METHOD AND DATA PROCESSING DEVICE FOR COMPUTER-AIDED DETERMINATION OF HAIR COLOR PROPERTIES

Description

[0001] The invention relates to a method for computer-aided determination of the properties of hair colors according to claim 1 and a data processing device for carrying out the method according to claim 8.

[0002] Hair dyes can have properties such as colorfastness, wash fastness, lightfastness, gray coverage, or other characteristics. Hair dyes are created by coloring hair with a hair dye, also known as the dyeing process.

[0003] Hair dyes can contain a mixture of different dye precursors and can therefore also be referred to as dye mixtures.

[0004] Predicting the aforementioned properties of hair colors, i.e., determining an expected coloring result without actually carrying out the coloring process, can be used commercially in many ways, for example, when calculating a color result based on an individual starting hair color, when calculating optimal colors, e.g., parameters of a hair color parameterized in a color space, for presentation, e.g., on packaging, in advertisements (e.g., commercials), on the internet and in apps, when compiling an individual ("customized") hair dye, and for subsequent recipe optimization in product development, for example, with regard to the wash fastness, light fastness, and / or gray coverage of a hair color.

[0005] While it is possible to calculate the exact color to be reproduced in related areas of color production, for example in precise photo printing using calibrated pigment printers, this has not been possible in the field of hair colors until now.

[0006] One main reason for this is that hair coloring, i.e., the creation of a hair color, often doesn't involve dyes directly, but rather dye precursors. During a coloring process, a multitude of different dyes can form, whose colorimetric properties as pure substances may not be fully known. Furthermore, the concentrations of the dyes in the colored hair may be unknown, and it may also be unknown which concentration of dyes in the colored hair corresponds to which concentration of dye precursors in the hair dye. This can be at least partially due to the fact that the dye precursors interact with each other during the formation of the various dyes.

[0007] Therefore, calculating the reflection spectra of dyed hair has not been possible until now.

[0008] Known methods for predicting, e.g., calculating, hair color utilize a principle of comparative dyeing and transferring the results obtained to hair with a similar starting hair color. See, for example, FR 2984569 B1, JP 2007-212140 A, JP 2004-144569 A, or WO 2001 / 87245 A2.

[0009] However, these methods are not suitable for calculating a color, or other properties, that would result from dyeing with an unknown, never-before-dyed mixture.

[0010] DE 2007 050434 A1 describes a method and arrangement for computer-aided determination of at least one property of a hair coloration based on a formulation of chemically reactive and / or unreactive raw materials.

[0011] In various embodiments, a method for computer-aided determination of the properties of hair colors is provided, whereby the determination may also be possible for hair colors that are achieved with a dye mixture that has not previously been dyed (or whose properties are not already known).

[0012] In various implementation examples, methods from the field of predictive analytics, for which the corresponding English term "Predictive Analytics" is usually used (also known as "Big Data", "Data Mining" or "Machine Learning"), are used to enable precise calculations of properties, such as the properties mentioned above, of hair colors despite potentially many unknowns in a dye mixture.

[0013] According to various embodiments, it is possible to provide a data set (also referred to as hair color data) by means of test dyeings, which has a plurality of dyeing prerequisite parameters and at least one dyeing result parameter for each of the test dyeings.

[0014] The majority of dyeing prerequisite parameters include at least two concentrations of dye precursors, and the majority of these parameters also include a base hair color, which is parameterized in a color space. Furthermore, the majority of these parameters also include a pre-existing hair damage. However, the dataset may also include other dyeing prerequisite parameters, such as concentrations of other dye precursors, other ingredients of a hair dye, a degree of hair graying, and / or other dyeing prerequisite parameters.

[0015] At least one coloring result parameter specifies a hair color that is parameterized in a color space.

[0016] In this context, "color" can be understood as an interaction of a hue (i.e., a spectral color impression, also referred to as hue, which can be understood as what is regarded as the "actual color"), a color intensity (i.e., how intense the color appears, e.g., compared to a neutral gray, which is also referred to as saturation, color saturation, hue, chromaticity, chromacity, or color depth), and a brightness (i.e., how light or dark the color appears).

[0017] In various embodiments, the color information can, for example, have a parameterization in a known color space, such as an L*a*b* color space (where L* indicates the brightness of a color, a* the green and red components, and b* the blue and yellow components of the color; sometimes the abbreviated notation Lab or individually L, a, and b is used here), an RGB color space by color components in red, green, and blue, a CMYK color space by color components in cyan, magenta, yellow, and black, or any other color space.

[0018] The term "hue" can be understood here, as described above, as the spectral color impression of a color, regardless of how it may be parameterized, for example as a point in a two-dimensional color space (e.g. a*b* of the L*a*b* system) or a ratio of color components (such as in the RGB color space or the CMYK color space).

[0019] In various embodiments, a color space from which the color information (e.g., the hair color information of the dyed hair or the hair before dyeing, also referred to as the base hair color) originates, or in which the color information is represented (for example, when a hair color is displayed, see below), can be designed such that a determined or displayed color is independent of the medium by which the color is determined or displayed (e.g., colorimeter, screen, printer, scanner, human eye, etc.). The color space can, for example, be an L*a*b* color space, and the color information a hue parameterized, for example, by a* and b*.The uniform representation in the medium-independent color space can, for example, make it possible to present a realistic expected coloring result, for example, by ensuring that a color achieved through dyeing leaves the same color impression on the viewer of the dyed hair as in a representation of the expected result, such as a packaging print, a display on a computer screen, etc.

[0020] The at least one coloring result parameter can, in various embodiments, also have further properties of the dyed hair color, for example, lightfastness, washfastness or the ability to cover grey.

[0021] The data set is used as the basis for applying a predictive analytics method.

[0022] For example, the staining prerequisite parameters or a part of the staining prerequisite parameters and the staining result parameters or a part of the staining result parameters associated with them can be used to create a model that describes the data set as accurately as possible.

