METHOD FOR PERFORMING TRANSFORMATIONS OF COLOR DATA WITH TRANSFORMATION PRECAUTION

Description

[0001] The present invention relates to a method for carrying out transformations of color data from a source color space with n colors into a target color space with m values ​​using a transformation rule TRV.

[0002] In the printing industry, color transformations are often required to create the data sets for the color print products to be printed.

[0003] The starting point is digital layouts, which can contain, for example, color images from photographs or scans, illustrations and text. These layouts are templates that are intended to be reproduced in the form of printed products. Such a layout is designed or developed in some form or other and, in terms of color appearance, represents what is desired for the intended purpose. A layout usually comprises several layout objects. Each individual layout object therefore has a color appearance that can be clearly identified. As is well known, color appearances are created by mixing just a few colors; for printed products, for example, by simply overprinting rasterized printing inks, with each color making a certain contribution through its effect, and the final result being created through the interaction of the different color layers.

[0004] The color appearance of each individual layout object is defined by its color values ​​(of image pixels, outline colors, or fill colors), which represent the data source. In the case of image pixels, these are often millions of pixels, each with its own color values. These source color values ​​always refer to a source color space with a specific number of colors. This means that each object in the layout is assigned a color space, which usually has a relationship to a device, often a very specific RGB color space such as sRGB or Adobe RGB, or a CMYK color space, only black (grayscale), or another printing color space. The object color values ​​are numbered accordingly as RGB values, CMYK or grayscale values ​​(0-100% color), or other color values. The total number of device colors in the source color space is denoted here by n.

[0005] Color spaces are usually described by ICC color profiles (ISO 15076-1:2010), layouts by PDF and PDF / X files (ISO 15930 series), which can contain or reference these ICC color profiles.

[0006] The layout may need to be reproduced for a variety of reasons and motives. A wide variety of printing processes and printing presses are available for reproduction. These have the capability of applying a specific number of colors to a substrate in a specific way. These are the device colors of the printing system. These define or form the basis of the target color space, which in turn encompasses all conceivable combinations of the printing system's device colors. These are referred to as the m device colors of the target color space. Transformations into device-independent values ​​are also often necessary, such as Lab or spectral colors (see below), for example, to be able to check the results of subsequent print measurements or to provide a device-independent database for display on any device. For this reason, the term used is generally referred to as the m values ​​of the target color space.

[0007] Typically, the source color space and the target color space do not match. Often, standardized color spaces are used as source color spaces when designing the layout, but printing takes place using specific output systems with their own individual color behavior. Therefore, for each screen point in the layout, it is necessary to transform the given combination of the n device colors of the source color space into a combination of the m device colors of the target color space required for printing. The goal is to produce a print result that is as close as possible to the original, i.e., the digital layout, and ideally creates the same appearance for the viewer.

[0008] A raster point or pixel is each point in a layout resulting from a predefined rasterization of the layout objects. This, in turn, is related to the so-called resolution of the display, for example, for creating the printing form. If a raster with i rows and j columns is placed over a layout, the resulting resolution is ixj pixels. Each individual pixel can then be identified based on its color construction.

[0009] Color transformations are used to convert layout data into print-ready data for a desired set of printing colors. This creates, for example, uniform, pure CMYK data for offset printing. Color transformations are also used to accurately simulate the color appearance of this print data before printing. The color appearance is usually described device-independently in the CIELAB color space (Lab for short) or using spectral reflectance factors. The print data, with its associated color space, is transformed into this device-independent color data. This is displayed on a calibrated screen or with special proof printing systems (a so-called "proof," usually with inkjet systems). Related to this is the task of adapting print data for other printing processes, i.e.To create new print data for other printing colors from existing print data by transforming it and thereby preserving the original color impression as best as possible.

[0010] In the usual printing processes, which usually involve no more than four colours, these transformation processes are largely developed and standardised by appropriate data systems.

[0011] However, in packaging printing, for example, more than four printing colors are often used. Traditionally, brand colors are supplemented as spot colors, and images are created with product-specific colors (e.g., dairy products with an additional light gray printing ink, chocolate products with various brown printing inks). For transparent film packaging, white printing ink is often used in addition. Each printing ink used occupies a printing unit in the printing press and requires its own printing form for ink transfer. The colors are applied one after the other, and in principle, all colors can be combined (overprinted). The totality of the separate printing forms leads to the term "separation" as a process or result of creating the print data.

