MXPA97005596A - Prediction system of proportions and method to obtain a mez - Google Patents

Abstract

The present invention relates to a system for predicting proportions, to realize a predetermined objective, by mixing a predetermined number of elements in predetermined proportions. The system includes an extractor (1) of characteristics of the objective, to extract a characteristic from the predetermined objective: an evaluator (2), operable by receiving proportion vectors, which are represented by the characteristics of the predetermined objective and the quantity of each one. of the elements, respectively, to determine the ideal form of the proportion vectors, based on the characteristic extracted from the predetermined objective, and a processor (3) of the genetic algorithm, to predict the proportion vectors based on the aptitude of agreement with a genetic algorithm, in which the quantity of each of the elements and each of the proportion vectors are represented by a gene and a chromosome, respectively. The proportion vectors, which have been predicted by the processor of the genetic algorithm, are entered into the evaluator, to cause this evaluator to repeat the evaluation of the proportion vectors, to determine the optimal proportion vectors.

Description

PREDICTION SYSTEM OF PROPORTIONS AND METHOD TO OBTAIN A MIXTURE FIELD OF TECHNOLOGY The present invention relates to a system for predicting proportions, for forecasting a ratio or a mixing ratio, in which a plurality of materials, colors, lights, acoustic signals, electrical signals, electromagnetic waves or the like, is combine to produce a desired objective, and also refers to a method for obtaining a mixture of the desired objective, using the ratio predicted by the proportions prediction system. PREVIOUS TECHNIQUE In discussing the prior art, reference is made, for example, to the field of prediction of color formulations to prepare a color of a particular specification, by mixing a plurality of pigments. In the description that follows, the original pigments that are to be mixed to produce the color of the particular specification, are referred to as the elements. The simplest method of the prior art for this purpose is that where an expert empirically determines the mixing ratio or ratio of the elements to be mixed, observing the target color. Another method of the prior art comprises the steps of analyzing a spectrum of the target color, making a search in the database to find a color, which may correspond to the color of a spectrum closer to the spectrum previously analyzed, and adjust, in fine form, the proportion of the elements making reference to the proportion of the pigments of elements that were used to produce the previous color. A further method of the prior art is disclosed, for example, by G. yszecki and WS Stiles in Color Science: Concepts and Methods, Quantitative and Tactical Dates and Formula ("Color Science: Concepts and Methods, Quantitative Data and Formulas"). ) (New York, NY: Wiley, 2nd Ed. (1982)), where a model, which mathematically simulates the relationship between the color spectrum and the projected image of the proportion, as represented by the Kubelka- Munk, is used to determine the proportion of the color spectrum. A more recent prior art method is disclosed, for example, by JM Bishop, MJ-Bushnell and S. Estonia in Applying Neural Networks to Computer Recipe Prediction ("Applying Neural Networks to the Prediction of Computer Formulations ") (Color Research and Application, Vol. 16, No. 1, pages 3-9 (1991), where the neural network is taught to produce the ratio in response to the input of the color spectrum.

In any case, according to the prior art, the elementary pigments are mixed together according to the proportion, thus obtained, to produce the objective color. While the above discussion is directed to the color recipe, a similar discussion applies equally to the field of perfumery, preparation of food products, sound effects designs, development of materials or the like. All the methods of the prior art have a common characteristic in that a color, light or material, which satisfies a given specification, is produced on a trial and error basis, with the use of human sensory perception, and experiences and bases of Experimental data. It will be easily understood that it is not easy to obtain high precision. The model, simulated mathematically, represented by the equations indicated in the theory of Kubelka-Munk, which are generally used in the field of color formulations, is limited in its application, since it is difficult to prepare the model that meets all conditions available. Although the conventional theory of Kubelka-Munk is widely used in forecasting color correspondences, some assumptions are made that limit the situations where the theory can be applied. It is really very difficult to configure a model that replaces the model discussed above.

The use of the neural network seems to be an effective technique to remove the difficulty in configuring the substitute model. Nevertheless, human vision is too sensitive to allow the neural network to display an accuracy required to predict small proportions, as low as 0.01%. EXPOSITION OF THE INVENTION In view of the foregoing, the present invention seeks to provide an improved system for forecasting proportions, to systematically predict, with great accuracy, the proportion or mixing ratio in which a plurality of elementary colors, elemental light, Elemental sound signals, elementary electrical signals or elementary electromagnetic waves are mixed to produce a target material, a target color, an objective light, a target sound signal, a target electrical signal or an objective electromagnetic wave. The present invention also seeks to provide an improved method for obtaining a mixture of the desired objective, such as a material, color, light, sound signal, electrical signal or electromagnetic wave, having a specification similar to that of the target material, objective color, light target, target sound signal, objective electric signal or objective electromagnetic wave.