[0023] In various implementation examples, the measurement data of the dataset, i.e., the measured values ​​of the hair color data, which describe the properties of the hair color produced by dyeing (e.g., L*, a*, b* for the color, wash fastness, light fastness, gray coverage, etc.), can be dependent variables. Using a complex mathematical model, which can be developed using predictive analytics, the dependence of the dependent variables on the independent variables (for example, the concentrations of the dye precursors, such as the concentrations present on the head) can be modeled. That is, using predictive analytics, a relationship between the independent and dependent variables (in other words, between the dyeing prerequisite parameters and the dyeing result parameters) can be determined. For color, this can be expressed, for example, as: L * a * b * = f c 1 , c 2 , c 3 , . . . , c n , where L*a*b* represents the color parameters, and ci (i=1,...,n, n>1) represents the concentrations of dye precursors. The function may or may not be analytically known. If no analytical function is known, the values ​​of the dependent variable (the staining result parameters) can also be calculated using numerical algorithms.

[0024] In addition to dye precursors, dyes can also be used in hair dyes. Accordingly, ci can also stand for ci concentrations of dyes.

[0025] Possible independent variables can be metric (cardinal), ordinal, or categorical.

[0026] In various embodiments, the independent variables (the dyeing prerequisite parameters) can be properties that influence the dyeing result, for example the concentration of one of the respective dye precursors, the base hair color, a state of damage and / or a degree of graying of the hair, or similar.

[0027] In various embodiments, predictive analytics can be used to create a model which, given given staining prerequisite parameters (independent variables, examples see above), predicts the staining result parameters (dependent variables, examples see above) as accurately as possible.

[0028] In various implementation examples, predictive analytics can identify independent variables that have no or only a negligible influence on the model. In other words, independent variables (dyeing prerequisite parameters) may be present in the hair color data that were assumed to influence the dependent variables (dyeing result parameters), even though this is not the case or only a negligible influence. These negligible variables can be identified using predictive analytics and, if necessary, disregarded in subsequent modeling with comparable prerequisites to improve model accuracy.

[0029] Using predictive analytics, a continuous model of the staining prerequisite parameters and the staining result parameters is generated, making it possible to determine a value for a staining result parameter for a value of a staining prerequisite parameter or a combination of values ​​for a plurality of staining prerequisite parameters that does not correspond to any of the corresponding experimental values ​​or combinations of values.

[0030] It is also possible to model categorical properties, such as "good" or "bad".

[0031] Predictive analytics can be generally described as a method for extracting information from large datasets and creating a model that allows predictions to be made even for samples not included in the dataset. When applying a predictive analytics method, a portion of the dataset is typically used as a training dataset (also called training set or training data). Based on this training dataset, one or more models can be created and then tested on the data not included in the training dataset, on the entire dataset, or on a specifically selected subset of the data.

[0032] For an evaluation of the model, i.e., a determination of the goodness of fit, for example a coefficient of determination R 2< , a mean absolute error, a mean squared error, a standard deviation and / or a mean deviation can be used.

[0033] The coefficient of determination R²< can correspond to a squared correlation coefficient for a linear regression model. For a different model (a different relationship), it may be defined differently.

[0034] For predictive analytics modeling, various functions or methods can be used, depending on the specific implementation. In a simple case, for example, multiple (linear) regression can be employed. Better results can typically be achieved using polynomial regressions, neural networks, support vector machines, decision trees (e.g., tree ensembles), or similar techniques.

[0035] In various embodiments, the described method for computer-aided prediction of hair color properties can be carried out using a data processing device.

[0036] The data processing device may, for example, be a computer, or any other data processing device suitable for storing and providing the data and for performing the predictive analytics procedure, such as any data processing device with a sufficiently large data storage capacity and a sufficiently powerful processor.

[0037] In various embodiments, the data processing device can have at least one input device for entering information into the data processing device, for example, for entering hair color data and, if necessary, for entering instructions, parameters, etc. for executing the method.

[0038] In various embodiments, the data processing device can have at least one output device for outputting information, for example, for outputting results of the process.

[0039] In various embodiments, the at least one output device can include a screen and / or a printer.

[0040] In various embodiments, for example, when the output dyeing result parameters include a hair color, the color can also be parameterized for output in a medium-independent color space, such as the L*a*b* color space. This allows the expected dyeing result, determined as described above and displayed on a screen or printed (e.g., on the packaging of a dyeing product), to appear essentially as it would after actual dyeing. If the output device requires a different color parameterization, the determined color can be transformed from one color space to another.

[0041] A method for the computer-aided determination of hair dye properties according to claim 1 is provided. The method includes, among other things: providing hair dye data, wherein the hair dye data for a plurality of dyeing processes each have values ​​for a plurality of dyeing prerequisite parameters and for at least one dyeing result parameter, wherein for each dyeing process of the plurality of dyeing processes the plurality of dyeing prerequisite parameters has a first concentration of a first dye precursor and a second concentration of a second dye precursor, wherein for each dyeing process of the plurality of dyeing processes the at least one dyeing result parameter has a measured value about a property of hair dyes, and determining a relationship between the plurality of dyeing prerequisite parameters and the at least one dyeing result parameter by means of predictive analysis based on the hair dye data.

[0042] The procedure further involves: Determining, by means of the determined relationship, a value for a dyeing result parameter to a combination of values ​​from one value each for a selected plurality of dyeing prerequisite parameters, wherein the combination of values ​​for none of the dyeing processes is identical to its combination of values ​​for the selected plurality of dyeing prerequisite parameters.

[0043] In various embodiments, predictive analytics can utilize at least one method from a group of methods, where the group of methods may include: linear or multi-linear regression, polynomial regression, multiple polynomial regression, neural network methods, support vector machine methods, and decision tree methods (including tree ensembles).

[0044] In various embodiments, the decision tree method can utilize decision tree ensembles.

[0045] Advantages of decision tree ensembles in this method are that they provide more accurate predictions and require less computational effort than other predictive analytics methods. This can be beneficial when multiple computer-aided methods for determining hair color properties are to be run in parallel on a single server.

[0046] Furthermore, the majority of coloring prerequisite parameters have a base hair color, whereby the base hair color is parameterized in a color space.

[0047] Furthermore, the majority of dyeing prerequisite parameters indicate pre-existing damage to the hair.