[0012] Press systems with 7-10 color units are not uncommon in flexographic, offset, and gravure printing. Designers of branded goods can choose their printing inks from thousands of spot colors, including over 2,000 shades in the PANTONE® system. These printing inks are produced and supplied by ink manufacturers. Digital printing uses inkjet and toner-based printing presses that often contain 6-8 color units, but operate with fixed ink sets, as inkjet inks and colored toners often require very specific, machine-dependent processing properties, or changing them would be technically too complicated. Typical color sets include CMYK, supplemented by orange or red, green, blue or violet, and white.

[0013] Digital printing processes are currently slower than conventional printing processes, but since no printing plates need to be created and inks are not changed, the setup effort is much lower, making short runs more economical and personalized printing possible. The desire here is therefore to print packaging that was initially produced conventionally in short runs or even personalized (run of 1) as needed, but with the color appearance being as identical as possible to previously printed products that were produced with individual spot colors. This is because different batches can be placed side by side on the sales shelf. To achieve this, the print data available for the conventional printing process is converted from its source color space into a target color space that describes the digital printing process.In the extreme case of variable, personalized printing, different source data must be transformed for each copy, and then the transformation becomes a bottleneck.

[0014] The color transformations mentioned above convert from a source color space Q (for example, with n device colors) to another target color space Z with m values ​​(for example, device colors or the three CIELAB values ​​L, a, b). This is done by applying a transformation rule TRV. This can be a calculation formula, transformation models, or transformation tables already created based on the TRV. They often have to be applied per pixel, so high throughput is important (millions of pixels per second). To achieve this, interpolation takes place in a custom table, such as the ICC industry standard. The table is structured for combinations of n color values. Each of the n input colors can assume values ​​between 0 and 100% ink coverage. The table cannot list all combinations, but samples each input color at a specific, predefined resolution.It is a regular rectangular n-dimensional grid containing all combinations of the sampled levels. N-dimensional interpolation is performed between the levels. The accuracy depends on the sampling. Typical samplings are 33 levels for n=3 (RGB / Lab) (33 3< ≈ 36000 table entries), and 17 levels for n=4 (CMYK) (all 6.25%, 17 4< ≈ 84000 table entries). The number of levels is the base of the power, and the number n of printing colors is the exponent. The generation of transformation tables involves measures familiar to those skilled in the art. They are based, for example, on the spectrometric measurement of printed colors, calculations for estimation and / or interpolation, and the like.

[0015] The table size increases exponentially with the number n of input colors. Because storage space is limited, the sampling must be coarser for practical reasons. For the common case n=7, often only 6 or 7 levels are used. An example of this is the widely used open-source color management system LittleCMS, which uses 7 levels (all 16.67%, 7< ≈ 820,000 table entries). The table structure therefore leads to the following problem: the more printing colors, the larger the table, and the fewer gradations can be stored. Users expect a comparable level of accuracy with CMYK + additional colors as they are used to with CMYK. 7 levels are considerably too coarse.

[0016] Known approaches to a solution are disclosed in DE 10 2004 003 300 A1, according to which assignment tables that have more than four input color components are divided into several assignment tables with a maximum of four input color components. This technique is intended to solve the problem of assignment tables that are too large. It is a concrete proposal to solve this problem by conventionally reducing the number of tables to four-component tables, since the state of the art utilized the fact that, even in a seven-color print, usually no more than four colors are printed on top of each other at one point on the printed sheet. For this reason, the reduction to four-component tables is being pushed forward. To ensure this in case of doubt, additional methods such as Gray Component Replacement (GCR) and Color Component Replacement (CCR) are used when creating the separation table.Similar prior art can be found in Boll H: "A COLOR TO COLORANT TRANSFORMATION FOR A SEVEN INK PROCESS", PROCEEDINGS OF SPIE, IEEE, US, Vol. 2170, 1 February 1994 (1994-02-01), pages 108-118, and US 5 892 891 A.

[0017] Known solutions use a model or a combination of table and model. A model is usually a mathematically based transformation rule. This allows the Lab color value of an overprint to be estimated from the individual printing colors. This is done, for example, when displaying layouts or PDF files on screen and is fast, but inaccurate. The ICC enhancement iccMAX proposes calculating individual colors spectrally or using tables for the CMYK content and modifying their value for each additional color. In all cases, adding individual colors is too imprecise because it does not, or insufficiently, take into account the interaction of the colors in overprints in which these colors are involved.

[0018] While there are more accurate models that better calculate the overprint behavior of color mixtures, they are far from fast enough to be useful for transforming millions of pixels. Furthermore, these models provide Lab values ​​or spectra, but not device color values, making them unsuitable for direct device-to-device transformation ("device-link"). The output device color values ​​are often subject to special separation rules (see below) and usually require special processing for high quality. However, transformation to device colors is an extremely common application.

[0019] Color transformations are often applied to high-dimensional spaces, occurring on both the input and output sides.

