured in the output by the measuring device in the canonical colour space. Inversely, for input devices (e. g. camera, scanner), the known canonical colours will be captured by the input device in the device colour space. Using these known device and canonical colour pairings, a mapping or transformation between these colour spaces is derived. Also derived from this is the gamut the totality of all representable colours of a device. Depending on the type of a device, these proles are generated in dierent ways.

Many device colour spaces, especially those with 8 bit per channel encoding, use gamma correction. To achieve good quality proles, as a rst step a set of linearisations is performed between the device colour and the rest of the transformations to undo the gamma correction (using the A curves, see Fig. 7.1 in Chap. 7). The device colour data is now present in a more uniformly distributed form, which largely increases the quality of the achievable transformation.

The enclosing volume of the scattered measurement data is used to dene the gamut of the device. From this data, a transformation between the linearised device colour and the device independent prole connection space data must be determined. This step is the core task of proling. Proling may be used to generate two basic types of proles. These are on the one hand the mathematically very straight shaper/matrix proles. They include a set of shaper curves (M curves), a linear transformation matrix to yield a best t, and optionally post-linearisation curves (B curves). These proles are only available for devices using exactly three colour channels. For a more detailed description see Chap.7. On the other hand, a prole may contain a multi-dimensional colour lookup table (CLUT). CLUT based proles are much more complicated to create, but their capabilities and matching accuracy

by far exceeds the shaper/matrix proles' for real world devices. Computations to determine a suitable CLUT usually involve a complex mathematical process using multi-dimensional spline tting, interpolation, smoothing, etc. Additionally, often testing, personal judgement and parameter tuning to obtain proles performing as good as possible are also involved. Again, see Chap.7for a much more detailed description and implementation notes.

Due to aging eects, hardware devices' colour accuracy may undergo a certain drift, so that proling is not a one-o procedure for a specic device (for given media or illumination conditions). Regular re-proling of these devices is usually necessary in these cases. Also, manufacturer provided proles are only valid to a certain extent for a certain device series under conditions that are not too well documented by the manufacturer. It is usually advisable to perform individual characterisations to generate ICC proles for specic given conditions.

The obtained ICC prole, as a result of the hardware characterisation, is either an input or an output prole. Input proles commonly only demand translation capabilities from device colour space to the PCS. As a minimum output proles must provide a PCS to device colour space transformation. In practice however, especially for the graphic arts industry, it is often necessary to also include a transformation in reverse. It is used to compute backwards the appearance of a print, and emulate its look on a dierent output device. This process is called proong: For screens, this is a soft proof, and for other physical output media it is a hard proof [45,47,48].

5.5 Colour Translation

In most practical applications and workows, a user is not interested in the colour informa- tion on the level of the PCS. ICC proles only describe the characterisation of the devices. As stated above, a colour management module (CMM) is responsible for conducting com- putationally the transformations described by the ICC proles on the colour information. Typical, simple colour workows would therefore have a form as illustrated in Fig.5.1. The intermediate PCS can be considered as a lingua franca one can translate every input de- vice colour encoding into, and from which one can convert into every other output device colour encoding. Captured imagery from a proled camera could be, for example, displayed (comparatively) true to original colour on a proled display. If the image is to be used on various unknown output devices, for example to be distributed on the Internet, a generic sRGB output prole could be used, as most computers know how to handle sRGB encoded colour information at least reasonably well.

For this reason, typical workows consist of two colour space transformations, which the CMS conducts through the CMM: A transformation to the PCS, and then from PCS to some output colour space. As this is a very common task, modern CMMs know how to smelter this multi step process into a single transformation operation to gain computational eciency and reduce an accumulation of repeated rounding errors [42]. If one is interested in

5.5. COLOUR TRANSLATION 67

Figure 5.1: Common simple colour transformation workows to retain colour delity on the path from an arbitrary input device towards another arbitrary output device.