[0048] Pre-existing damage can be characterized by physical quantities, such as the modulus of elasticity, or chemical quantities, such as the cysteine ​​acid content.

[0049] In various embodiments, the majority of dyeing prerequisite parameters can also exhibit a degree of graying.

[0050] In various embodiments, the at least one coloring result parameter can have at least one property of hair colors, wherein the properties of hair colors can have: one or more parameters of a hair color parameterized in a color space, a wash fastness, a light fastness, and a gray coverage capability.

[0051] In various embodiments, the at least one dyeing result parameter can have a plurality of dyeing result parameters.

[0052] A data processing device according to claim 8 is provided for performing a computer-aided determination of the properties of hair colors, wherein the data processing device is configured to perform the method according to claim 1.

[0053] The data processing device includes a processor, the processor of which may be configured to perform the determination of the relationship.

[0054] Exemplary embodiments of the invention are shown in the figures and are explained in more detail below. They show

[0055] Figure 1 is a schematic representation of a predictive analytics method according to the CRISP model; Figure 2 is a flowchart illustrating functional components and their interaction in an implementation of a method for the computer-aided determination of hair color properties according to various embodiments; Figure 3 is a diagram illustrating a result of a method for the computer-aided determination of hair color properties according to various embodiments; Figure 4 is a flowchart illustrating functional components and their interaction in an implementation of a method for the computer-aided determination of hair color properties according to various embodiments; Figures 5A and 5B are two diagrams illustrating, in different ways, a result of a method for the computer-aided determination of hair color properties according to one embodiment.Figures 6A and 6B are two diagrams that, in different ways, represent a result of a method for the computer-aided determination of hair color properties according to an embodiment; Figure 7 is a diagram that represents a result of a method for the computer-aided determination of hair color properties according to an embodiment; Figure 8 is a diagram that represents a result of a method for the computer-aided determination of hair color properties according to an embodiment; Figure 9 is a diagram that represents a result of a method for the computer-aided determination of hair color properties according to an embodiment; Figures 10A, 10B, and 10C are three diagrams that represent the relationship between measurement points, true function, and model for different models used in a method for the computer-aided determination of hair color properties according to different embodiments;Figure 11 shows a flowchart illustrating a method for computer-aided determination of hair color properties according to various embodiments; Figure 12 shows a schematic representation of a data processing device according to various embodiments; and Figure 13 shows hair color data according to one embodiment.

[0056] The following detailed description refers to the accompanying drawings, which form part of this application and illustrate specific embodiments in which the invention can be implemented. It is understood that other embodiments may be used and structural or logical modifications may be made without altering the scope of protection of the present invention. It is understood that the features of the various exemplary embodiments described herein may be combined unless specifically stated otherwise. Therefore, the following detailed description is not to be interpreted as restrictive, and the scope of protection of the present invention is defined by the attached claims.

[0057] Figure 1 Figure 100 shows a schematic representation of a predictive analytics procedure.

[0058] The terms Predictive Analytics, Big Data and Data Mining are used synonymously here.

[0059] As a starting point for any predictive analytics method, as described in Fig. 1 The illustration shows an understanding of the business, for example, what is to be achieved with the predictive analytics method.

[0060] Predictive analytics requires data to function. This data may already be available or may need to be collected in a targeted manner. Understanding which data can be helpful for the defined goal and what information is contained within this data can be useful, for example, before preparing the data for use in predictive analytics, such as before the data is entered into a computer program to execute the procedure.

[0061] Subsequently, modeling can be carried out using predictive analytics. A subsequent evaluation of the result can, for example, lead to its immediate application, or it can influence the understanding of the business, for instance, by revealing that further parameters should be included, or that the task cannot be solved in its current form.

[0062] In Fig. 1 Frame 110 indicates which parts of such a predictive analytics method the computer-aided determination of hair color properties according to various embodiments can influence when the data set is available, namely the modeling and the evaluation.

[0063] Table 1 and Table 2 present two exemplary hair color data sets (or excerpts from a more extensive data set of hair color data).

[0064] To generate measured data for the hair colors, the respective hair dyes (the product listed in column 1, 74 different products (formulations) in Table 1 and 53 different products (formulations) in Table 2) were dyed according to the instructions for use (on Kerling Euronaturhaar white) and measured colorimetrically (light type D65 / 10, diffuse with gloss, 8°).

[0065] In Table 1, column 2 shows the immediate result of the colorimetric measurement as a parameter in the L*a*b* color space.

[0066] In Table 2, colorimetric measurements were performed twice: once immediately after dyeing and once after 12 washes. The resulting color difference ΔE 00 is listed in column 2 of Table 2. This color difference can be used as a measure of the wash fastness of the hair color.

[0067] Columns 3 and 4 of Tables 1 and 2 list the concentrations of two dye precursors (p-toluenediamine sulfate and m-aminophenol) as they would appear upside down. Concentrations were determined for a total of twenty dye precursors, of which only two are shown here as examples.

[0068] Table 2 is also in Fig. 13 The exemplary hair color data 1300 shows, for a number of coloring processes (two of which are marked as examples with 1310), a number of coloring prerequisite parameters 1330 (the concentrations of the two dye precursors) and one coloring result parameter 1320 (the color difference after 12 hair washes). An example value for one of the coloring prerequisite parameters is marked with 1331. An example value for the coloring result parameter is marked with 1321.

[0069] The data presented in the tables are intended solely as illustrative examples. The specific hair color measurement data generated and used in different implementations depends on the application. For instance, different, fewer, or more products may be used, which may contain different, more or fewer dye precursors. Instead of color or color difference, other parameters may be determined and included in the hair color data, such as lightfastness, base hair color, etc.