[0020] On the input side, for example, this could be the color combinations of a printing process. Traditionally, the colors cyan, magenta, yellow, and black were the standard in multi-color printing. For several years, additional colors have been used to expand the color range; in particular, efforts are underway to standardize seven-color printing with additional orange, red, green, blue, or violet (known as ECG printing = Expanded Color Gamut). Even more colors are used in packaging printing and specialty applications such as banknote printing.

[0021] On the output side, there are m values, such as CIELAB color values ​​(three components), printing ink combinations of a target system or spectral reflection factors (with values ​​for wavelength ranges; e.g. m=31 values ​​for 10 nm bands from 400 to 700 nm).

[0022] A common application is predicting the resulting color effect based on the reflection spectra (output side) of proportionally overprinting colors (input side) in an industrial printing process. Only with knowledge of such printing characteristics can further transformations be created that adapt the data from one printing process to another in a color-consistent manner. Color transformations are based, as explained, on tables or models.

[0023] Models can be either empirical-mathematical approaches or mechanistic-physical simulations of color behavior. While both approaches involve adjusting model parameters, there is usually a residual deviation between the model and reality that one would like to reduce. Thus, one would like to correct the rigid model, which follows its internal laws, to independent target data.

[0024] Tables in which multi-dimensional interpolation is performed offer greater flexibility, as the entries can be freely adjusted. The problem here is that with an increasing number n of input dimensions (printing colors), the storage space for the color combinations increases exponentially, making detailed storage no longer possible. One proposed solution is to only store the combinations that occur in the same place when overprinting. This is usually only four overprinting colors. Consequently, a few four-dimensional tables are sufficient instead of one large n-dimensional one. However, it can occasionally happen that more than four color components occur in the color processing chain that need to be transformed. Our proposal, outlined elsewhere, is to estimate the output values ​​by assembling several contributions from the existing tables: for example, usingthe largest 4-dimensional contributions and adds the missing colors by determining their color effect from other tables in which they appear and applying it to the large 4-dimensional output value. This estimation method is ultimately a simple model that usually deviates from reality. However, a direct fit is not possible because there are no table spaces for more than 4 dimensions.

[0025] In this respect, it is desirable to correct the results of a table or model calculation to given target values. The correction should not only have a selective effect, but should also smoothly adapt the entire behavior of the neighborhood.

[0026] Correction methods that compensate for differences between target and actual values ​​are available in the state of the art. US 2005 / 0128498 provides expected reference values ​​in the device-independent target color space Lab (m=3) for some printed color combinations consisting of color components in the printer source color space (here CMYK, n=4). The color combinations are printed using a specific printing device, the actual Lab values ​​are measured, and the deviations are determined as the difference. Based on the differences, a reverse transformation table is corrected from the absolute Lab color space back to the printer color space. This improves transformations of given device-independent Lab color values ​​for output on this specific device, but it does not correct the original transformation from the source to the target color space to the actual values.

[0027] US 2006 / 0280537 also deals with a specific printing device and describes how, after determining the need for calibration, a non-specifically defined difference profile is generated and "synthesized" with the standard profile. Using the synthesized profile, color-corrected print data is generated and printed. Here, too, the goal is to compensate for color changes and individual device differences. Here, too, the original transformation from the source to the target color space is not corrected to the actual values.

[0028] Based on the above-described prior art, the invention is based on the object of improving a method for carrying out color space transformations using transformation rules in such a way that the results of a table or model calculation (n input values ​​to m output values, hereinafter referred to as core model) can be corrected to given target values.

[0029] The technical solution to this problem consists in a method having the features of patent claim 1. Further advantages and features emerge from the subclaims.

[0030] The invention is based on the additional use of a correction method.

[0031] The invention also involves storing corrections in multidimensional tables, but these require very few control points, ideally only 2 (for 0 and 100% of the input color), and contain only as many dimensions as necessary, not necessarily all n input colors. Several such corrections can be applied sequentially to different, even overlapping, subspaces of the n colors. Due to the small number of control points, the interpolated correction is automatically smooth and has a wide radiance.