It will be noted that the term "element", used herein, will be understood as meaning a material, color, light, an acoustic signal, an electrical signal or an electromagnetic wave, having a variety of physical and / or chemical characteristics, having a capacity to be mixed and the term "objective", also used here, will be understood as meaning a material, color, light, acoustic signal, an electrical signal or an electromagnetic wave, produced by mixing these elements. Likewise, the term "mixture", mentioned here, will be understood as meaning not only a direct mixture to a physical matter, chemical matter or electrical matter, but also a mixture of matter which the human sensory organs perceive, although no direct mixture of matter take place. By way of example, the mixture of matter which the human sensory organs perceive, includes the preparation of a mixed color, spatially arranging particles of a color tone agent in a given area, in varying proportions, by means of, for example , a color printer, or by the display by means of, for example, a color display, colors of varying intensities of light, spatially distributed in the color display. Therefore, the term "mixture", used herein, shall be understood as meaning a material, color, light, acoustic signal, electrical signal or electromagnetic wave, thus prepared, by the specific mixture. According to one aspect of the present invention, a proportions prediction apparatus is provided, which includes a proportional characteristic extractor to determine a characteristic of a target, a genetic algorithm (GA) calculator, to predict the proportion or mixing ratio of two or more elements, and an evaluation element to calculate a similarity between two characteristics. According to another aspect of the present invention, a method is provided to obtain a mixture, which includes a ratio characteristic extractor to determine a characteristic of a target, a genetic algorithm (GA) calculator, to predict the mixing ratio or ratio of two or more elements, an evaluation element for calculating a similarity between the two characteristics and a mixer for mixing the elements together in respective proportions finally calculated by the AG processor. The evaluation element employed in the present invention includes at least one of the evaluation section of the mixing element, adapted to receive the proportions predicted by the AG processor to evaluate the type of elements of which the proportion is not zero, and / or a mixing characteristic evaluation section, to forecast a characteristic of the mixture formed by mixing the elements in the respective proportions and to compare it with the characteristic of the objective. The evaluation section of the mixing element of the evaluation element used in the present invention includes at least one of an evaluation section of the element number and an evaluation section of the unnecessary element number. This evaluation section of the element number includes a selector of the number of mixing element, adapted to receive the proportions predicted by the AG processor, to digitize the predicted proportions, a forecaster of the mixing element, adapted to receive the characteristic extracted by the extractor characteristic of the mixture, to predict the type of elements that form the objective, and a calculator of the distance of the element number to compare the number of the element, obtained from the selector of the number of mixing element for use in the mixture, with the element number obtained from the forecaster of the mixing element and to produce a similarity. On the other hand, the evaluation section of the unnecessary element number includes a selector of the number of the mixing element, a knowledge base in which unnecessary combinations for the mixture are described in the form of knowledge, and a penalty section, adapted to receive the element number obtained from the selector of the number of the mixing element for use in mixing and for obtain reference to the knowledge base to decrease an aptitude in the case of the presence of the unnecessary element number. The system for evaluating the mix characteristic includes a forecaster of the mixing characteristic, adapted to receive the predicted proportions, calculated by the AG processor and to predict the characteristics of the mixture obtained by mixing the elements in the predicted proportions, and a calculator of the distance of the characteristic of the mixture to compare the characteristic extracted by the extractor of the mixing characteristic, with the characteristic of the mixture predicted by the forecaster of the characteristic of the mixture and to produce a similarity. The AG processor, used in the present invention, includes a section for determining the initial value of the AG, for determining the initial values for predicting the proportions according to the genetic algorithm, and a dynamic AG processor for determining in sequence the proportions. The section for determining the initial value of AG includes at least one section for determining the initial proportion value, adapted to receive the characteristic of the target and forecast the proportions that are then made to be the initial values. The initial value determination section of AG, mentioned above, may include at least one element to determine the value of the initial proportion to receive the characteristic of the objective to forecast the proportions that are then made to be the respective initial values, related combinations with a base of knowledge of the elements and a generator of multiple optimal selections to compare an output of the element of determination of the value of initial proportion with the knowledge to formulate new initial values of the AG processor, replacing the proportions of the candidates of unnecessary elements with zero. Alternatively, the initial value determination section of the AG may include an extractor of the color space characteristic, to produce coordinates in a color space, such as L * -a * -b * of that target color, a classifier of the color space, adapted to receive the coordinates in the color space, obtained by the color space characteristic extractor, to determine in which representative color area the objective color is included, a knowledge base, in which combinations related to the knowledge of colors of the elements are described, a generator of a random initial value, to produce random proportions, and a random initial value corrector, adapted to receive the ratio generated by the random initial value generator and the information in the representative color area determined by the color space classifier and operable to make a search in the cone base The basis for determining whether an inappropriate color combination in the color area is included in a random value of the proportion obtained from the initial random value and