[0070] Providing hair color data as a table is merely an example. The hair color data can be provided in any format that allows for the mapping of coloring result parameters to their respective corresponding coloring prerequisite parameters and enables computer-aided use of the hair color data. Table 1 product L ab p-Toluene diamine sulfate [µmol / 100g] m-Aminophenol [µmol / 100g] N&E 542 Medium Ash Blonde 40,4 3,8 11,5 859,5 130,6 N&E 545 Medium Golden Blonde 38,7 7,3 12,7 1369,6 75.5 N&E 550 Dark Blonde 42,7 4 0 15,2 862,4 66.4 N&E 555 Dark Golden Blonde 38,5 7,6 13,0 089.8 72,9 N&E 557 Multi-Reflex Brown 31,7 8,9 13.7 1859,6 0,0 N&E 550 Light Brown 26,5 3,5 7,6 2117,1 400.9 N&E 562 Light ash brown 29,5 3,2 7,4 1505,7 184.5 M&E 555 Light Golden Brown 33,4 5,5 11,4 1811,S 166,3 N&E 566 Cinnamon Golden Brown 25,2 6,7 10,0 3453,3 234,0 N&E 568 Intense Red 31,9 30,8 20,6 0,0 2000,1 N&5 570 Medium Brown 27,0 3,7 9,6 2850,1 422,7 N&E 574 Dark Chocolate 25,9 3,0 10,1 3861,5 0,0 N&E 575 Chestnut Reddish Brown 25.5 8,3 7,7 2810,3 304,3 N&E 580 Dark Brown 20,8 2,1 4,5 3758,4 989,7 N&E 584 Mocha Chocolate 23,7 6 5 7,3 3595,9 334,8 N&E 585 Multi-Reflex Brown 23,3 8,3 5,2 2923,5 229,1 N&E 586 Cinnamon Dark Brown 23,2 7,1 8,7 5616,1 0,0 N&E 588 Glossy Acaiberry 22,9 15,5 2.9 1999,3 0,0 N&E 590 Black 15,3 0,1 -1.7 8475,1 918,4 Nectra 688 25,6 10 4 6,6 0,0 2776,8 Nectra 499 20,7 11,6 -2,0 447,2 336,7 Nectra 568 29,6 10,4 6,6 1804,3 401,7 Nectra 400 21.9 1,82,7 4133,8 1093,5 Syoss Oleo 3-10 20,3 2,3 2.7 3759,4 938,7 Syoss Oleo 5-92 30,5 31,7 13,2 0,0 1963,3 Nectra 320 19,2 1,3 1.9 5650,4 1484,6 Nectra 468 29,2 11,5 10,5 2338,3 0,0 Colour cream 1 19,3 6,4 -2,9 1001,1 0,0 Colour cream 2 20,4 7,3 -4.6 998,& 0,0 Color cream 3 19,6 5.8 -7,1 998.9 0,0 Colour cream 4 22,0 15,7 2.6 497,2 0,0 Colour cream 5 20,3 9,4 -3,5 501,7 0,0 Colour cream 6 22,6 10,5 -5,1 499,4 0,0 Colour cream 7 20,4 3,3 -4,5 751.4 0,0 Colour cream 8 23,3 10,1 -4,8 499,4 0,0 Colour cream 9 22.5 3,3 -9,3 499,4 0,0 Colour cream 10 27,8 19,8 2,6 249,7 0,0 Colour cream 11 24,5 13,5 -5,1 252,0 0,0 Colour cream 12 30,1 13,7 -5,8 249,7 0,0 Colour cream 13 25,2 12,5 -4,8 376.3 0,0 Colour cream 14 25,2 12,5 -5,2 333,7 0,0 Colour cream 15 29.1 12.7 -5.0 333,7 0,0 Colour cream 16 27,9 9,1 -9,7 333,7 0,0 Colour cream 17 25.1 20,7 2.6 165,7 0,0 Colour cream 18 26,0 15,1 -5.5 168,0 0,0 Colour cream 19 35,1 14,4 -4,1 165.7 0,0 Cashmere red variant 2 31,4 36,1 22,6 OC 0,0 Syoss Color 2014 5-22 25,3 25,3 12,5 491,7 2812,0 Syoss Color 2014 1-4 15.0 12 -3,3 4994,3 0.0 Syoss Color 2014 5-28 23,2 9,1 9,6 1756,7 166,0 Syoss Oleo 4-29 25,7 26,0 13,7 681,3 227,7 Syoss Oleo 1-40 15,3 1.4 -3,3 4994,3 0,0 Igora Royal 5-88 25,1 26,5 12,6 481.7 2812,0 Nectra 600 29,8 4,6 10,0 1770,7 252,0 Nectra 662 29,3 6,7 9,4 2179,3 228,1 Nectra 777 45,3 29,7 33,2 0,0 620,0 Beats Royal 3-0 18,3 1,1 1,8 5788.6 1512,1 Leaves Royal 5-5 31,0 6,7 11,3 2043,1 165,0 Igor Royal 6-65 33,0 7,1 11,8 1793,4 183,3 Beats Royal 6-0 28,84,1 8,5 1922.6 440,3 Beats Royal 5-1 28.3 1,4 5,2 1566,4 247,4 Igora Royal 4-88 25.1 26,5 14,4 794.6 1374,6 Igora Royal 7-887 35,1 40,7 29,0 149.5 0,0 Syoss Color 2012 4-2 25.5 242 13.3 1589,1 458,2 Syoss Color 2012 5-22 3D,3 31.6 19,3 612.9 481,1 Syoss Color 2012 1-4 16,8 1,7 -5,4 4904,3 0.0 Syoss Color 2D12 5-29 33,5 36,1 22.7 0,0 2000,1 Syoss Color 2012 3-55 27,7 6,9 9,3 3904,7 412,4 Syoss Color 2012 5-0 25,6 3.0 6.6 3132.8 710,2 Brilliance 880 25,2 25 6.3 3408.6 740,9 Beats Royal 1-0 16,8 0,1 -1.6 7355.3 733,1 Syoss Oleo 2-10 15.7 1,1 0,3 5012,5 1319,6 Syoss Oleo 4-18 24,9 6,3 8,2 2042,8 131,5 Syoss Oleo 6-10 27,4 46 10,6 2144,4 163,1 Table 2 Product of 00 (12HW) p-Toluylene-diamine sulfate [µmol / 100g] m-Aminophenol [µmol / 100g] N&E 542 Medium Blonde 2,7 859,5 130,6 N&E 545 Medium Golden Blonde 1,4 1369,6 75,6 N&E 550 Dark Blonde 2,4 862,4 56,4 N&E 555 Dunkles Goldblond 2,7 989.6 72,9 N&E 557 Multi-Reflex-Braun 1,8 1959,6 0,0 N&E 560 Hellbraun 2,8 2117,1 400,9 N&E 562 Heilaschbraun 2.9 1505,7 184,5 N&E 565 Hellgoldbraun 2,0 1811,9 166,3 N&E 566 Cinnamon Golden Brown 1,6 3453,3 234,0 N&E 568 Intensive-Roth 3,4 0,0 2000,1 N&E 570 Mittelbraun 1,8 2650,1 422,7 N&E 574 Bittersweet Chocolate 1,6 3861,9 0,0 N&E 576 Kastanie Rotbraun 1,4 2810,3 304,3 N&E 580 Dark Brown 2,3 3759,4 989,7 N&E 584 Mocha Chocolate 1,4 3595,9 384,9 N&E 585 Multi-Reflex-Braun 1,9 2928,5 229,1 N&E 586 Cinnamon Dank Brown 1,3 5616,1 0,0 N&E 588 Glossy Acaiberry 3,1 1999,3 0,0 N&E 590 Black 0,5 6475,1 916,4 Nectra 688 2,2 0.0 2776,8 Nectra 499 1,9 4472 386,7 Nectra 568 1,9 1804,8 491,7 Nectra 400 1,8 4133,9 1090,5 Syoss Oleo 3-10 2,3 3759,4 989,7 Syoss Oleo 5-92 2,8 0,0 1963,9 Nectra 300 1,1 5650,4 1484,6 Nectra 468 1,4 2338.3 0,0 Kashmirrot Variant 2 2,8 0,0 0,0 Syoss Color 2014 5-22 2,8 491,7 2812,0 Syoss Color 2014 1-4 2,1 4994,3 0,0 Syoss Color 2014 5-28 1,7 1736,7 165,0 Syoss Oleo 4-29 4,4 681,3 227,7 Syoss Oleo 1-40 2,5 4994,3 0,0 Igora Royal 5-88 2,8 491,7 2812,0 Nectra 777 7,4 0,0 630,0 Igora Royal 3-0 2,2 5768,9 1512,1 Igora Royal 5-5 1,7 2043,1 165,0 Igora Royal 6-65 1,8 1793,4 183,3 Igora Royal 6-0 2,9 1922,6 440,3 Igora Royal 6-1 3,7 1566,4 247,4 Igora Royal 4-88 2,6 794,6 1374,6 Igora Royal 7-887 2,2 149,8 0,0 Syoss Color 2012 4-2 4,1 1589,1 458,2 Syoss Color 2012 5-22 2,8 612,9 481,1 Syoss Color 2012 1-4 2,4 4994,3 0,0 Syoss Color 2012 5-29 3,2 0,0 2000,1 Syoss Color 2012 3-65 1,3 3904,7 412,4 Syoss Cofor 2012 5-0 1,9 3132,8 710,2 Brillance 880 1,5 3408,6 740,9 Igora Royal 1-0 1,0 7355,3 733,1 Syoss Oleo 2-10 1,8 5012,5 1319,6 Syoss Oleo 4-18 2,6 2042,9 181,5 Syoss Oleo 6-10 2,4 2144,4 163,1