[0032] The invention accordingly proposes a method for carrying out transformations of color data from a source color space with n colors into a target color space with m values ​​using a transformation rule TRV, characterized by the following steps: a) Selecting a set of color data Q1 of the n colors of the source color space with the color components q(1), q(2), ..., q(n), for which a target value data set Z1 is available in the form of combinations of components z(1), z(2), ..., z(m) of the target color space; b) Generating a second set of color data Q2 of the n colors of the source color space with the color components q'(1), q'(2), ..., q'(n), which are initially set identically to Q1; c) Transforming the set of color data Q2 using the transformation rule TRV to generate an actual value data set Z2 of the m values ​​of the target color space z'(1), z'(2), ..., z'(m); d) Applying an n-dimensional optimization method which, by suitably modifying the set of color data Q2 and repeating step c), minimizes the differences between the target value data set Z1 and the actual value data set Z2 in the sense of the least squares, so that Q2 finally contains the resulting optimized color components q'(1), q'(2), ..., q'(n); e) Calculating the differences Dq(1) = q'(1)-q(1), Dq(2) = q'(2)-q(2), ..., Dq(n) = q(n)-q(n) of each color component of the n colors of the optimized color data set Q2 to the respective color component of the color data set Q1; f) Calculating the differences Dz(1) = z(1)-z'(1), Dz(2) = z(2)-z'(2), ..., Dz(m) = z(m)-z'(m) of each component of the color data set Z1 to the respective component of the color data set Z2; g) Applying the differences Dq(1), Dq(2), ..., Dq(n) and Dz(1), Dz(2), ..., Dz(m) for transformations of sets of color data from the source color space with n colors into the m values ​​of the target color space, by adding the difference Dq(i) to the color data q(i) of the n colors of the source color space for the i=1 to n values ​​when transforming a data set, transforming the modified color data q'(i) into the m components z'(j) for the j = 1 to m values ​​of the target color space using the transformation rule TRV and then adding the corresponding difference Dz(j) to the respective component z'(i). .

[0033] For a correction, a suitable, robust method for adjusting the output values ​​must first be selected. Simple options include differences to be added or factors to be multiplied, or other mathematical functions of value and correction value. This choice determines the type and application of the table entries. Differences are suitable when the correction in the target color space is largely independent of the size of the output values ​​and have the advantage that the corrected values ​​are more limited. Factors are suitable when the nature of the correction is more relative, i.e., value-dependent, e.g., for reflectance spectra, where model errors can often be easily understood as deviations in transmission, which is a multiplicative process. Where no corrections are required, the ineffective entries are used (e.g., 0 for addition, 1 for multiplication).Factors can easily be reduced to differences by making corrections in logarithmic space.

[0034] The invention assumes that for a set of color data (called Q1) from the n colors of the source color space, a secure, i.e., verified, target value data set of the m values ​​of the target color space is available. This data set represents a type of reference and is designated Z1. The verified data set is obtained, for example, from the unique processing result by printing the n color components and spectrally measuring the result. Or it can be obtained, for example, from selective, complex model calculations for the n color components. If the target color space represents device-independent Lab values ​​or spectra, spectral overprint models can be used for this purpose. If the target color space represents m device colors, color separation algorithms can be used to calculate a suitable target color structure with m device color components in the target color space, taking printing rules into account.

[0035] A second set of color data Q2 is created as a copy of the set of color data Q1 so that it can subsequently be modified independently. Q2 is transformed using the specified transformation rule, for example, using transformation tables. This creates another data set of the m values ​​of the target color space. It is denoted by Z2 and represents the actual values ​​of the transformation.

[0036] The differences between the two data sets Z2 (the transformation result of Q2) and Z1 (the target value data set) are determined.

[0037] These differences can now be applied to the transformation of color data from a source color space with n colors to a target color space with m values ​​using the TRV transformation rule. They are recorded, for example, in a correction table.

[0038] According to the invention, the data in the set of color data Q2 of the n colors of the source color space are modified and transformed using the specified transformation rule. This creates a new Z2. This modification of Q2 and transformation in Z2 is carried out until the differences Dz between the two data sets Z1 and Z2 are minimized in the sense of least squares. Well-known methods for n-dimensional nonlinear optimization in numerical mathematics, such as the Levenberg-Marquardt algorithm, are suitable for this optimization task.

[0039] This ultimately results in an optimized set of color data Q2 from the n colors of the source color space, and the difference Dq of the modified data Q2, which has generated minimized differences, to the unchanged data Q1 can now be determined.

[0040] These differences Dq and Dz can now be applied to the transformation of color data from a source color space with n colors to a target color space with m values ​​using a transformation rule TRV.

[0041] If several correction steps are carried out consecutively that affect overlapping subspaces, the result generally depends on the order. This means that the correction values ​​of later steps also depend on the previously performed corrections. It is therefore advantageous according to the invention to generate the correction tables from the target values, the core model TRV and the sequence as follows: The available target output values ​​are given in the form of one or more coarse-level multidimensional tables, which expediently cover larger subspaces (with u colors) than the stored fine-level (e.g., k-dimensional) table or model calculations, up to the full n colors, if desired. A correction chain is constructed in the order of these tables, which is initially empty.

[0042] Using the first setpoint table, the core model is called for all entries (as input data sets), the result is compared with the respective setpoint in all m components, the input values ​​are optimized if desired, the required correction value of the input is calculated based on the optimization and the output based on the remaining difference to the setpoint, and the two corrections are entered into two new correction tables. These tables are placed around the core model as additional correction transformations on the input and output sides.

[0043] The next setpoint table is then processed in the same way. The resulting second pair of correction tables takes the effect of the first table into account, preventing incorrect multiple corrections from occurring in the case of overlapping entries. Overlapping, redundant setpoint entries should, of course, contain the same output values, so the set of setpoint tables should be internally consistent.

[0044] At the end, a procedure is available that performs soft adjustments to the target values ​​of a comparatively small number of grid-based control points by calling the core model and a chain of upstream and downstream corrections.

[0045] It is, in a sense, a collection of concentric shells around the core model, each shell consisting of an input and an output adjustment, both defined on the same grid of input values. In other words, the table pairs have the same sequence of table entries; each entry contains both n correction values ​​for the input and m correction values ​​for the output. When transforming input values, the input corrections are applied first, then the next innermost shell (or the core model at the very center) is called, and the output corrections are applied upon return.

[0046] Depending on the application, it may be appropriate to apply only one or both of the differences Dz(i) and Dq(i) in step g). In general, the use of both difference tables results in, firstly, the input data sets being precisely transformed to the target values. Furthermore, this not only results in a simple interpolation of the deviations, but also in the compensation of systematic deviations on the input side. This provides a significant advantage over the state of the art by expanding the correction options and improving the corrections.

[0047] In order to reduce the number of work steps, it is advantageously proposed that in step d) only those color components of the n colors that are not zero are changed.

[0048] The method according to the invention is carried out according to a proposal of the invention on a computer unit by means of control software, wherein the computer unit comprises an input unit for providing the digital color data of the project and an output unit for outputting the transformed color data as well as a memory on which the transformation tables are stored, wherein by means of the control software for input color data of the project using the transformation tables, color data for the target color space is generated and provided in a data set

[0049] A more formal, general description of the procedure now follows: Let Rn be the n-dimensional space of input vectors, and Rm be the space of m-component output vectors. Target values ​​T from Rm are known for points P from a subspace Ru of the input space Rn with u ≤ n dimensions. A correction shell is placed around the core model M (Rn -> Rm). A correction shell generally consists of an adjustment A of the input values ​​(Ru -> Ru) before the model and a coupled adjustment B of the output values ​​(Ru -> Rm) after the model. The corrected model (A -> M -> B) is still a mapping n -> m. Further shells can be placed around the corrected model. A or B can be omitted. Formally, they can then be regarded as identical mappings. In implementation, the respective step would then be skipped. A and B are determined by processing all known points P as follows: The input point P from Ru has its target value T from Rm.The model returns a model output value V for this input vector (supplemented by zero entries in the missing nu dimensions).

[0050] When mapping A is used, it serves as a preparation for the input values. Then, for each point P, a modified input value P' is sought whose model output value V' is closer to T. A frequently used method is the Levenberg-Marquardt method, a multidimensional nonlinear optimization of the input parameters P, in which constraints such as the valid domain are used as needed. The mapping A thus consists of pairs (P, P'). This step occurs before determining B.

[0051] If mapping B is used, the model (possibly pre-corrected by A) is queried for each point P. This means that if mapping A is used, P -> P' is mapped with it and P' is fed into the model. If mapping A is not used, P is fed directly into the model. The model provides a model output value V with m components. The deviation from the target value T can be seen, for example, as a component-wise difference D=TV or as a component-wise factor F=T / V, so that the model would be corrected exactly to the target value T by V+D=T or by V*F=T. The mapping B then consists of pairs (P, D) or (P, F) and provides the correction value that is applied to V.

[0052] To ensure that these point-by-point corrections can also smoothly influence the neighborhood of known points, the images A and B must be interpolable. Topological methods such as "natural neighbor interpolation" or "thin plate splines," which can be configured with arbitrary points, are possible here. However, the most efficient and easiest to control are regular rectangular u-dimensional grid tables, as established in color transformations, for example, in ICC profiles, or combined collections thereof.

[0053] Application examples are described and illustrated below: Example 1: The aim is the color transformation of 7 printing ink values ​​into spectral reflectance values.

[0054] For this purpose, there is a complex (e.g., physical) prediction model that can calculate reflectance values ​​for any overpressure combination. However, it is too slow to be applied to millions of pixels in print files, and it is also too time-consuming to populate a detailed 7-dimensional table with predictions, as such a table would be enormous.