[0071] In the example shown in Table 1, the concentrations of the dye precursors (from columns 3 and 4 of Table 1) can represent two dyeing prerequisite parameters, i.e., independent variables for the predictive analysis method (actually, there would be more dyeing prerequisite parameters, approximately 100, corresponding to the approximately 100 dye precursors). The hair color data can have one value for each dyeing prerequisite parameter for each dyeing process; in this example, therefore, 74 values ​​(corresponding to the 74 test dyeings) for each dye precursor concentration.

[0072] The L*a*b* values ​​from column 2 of Table 1 can, for example, constitute staining result parameters, e.g., three staining result parameters L*, a*, and b*, or one (three-dimensional) staining result parameter. The staining result parameters can represent dependent variables for the predictive analytics procedure. One or more values ​​can be specified for each staining result parameter, for example, 74 values ​​corresponding to the 74 test stainings (also referred to as staining processes) for each staining result parameter. Each value of a staining result parameter corresponds to a row entry in the column corresponding to that staining result parameter. A staining prerequisite parameter can also have a value of zero. A staining result parameter can also have a value of zero.

[0073] The example shown in Table 2 is similar to that in Table 1; however, column 2 contains a color difference ΔE 00 (also denoted as dE 00) instead of a color. This represents the difference in the Lab color space that a hair color achieves after a specified number of washes (here, for example, 12; however, 24 or 36 washes are also common, and any other number of washes that provides a meaningful result would be possible). The color difference can serve as a measure of the wash fastness (also referred to as wash resistance) of the hair color. The smaller the achieved color difference, the more wash-resistant the color.

[0074] Based on provided hair color data (for example, as shown in excerpts in Table 1), predictive analytics can be used to determine a relationship between the coloring prerequisite parameters (in the present example 14, two of which are shown in Table 1 and Table 2 respectively) and the desired coloring result parameter (for example, L*, the brightness of the color, also known as luminance, is used here).

[0075] Alternatively, in various embodiments, it would be possible to use other dyeing prerequisite parameters, for example, to use only a portion of the dye precursor concentrations, to add others, to determine further parameters, for example, by measurements and add them to the hair color data (e.g., base hair color, degree of graying, pre-damage to the hair, etc.), and / or to use other dyeing result parameters, for example, the hair color (as a three-dimensional dyeing result parameter), the wash fastness (see the example from Table 2), etc., and to use them to determine the relationship.

[0076] In various implementation examples, any program that provides such functionality can be used for modeling using predictive analytics.

[0077] In the examples described below, predictive analytics was performed using the software KNIME 2.11.2.

[0078] Figur 2 Figure 200 shows a flowchart illustrating functional components (so-called nodes) and their interaction in the execution of a method for computer-aided determination of hair color properties according to various embodiments.

[0079] The functional components (nodes) can represent individual processes carried out during the execution of the predictive analytics procedure.