[0055] Therefore, this prediction model only populates a collection of detailed tables for all 4-dimensional subspaces, as this covers the commonly occurring, important color combinations in printing. Additionally, there is a component that can quickly (but roughly) estimate queries with 5 to 7 colors from the known 4-dimensional tables. The collection plus the estimation component results in model M.

[0056] Since this estimate is a kind of extrapolation (into untabulated additional dimensions), it is advantageous to check it selectively. One can now add a collection of coarsely graded 5-dimensional subspace tables as Figure B. Here, too, there is an estimation component for queries with 6 or 7 colors. For each entry P, the correction D or F is calculated from the estimated result V from M relative to the target value T from the prediction model. This already improves 5-dimensional queries. (For 4-dimensional entries, one can directly set the correction D=0 or F=1).

[0057] As a further correction shell, another figure B2 can be supplemented with a 7-dimensional table that only contains the vertices (0 or 100%), i.e., only 2 7< entries. These few can still be easily calculated with the prediction model. This fully captures free extrapolation and delivers meaningful, controlled results for all queries.

[0058] Example 2 ( Figure 1 and 2 ): The goal is to adapt a physical prediction model to measured values ​​from a test chart.

[0059] A physical prediction model describes the optical behavior of colors printed on top of each other or mixed together. The printing substrate serves as the basis. The properties of the individual colors are characterized, for example, by effective halftone dot area ratios, layer thicknesses, and transmission and scattering spectra, using color patches consisting only of gradations of individual printing colors. The ink transfer from ink to paper and from ink to previously printed ink is process-dependent. This so-called ink acceptance behavior is usually determined using color patches consisting of a pair of colors applied over the entire surface. This essentially physically captures the printed image. The reflection spectrum can then be calculated for any color combination.

[0060] However, this model is only a simplified representation of reality. Therefore, one would like to smoothly adjust the model to the actual measured values ​​using additional color patches. Established test charts exist for this purpose, such as the ECI2002 test chart for CMYK printing processes. This test chart includes, among other things, all CMYK combinations of 0, 40, or 100% tonal values. Using these 3 4< = 81 patches, an image A and an image B can be adjusted.

[0061] Image A corresponds to an interaction in the halftone that is not adequately described by the model. For example, an overprint of 40% cyan followed by 40% magenta could be more reddish in the model than in reality, because in reality, the magenta dots adhere less well to the cyan dots than would be the case with full-area ink transfer. In this case, the input dot (40,40,0,0) would be mapped to a modified dot (40,36,0,0). The physical interpretation could be that magenta is transferred 10% less strongly to cyan in the halftone.

[0062] After this pre-fitting, Figure B takes over the residual error of the model, so that overall the model exactly reproduces the measured values ​​on the (0,40,100) 4< grid.

[0063] This adjusts the model at key points. The ECI2002 test chart also contains even finer CMYK combinations, although these are incomplete in the black direction, so some target values ​​are missing. Next, you can use the combinations of CMY (0, 20, 40, 70, 100) with K (0, 20, 40, 80, 100) and place another correction shell around the model. The previously corrected (0, 40, 100) combinations will then receive a zero correction in the outer shell. The new steps will only require a relatively small correction because the inner (0, 40, 100) shell has largely adjusted the intermediate tones through interpolation of the corrections. The missing target values ​​leave "holes" in correction tables A and B, which are then smoothly filled by interpolating set neighbors.Furthermore, a third tray can be filled with the finest gradations CMY (0,10,20,30,40,55,70,85,100) and K (0,10,20,40,60,80,100) and thus over 90% of the color fields of the ECI2002 test chart can be reproduced exactly.

[0064] Finally, a concrete application case is described using subspace tables that store correct target values ​​for input data with up to k (4) color components, and for which an estimation method provides estimated target values ​​for input data with more than k color components (up to n).

[0065] Given a transformation to be corrected, represented by subspace tables, the transformation yields a "result" of m values ​​for input color data consisting of n color components. This can be achieved for any input (combinations of components between 0 and 100% of the n colors) by interpolation.

[0066] For a set of inputs, the correct results exist. These are correct, verified results that should match those generated by the transformation. The results were obtained, for example, by printing and measuring the input color data or by transforming it with another model with good results.

[0067] The following is an example of a concrete calculation for subspace tables with k=1, only one color, for a total of n=2 colors. The target color space is Lab, meaning the m=3 values: brightness, reddish value, and yellowish value. Subspace table 1: for printing ink yellow 0% yellow 95 0 0 (very bright, not reddish, not yellowish, unprinted paper) 100% yellow 90 0 100 (light, not reddish, very yellowish) Subspace table 2: for printing ink magenta 0% Magenta 95 0 0 (very bright, not reddish, not yellowish, unprinted paper) 100% Magenta 55 70 0 (medium brightness, very reddish, not yellowish)

[0068] The correct target value for the input 100% yellow + 100% magenta is also known: 100% yellow + 100% magenta 54 65 62 (by model or pressure+measurement)

[0069] The transformation delivers 100% yellow + 100% magenta 52 67 64

[0070] The difference to the target value is 100% yellow + 100% magenta +2 -2 -2 (make it brighter, less reddish+yellowish)

[0071] An optimization process would now reduce the color components of yellow and magenta. The result could be exemplary 98% yellow + 96% magenta 53 64 62

[0072] The difference to the target value is minimized, but not zero. Even less magenta would make it brighter, but it would further reduce the redness, which is already too low. 98% yellow + 96% magenta -1 -1 0 (still too dark, not red enough)

[0073] The difference for optimized input values ​​is assigned to the original input value, 100% yellow + 100% magenta -2% Yellow -4% Magenta

[0074] By applying this difference to the result, the transformation for this input can be corrected.

[0075] A different input, such as 50% yellow + 50% magenta, is far from the known input value. A smaller correction is probably more effective. It could, for example, fade with distance.

[0076] The two tables from the set of inputs and their input differences and target value differences could then be as follows: Yellow magenta ΔYellow ΔMagenta ΔL Δa Δb 0% 0% 0 0 0 0 0 100% 0% 0 0 0 0 0 0% 100% 0 0 0 0 0 100% 100% -2 -4 -1 -1 0

[0077] You can interpolate in this: 50% 50% -1 / 2 -1 +1 / 4 -1 / 4 0 (Contribution here is 1 / 4 each)

[0078] This is the correction to be applied to both the input and the result of the transformation (estimation procedure). The result would first be calculated as follows: The input values ​​for yellow and magenta are corrected from 50% + 50% to 49.5% + 49%. In subspace table 1: interpolated for printing ink yellow 49.5% Yellow 92.5 0 50 (very light, not reddish, moderately yellowish) In subspace table 2: interpolated for printing ink magenta: 49% Magenta 75 35 0 (rather light, moderately reddish, not yellowish)

[0079] The estimation method provides by combination (from XYZ color space) 49.5% Yellow + 49% Magenta 73 34 41

[0080] Final result after correction of the target values: 50% yellow + 50% magenta 73.25 33.75 41 (interpolated Δ were added)

[0081] The invention describes a solution that is practical and implementable for a person skilled in the art, with which a sufficiently fast and high-quality color space transformation into the color space of the specific printing system is possible even with a very high resolution of a layout and a large number of input colors.

[0082] Further advantages and features are evident from the following description based on the figures. They show: Figure 1 shows the source- and target-side correction of a transformation based on target values. Figure 2 shows the nesting of corrections until the desired accuracy is achieved.

[0083] Figure 1shows the steps of the correction process with iterative adaptation of the source data and subsequent residual correction of the target data for a selected color data set Q1 of the n colors of the source color space (101) with the color components of the colors q(1), q(2), ..., q(n), for which a verified, secured target value data set Z1 of the m values ​​of the target color space (102) with the components z(1), z(2), ..., z(m) is available, for example by measurement. In step 103, the current source color data set Q2 is initialized with the start value (101). This is transformed in step 104 using the provided transformation rule, for example a model. The data set Z2 of the m values ​​of the target color space 105 with the components z'(1), z'(2), ..., z'(m) is generated.

[0084] In step 106, the differences of the individual values ​​from Z1 and Z2 are calculated. These differences are checked in 107 to determine whether they are sufficiently small. If not, the color data set Q1 from step 103 is modified in step 108 to generate the color data set Q2. The modified color data set is transformed in step 104, and steps 105 to 107 are repeated. The steps are meant to be schematic; the concrete processes are part of the optimization method used, in particular how the modification is determined in step 108. If the differences are ultimately minimal, the differences between the source color data sets Q2 and Q1 are calculated in step 109. They are managed, for example, for a large number of pairs of Q1 and Z1 and their differences Z1 and Z2, as well as Q2 and Q1, in transformation tables and are interpolatable.The source differences are applied to the input data in the overall transformation 110, followed by the original transformation 104 into the target color space and the final correction of the residual error 106.

[0085] Figure 2 refers to Example 2 in the description starting on page 15. It describes the multiple hierarchically divided correction of a TRV 201 model using measurement data from a printed CMYK test chart 202. For the first correction at the corners of the CMYK color space, the 2^4 = 16 combinations of 0 and 100% and their measurement data are taken in step 203. For each combination (204), the source and target corrections are made as in Figure 1 determined (205) and tabulated, resulting in tables A1 (206) and B1 (207) with 16 entries each. These are combined with the model transformation rule TRV to form a new, corner-corrected transformation TRV1 (208).