[0080] For example, node 1 represents a so-called XLS reader 210, which, according to various embodiments, can read hair color data. The hair color data can be in Excel format, for example, or can be read into such a format using the XLS reader 210. For instance, Table 1 could be read into four columns with 74 rows each (if the three color values ​​L*, a*, and b* are read into a single column), or into six columns with 74 rows each (if the three color parameters L*, a*, and b* are each read into a separate column). For the further description of the example, it is assumed that each of the color parameters L*, a*, and b* has been read into its own column.

[0081] In node 2, a column filter 220 can be provided according to various embodiments, which can be used to select columns to be used as dyeing prerequisite parameters and as dyeing result parameter(s). The columns can be selected after the hair color data has been read in. In the example from Table 1, the columns with concentrations of dye precursors can be selected as dyeing prerequisite parameters (independent variables), and the column with brightness L* as a dyeing result parameter. The selected data can also be referred to as a training set.

[0082] In node 10, a functional component 230 for performing a simple linear regression can be provided according to various embodiments. The simple linear regression can be performed after selecting the columns to be used for the regression. Using the regression, a relationship between the majority of staining prerequisite parameters and the at least one staining result parameter can be determined. This relationship can also be referred to as a model. One principle for fitting the model to the data can be least-squares error optimization. In other words, an error can be minimized using the least-squares method.

[0083] In various embodiments, a so-called Weka Predictor 240 can be provided at node 11, to which both the results of the simple linear regression from node 10 and the unused hair color data from node 2 can be fed. The Weka Predictor 240 can determine values ​​for the coloring result parameter (the brightness L*) for all selected dyeing prerequisite parameters (in the example from Table 1, the two concentrations of the dye precursors) using the previously determined relationship (of the model).

[0084] In various embodiments, a so-called numerical evaluator 250 can be provided at node 18. This evaluator uses the values ​​for the staining result parameter determined by the Weka predictor 240 and the actually measured staining result parameter(s) to calculate values ​​that can be used as a measure of the model's goodness of fit. For example, a coefficient of determination R²< (which corresponds to a squared correlation coefficient for linear regression; generally, good values ​​for the coefficient of determination R²< are approximately between 0.9 and 1.0), a mean absolute error, a mean squared error, a standard deviation, and / or a mean absolute deviation can be determined. In various embodiments, particular attention can be paid to the coefficient of determination R²< and the mean absolute deviation.

[0085] Such a method for computer-aided determination of the properties of hair colors according to various embodiments was carried out using the 14 dye precursors as dyeing prerequisite parameters and the brightness L of the hair color as a dyeing result parameter.

[0086] Figur 3 Diagram 300 shows a result of the procedure for computer-aided determination of hair color properties.

[0087] The model itself cannot be represented, as that would require a 15-dimensional representation. Therefore, in Figur 3 The modeled (predicted) brightness values ​​L are plotted as a function of the measured brightness values. Ideally, the data points would lie on a line with a slope of 1.

[0088] The exemplary values ​​determined for the goodness of fit are R 2<= 0.436, mean absolute error = 3.704, mean squared error: 21.542, standard deviation 4.641 and mean absolute deviation 0.

[0089] Even though the coefficient of determination R² may be low in this example and the mean absolute error may be quite high at 3.7, reasonably usable values ​​for the properties (brightness values) of unknown dye formulations can be determined using a relatively simple model (linear regression). To determine brightness values ​​for unknown dye formulations, these can be fed into the prediction node, also known as the predictor node.

[0090] In Figur 4 Figure 400 shows a flowchart which illustrates functional components and their interaction when carrying out a method for computer-aided determination of hair color properties according to various embodiments.

[0091] The in Figur 4 The flowchart shown, 400, can in many ways be compared to flowchart 200. Fig.2 correspond. For example, the XLS reader 210 in node 1 can be the same as the XLS reader 210 from Fig.2 Similarly, the column filter 220 in node 2 can be the same as the column filter 220 from Fig.2 .

[0092] In node 23, according to various embodiments, a functional component 330 can be provided to determine a relationship between the dyeing prerequisite parameters (the 14 dye precursors) and the dyeing result parameter (the brightness of the hair color L*) using a tree ensemble learner.

[0093] A tree ensemble learner (also known as Tree Ensemble Learner or Random Forest (protected term)), which can be used for categorical classification or regression, can be considered one of the most powerful algorithms in the field of predictive analytics.

[0094] In general, a Random Forest can be described as a classification method consisting of several different, uncorrelated decision trees. All decision trees can have grown under a specific type of randomization during a learning process. For classification purposes, each tree in this forest can make a decision, and the tree with the most votes determines the final classification. Besides classification, the Random Forest can also be used for regression analysis.

[0095] A procedure executed by the Random Forest (the Tree Ensemble Learner) to find the relationship between the staining prerequisite parameters and the staining result parameter can be intuitively described as finding relationships between different or partially overlapping subsets of the selected data. From the relationships found for the subsets, a relationship for all data selected at node 2 is determined or created, which best represents most of the relationships for the subsets. The relationships for the subsets can be determined, for example, using regression (e.g., linear or polynomial). This found relationship can also be referred to as a model.

[0096] In this example, default values ​​from the KNIME 2.11.2 software were used for the configuration of the Tree Ensemble Learner.

[0097] The application of the Tree Ensemble Lerner 330 can be performed after selecting (in node 2) the columns to be used for determining the relationship.

[0098] In various embodiments, a so-called Tree Ensemble Predictor 440 can be provided at node 24, to which both the results of the model creation (determining the relationship) from node 23 and the unused hair color data from node 2 can be fed. The Tree Ensemble Predictor 440 can determine values ​​for the coloring result parameter (the brightness L*) for all selected dyeing prerequisite parameters (in the example from Table 1, the two concentrations of the dye precursors) using the previously determined relationship (of the model).

[0099] In various embodiments, a Numerical Evaluator 450 (similar to the Numerical Evaluator 250 from) can be installed in node 22. Fig.2 ) be provided which, using the values ​​for the staining result parameters determined by the Baum Ensemble Predictor 440 and the actually measured staining result parameter(s), calculates values ​​that can be used as a measure of the model's goodness of fit. For example, a coefficient of determination R², a mean absolute error, a mean squared error, a standard deviation, and / or a mean absolute deviation can be determined. In various embodiments, particular attention can be paid to the coefficient of determination R² and the mean absolute deviation.