[0086] The second correction uses the CMYK combinations of 0, 40, and 100% (209) available in many test charts and their measured values. For each of the 3^4 = 81 combinations (210), the corrections for TRV1 are determined (211) and stored in tables A2 (212) and B2 (213). Since the corners were already correctly corrected in TRV1, these 16 of the 81 correction entries are zero. A2 and B2 are placed around TRV1 as the next correction shell, so that there are now two shells around TRV (214).

[0087] The TRV2 (214) model corrected in this way is accurate at all corners and in the midtones and requires only minor adjustments. Further trays, in the case of the ECI2002 or IT8.7 / 4 test charts, use the combinations 0, 20, 40, 70, 100 for CMY and 0, 20, 40, 60, 100 for K, followed by 0, 10, 20, 30, 40, 55, 70, 85, 100 for CMY and 0, 10, 20, 40, 60, 80, 100 for K. This optimally utilizes the grid structure of these standard test charts. With a model corrected in this way, a project's CMYK print data can be spectrally simulated and subsequently displayed on screen or on inkjet proofing systems using standard methods.

Claims

1. Method for the computer-aided execution of transformations of colour data from a source colour space into a target colour space using a given transformation rule TRV, wherein the source colour space comprises n colours which are present in each dot to be printed in combinations of colour proportions q(1) to q(n), the target colour space comprises m values which can be combined to form combinations of components z(1) to z(m), and there are at least some combinations of the colour components q(1), q(2), ..., q(n) of the source colour space, for which combinations of components z(1), z(2), ..., z(m) of the target colour space are assigned, characterised by the following steps: a) selecting a set of colour data Q1 of the n colours of the source colour space with the colour components q(1), q(2), ..., q(n), for which there is a set of target value data Z1 in the form of combinations of components z(1), z(2), ..., z(m) of the target colour space; b) generating a second set of colour data Q2 of the n colours of the source colour space with the colour components q'(1), q'(2), ..., q'(n), which are initially set identically to Q1; c) transforming the set of colour data Q2 using the transformation rule TRV to generate an actual value data set Z2 of the m values of the target colour space z'(1), z'(2), ..., z'(m); d) applying an n-dimensional optimisation method which, by suitably modifying the set of colour data Q2 and repeating step c), minimises the differences between target value data set Z1 and actual value data set Z2 in terms of the least squares error, so that Q2 finally contains the resulting optimised colour components q'(1), q'(2), ..., q'(n); e) calculating the differences Dq(1) = q'(1)-q(1), Dq(2) = q'(2)-q(2), ..., Dq(n) = q'(n)-q(n) of each colour component of the n colours of the optimised colour data set Q2 to the respective colour component of the colour data set Q1; f) calculating the differences Dz(1) = z(1)-z'(1), Dz(2) = z(2)-z'(2), ..., Dz(m) = z(m)-z'(m) of each component of the colour data set Z1 to the respective component of the colour data set Z2; g) applying the differences Dq(1), Dq(2), ..., Dq(n) and Dz(1), Dz(2), ... , Dz(m) when transforming sets of colour data from the source colour space with n colours to the m values of the target colour space by adding the difference Dq(i) to the colour data q(i) of the n colours of the source colour space for the i=1 to n values when transforming a data set, the colour data q'(i) modified in this way are transformed into the m components z'(j) for the j = 1 to m values of the target colour space using the transformation rule TRV and then the corresponding difference Dz(j) is added to the respective component z'(j).

2. Method according to claim 1, characterised in that, to determine the target value data set Z1, the data set Q1 is printed out and the target value colour data set Z1 is determined by spectral measurement.

3. Method according to one of the preceding claims, characterised in that in step d) only those colour components of the n colours are changed which are not zero.

4. Method according to one of the preceding claims, characterised in that the method according to claim 1 is carried out for several sets of colour data Qj and in that the determined differences Dq(i)j and Dz(i)j for several sets Qj are managed in transformation tables and, in the case of transformations in step f), the differences to be applied are interpolated in these transformation tables.

5. Method according to one of the preceding claims, characterised in that the method is carried out on a computer unit by means of control software, wherein the computer unit comprises an input unit for providing the digital colour data of the source colour space of the project and an output unit for outputting the transformed values of the target colour space as well as a memory on which transformation tables are stored, wherein values for the target colour space are generated by means of the control software for input colour data of the project using the transformation tables and are provided in a data set.