[0100] Such a method for computer-aided determination of the properties of hair colors according to various embodiments was carried out using the 14 dye precursors as dyeing prerequisite parameters and the brightness L of the hair color as the dyeing result parameter Z.

[0101] Figuren 5A und 5B Two diagrams, 500 and 501, show results of this procedure in different ways.

[0102] In Fig.5A The modeled (predicted) brightness values ​​L are plotted as a function of the measured brightness values. Ideally, the data points would lie on a line with a slope of 1.

[0103] In Fig.5B Residuals, i.e., absolute deviations between the modeled and measured brightness values, are represented as a bar chart.

[0104] The exemplary values ​​determined for the goodness of fit are R 2<= 0.939, mean absolute error = 1.073, mean squared error = 2.337, standard deviation = 1.529 and mean absolute deviation -0.06.

[0105] In contrast to the one related to Fig.2 and Fig.3 In the described example, where linear regression was used to determine the relationship, significantly improved values ​​are shown when using the tree ensemble, for example a coefficient of determination R 2<= 0.939, which lies between 0.9 and 1 as desired, and a mean absolute error of the brightness (luminosity) of approximately 1, which is below the perception threshold, so that calculated values ​​could no longer be visually distinguished from experimentally determined values.

[0106] Not shown in the figures is an exemplary modeling of a* and b* values ​​using the Tree Ensemble Learner.

[0107] For this purpose, the procedure for computer-aided determination of hair color properties is carried out essentially in the same way as described above for determining the brightness L*, with the difference that in column filter 220, the column with the brightness L* is not determined as the coloring result parameter, but a red / green parameter a* or a blue / yellow parameter b*.

[0108] When determining the goodness of fit using the Numerical Evaluator 450 at node 22, the following values ​​are obtained, which show a similar goodness of fit (accuracy) to the values ​​for the model for L*: The exemplary values ​​determined for a* are R 2<= 0.959, mean absolute error = 1.236, mean squared error = 4.067, standard deviation = 2.017 and mean absolute deviation -0.115.

[0109] The exemplary values ​​determined for b* are R 2<= 0.94, mean absolute error = 1.366, mean squared error = 4.763, standard deviation = 2.183 and mean absolute deviation - 0.063.

[0110] Instead of linear regression or a tree ensemble learner, other methods can also be used in various implementation examples, such as support vector machines or neural networks.

[0111] In an exemplary procedure for the computer-aided determination of hair color properties using a neural network (the Multi Layer Perceptron model with standard parameters was used), the following were observed: Fig.6A und Fig.6B The results shown were achieved.

[0112] Figur 6A und 6B Two diagrams, 600 and 601, show these results of the computer-aided determination of hair color properties in different ways.

[0113] In Fig.6A The modeled (predicted) brightness values ​​L are plotted as a function of the measured brightness values. Ideally, the data points would lie on a line with a slope of 1.

[0114] In Fig.6B Residuals, i.e., absolute deviations between the modeled and measured brightness values, are represented as a bar chart (i.e., a frequency distribution of the residuals).

[0115] The goodness of fit is also better here than in the case of linear regression, but slightly worse than in the case of the tree ensemble learner.

[0116] Figur 7 Diagram 700 shows a result of a method for computer-aided determination of the properties of hair colors according to an exemplary embodiment.

[0117] In the present example, the procedure was essentially carried out as described in connection with Fig.4 , Fig.5A und Fig.5B The procedure described above differs in that, instead of the brightness L* of the hair color, the hair color itself was selected as the coloring result parameter in the form of the three parameters L*, a*, and b* (using the column filter 220). The model was thus determined (using the tree ensemble learner 430) simultaneously for all three coloring result parameters L*, a*, and b*.

[0118] As a result of the exemplary embodiment, in Fig.7 A bar chart for an (unsigned) color difference ΔE, in which the differences in the three individual quantities L*, a*, and b* are used to calculate a Euclidean distance, is shown between the determined color and the measured color. For 75% of the values, ΔE < 2.5.

[0119] Figur 8 Figure 800 shows a diagram which represents a result of a method for computer-aided determination of the properties of hair colors according to an exemplary embodiment.

[0120] In the present example, the procedure was essentially carried out as described in connection with Fig.4 , Fig.5A und Fig.5B The process described above differs in that, instead of the brightness L* of the hair color, a wash fastness after 24 washes (not shown in Table 1 or Table 2) is used as the coloring result parameter. A color difference ΔE (designated as ΔE 00) can be used as a measure of wash fastness. Fig.8 The color difference calculated according to the specified model is plotted against the experimentally measured color difference, with the values ​​essentially scattering around a straight line with a slope of 1, as desired.

[0121] Numerical results for the adaptation quality of a further embodiment in which the color difference after 12 hair washes (see Table 2) was used as a dyeing result parameter (again using a tree ensemble learner, as above in connection with Fig.4 and Fig. 5A und 5B As described, R² ≤ 0.883, mean absolute error = 0.22, mean squared error = 0.124, standard deviation = 0.353, and mean absolute deviation -0.01. Here, the mean absolute error, which corresponds to a mean Δ(ΔE) between measured and calculated wash fastness, is Δ(ΔE) = 0.22, far below any visual distinguishability.

[0122] Figur 9 Diagram 900 shows a result of a method for computer-aided determination of the properties of hair colors according to an exemplary embodiment.

[0123] In the present example, the procedure was essentially carried out as described in connection with Fig.4 , Fig.5A und Fig.5B as described, with the difference that a measured grey coverage is used as the coloring result parameter instead of the brightness L* of the hair color.

[0124] Figuren 10A, 10B and 10CThree diagrams show the relationship between measurement points, true function and model for different models used in a method for computer-aided determination of hair color properties according to different embodiments.

[0125] This demonstrates that while the method for computer-aided determination of hair color properties will typically provide a model, this model does not necessarily represent the hair color data well. For a good model, various implementations can involve selecting a large dataset, allowing for the selection of training and validation datasets from within the dataset.

[0126] The ones related to the Figuren 2 bis 9 The illustrated embodiments are for illustrative purposes only. The method for computer-aided determination of hair color properties can be varied or extended in many ways compared to the examples above, according to various embodiments. For example, the dyeing prerequisite parameters and dyeing result parameters used can be changed, extended, or limited; algorithms for determining the relationship between the dyeing prerequisite parameters and the dyeing result parameter(s) can be varied; parameters for determining the relationship can be defined differently, for example, from the standard parameters specified by KNIME; for example, different dyeing prerequisite parameters or dyeing result parameters can be assigned different weightings; a different computer program can be used to execute the method, or similar modifications.

[0127] Additionally, it is possible to divide the data into test, training, and validation sets.

[0128] The Figuren 10A bis 10C illustrate that, for example, the degree of an adaptable function can be limited, because although the method for computer-aided determination of hair color properties can also be used for first-order functions according to various embodiments ( Fig.10A ) and 5 . Grades ( Fig.10C ) would deliver a result, a comparison of the determined goodness of fit of the three models (relationships) would show that a model using a fourth-degree function delivers the best result of the three models.

[0129] Figur 11 shows a flowchart which represents a method for computer-aided determination of hair color properties according to various embodiments.

[0130] In various embodiments, the method for computer-aided determination of hair color properties can include providing hair color data, wherein the hair color data for a plurality of dyeing processes each have values ​​for a plurality of dyeing prerequisite parameters and for at least one dyeing result parameter, wherein for each dyeing process of the plurality of dyeing processes the plurality of dyeing prerequisite parameters has a first concentration of a first dye precursor and a second concentration of a second dye precursor; wherein for each dyeing process of the plurality of dyeing processes the at least one dyeing result parameter has a measured value about a property of hair colors (at 1110), and determining a relationship between the plurality of dyeing prerequisite parameters and the at least one dyeing result parameter by means of predictive analysis based on the hair color data (at 1120).

[0131] Figur 12 is a graphic representation 1200 of a data processing device 1210 for computer-aided determination of properties of hair colors according to various embodiments.

[0132] The data processing device 1200 can, for example, be or comprise a PC, a laptop or any other data processing device suitable for carrying out the method for computer-aided determination of hair color properties, i.e., having, for example, a sufficiently large memory and a sufficiently powerful processor.

[0133] The data processing device 1200 includes a processor 1220. The processor 1220 can, for example, be a microprocessor of the data processing device 1200 or include such a microprocessor.

[0134] In various embodiments, the data processing device 1200 can include a data storage device 1230. The data storage device can be an internal or external data storage device 1230 of one of the aforementioned data processing devices 1200, or it can include such a data storage device 1230. The data storage device 1230 can be configured to store data that is stored and / or retrieved during the execution of the method for the computer-aided determination of hair color properties, for example, the hair color data.

[0135] In various embodiments, the data processing device 1200 can include a display device 1240. The display device 1240 can, for example, be a screen of a PC, a laptop, or any other data processing device 1200. The display device can be used, for example, to display results of the method for computer-aided determination of hair color properties, to request input parameters for executing the method, or similar purposes. The display device can be used, in particular, at a point of sale for hair treatment products.

[0136] In various embodiments, the data processing device 1200 can have an input device 1250 for providing information to the data processing device 1200, for example a keyboard, a mouse, a touch-sensitive surface of the display device 1240, or the like.

[0137] Further advantageous embodiments of the method result from the description of the device and vice versa.

Claims

1. A method for computer-assisted determination of hair colour characteristics, comprising: • providing hair colour data suitable for application of predictive analytics, o wherein the hair colour data for a plurality of colouring processes comprise values for a plurality of colouring prerequisite parameters and for at least one colouring result parameter , o wherein, for each dyeing process of the plurality of dyeing processes , the plurality of dyeing prerequisite parameters comprises a first concentration of a first dye precursor and a second concentration of a second dye precursor , wherein the plurality of dyeing prerequisite parameters further comprises a base hair colour, wherein the base hair colour is parameterised in a colour space , wherein the plurality of dyeing prerequisite parameters further comprises prior damage to the hair; o wherein, for each dyeing process of the plurality of dyeing processes, the at least one dyeing result parameter comprises a measured indication of a property of hair colours produced by dyeing hair, wherein the properties of hair colours produced by dyeing hair comprise: one or more parameters of a hair colour parameterised in a colour space; • determining a relationship between the plurality of dyeing prerequisite parameters and the at least one dyeing result parameter by means of predictive analytics based on the hair colour data , wherein the relationship comprises a model that, for each value of a combination of values for a plurality of dyeing prerequisite parameters , including values that do not correspond to experimentally determined values, a value for at least one dyeing result parameter is determined .

2. Method according to claim 1, further comprising: determining, by means of the determined relationship, a value for a dyeing result parameter for a combination of values, each comprising a value for a selected plurality of dyeing prerequisite parameters, wherein the combination of values is not identical for any of the dyeing processes with its combination of values for the selected plurality of dyeing prerequisite parameters.

3. Method according to claim 1 or 2, wherein the predictive analytics uses at least one method from a group of methods, wherein the group of methods comprises: linear or multi-linear regression, multiple polynomial regression, neural network methods, support vector machine methods; and decision tree methods.

4. Method according to claim 3, wherein the decision tree method uses decision tree ensembles ("tree ensembles").

5. Method according to one of claims 1 to 4, wherein the plurality of colouring prerequisite parameters further comprises a degree of greying.

6. Method according to one of claims 1 to 5, wherein the at least one dyeing result parameter comprises at least one further property of hair colours produced by dyeing hair, wherein the further properties of hair colours produced by dyeing hair comprise : a wash fastness; light fastness; and an ability to cover grey hair.

7. Method according to any one of claims 1 to 6, wherein the at least one dyeing result parameter comprises a plurality of dyeing result parameters.

8. Data processing apparatus for performing computer-aided determination of characteristics of hair colours, wherein the data processing apparatus is configured to perform the method according to claim 1.