for correcting the inappropriate proportion of random. In accordance with the present invention, in order to prepare a mixture having a substantially identical characteristic to that of the given objective, the characteristic of the given objective is first determined by the characteristic extractor of the mixture. Such a feature may be, in the case of sound or color, a physical characteristic, such as a spectrum. The GA processor then predicts, based on the genetic algorithm, the mixing ratio of the mixture of a formulated formulation by mixing two or more elements. The evaluation element compares the characteristic determined by the extractor of the characteristic of the mixture with this characteristic of the mixture which is finally prepared, by mixing the elements in the respective proportions predicted by the GA calculator and also evaluating the degree of accuracy of the proportions predicted by the AG processor. The ratio prognostic is comprised of the extractor of the characteristic of the mixture, described above, the AG processor and the evaluator element. The mixer serves to mix the elements in the respective predicted proportions to supply the mixture. The AG processor uses, as the respective initial values, the proportions determined by the section that determines the initial value, so that the accuracy of the prediction of the mixing ratio can be progressively increased based on the genetic algorithm. Likewise, in order to evaluate the validity of the proportions predicted by the AG processor, the evaluation element produces a result of the evaluation obtained by the evaluation section of the training element number, the evaluation section of the element number unnecessary or the evaluation section of the mix characteristic. The evaluation section of the element number is comprised of the selector of the number of the mixing element, the forecaster of the mixing element and the calculator of the distance of the element number. The mix element number selector is operable, in response to the proportions predicted by the AG processor, to digitize the forecasted mixing ratio, that is, to cause the predicted proportions to be shifted ("OFF"), if they they are less than a predetermined threshold value, but are active ("ON"), if they are greater than the predetermined threshold. The forecaster of the mixing element receives the characteristic extracted by the extractor of the mixing characteristic, so that the type of elements that make up the objective can be predicted. The item number distance calculator compares the item number, which is obtained from the selector of the mix item number and which is used in the mix, with the item number obtained by the forecaster of the mix item and they produce a similarity between them. The evaluation section of the unnecessary item number is comprised of the mix element number selector, the knowledge base and the penalty or penalty section. The knowledge base has the necessary knowledge to eliminate redundant elements, while the penalty section receives the element number obtained from the selector of the mix element number for use in the mixture and refers to the knowledge base so that the suitable form can be reduced in the case of the presence of combinations of unnecessary elements. The section for evaluating the characteristic of the mixture is comprised of the predictor of the characteristic of the mixture and the calculator of the characteristic distance of the mixture. The forecaster of the characteristic of the mixture receives the proportions predicted by the AG processor and forecasts the characteristic of the mixture, which is obtained by mixing the elements in the respective predicted proportions. The distance calculator of the characteristic of the mixture compares the characteristic obtained by the extractor of the characteristic of the mixture with the characteristic of the mixture obtained by the forecaster of the characteristic of the mixture and then supplies an output indicative of the similarity between they. BRIEF DESCRIPTION OF THE DRAWINGS This and other objects and features of the present invention will become clear from the following description, taken in conjunction with its preferred embodiments, with reference to the accompanying drawings, in which similar parts are designated by numbers similar references, and in which: Figure 1 is a block diagram, showing a proportions prediction apparatus, according to the present invention; Figure 2 is a block diagram, showing an output of an AG processor used in the apparatus of the present invention; Figure 3 is a flow diagram showing a genetic algorithm employed in the practice of the present invention; Figure 4 is a block diagram showing a method of obtaining a mixture, according to the present invention; Figure 5 is a block diagram showing the details of an evaluation element, used in the apparatus of the present invention; Figure 6 is a block diagram showing the details of a forecaster of the mixing element, used in the apparatus of the present invention; Figure 7 is a block diagram, showing the details of a characteristic forecaster of the mixture, used in the apparatus of the present invention; Figure 8 is a block diagram, showing the details of the AG calculator, used in the apparatus of the present invention; and Figure 9 is an explanatory diagram, showing a color space a * -b * and a confusing classification of the color space a * -b *. BEST MODE FOR CARRYING OUT THE INVENTION In describing some of the preferred embodiments of the present invention, it will be described as being applied to the prediction of dye proportions, that is, the prediction of the mixing ratio of the dyes that are going away. to mix to produce a desired color, as is the case with the prior art, discussed above. Referring first to Figure 1, a schematic representation of blocks of a proportions prediction system is shown, according to a preferred embodiment of the present invention. The proportions prediction system, shown there, comprises an extractor 1 of the mixture characteristic, an evaluation unit 2 and a genetic algorithm processor 3 (AG). Assuming that the color of an objective paint is determined in accordance with a specification, the feature extractor 1 of the mixture analyzes and produces a color spectrum of the objective color. Although the spectrum represents the physical properties that appear continuously on the frequency axis, in most cases the spectrum is expressed in terms of n values, using n filter banks and n discrete Fourier transformations. The AG processor 3 makes use of a genetic algorithm to predict how to mix elementary pigments (m is greater than 1), in a particular mixing ratio to produce the target color. Figure 2 illustrates examples of mixing ratios or proportions of the elementary pigments, which have been predicted by the AG processor 3, and those mixing ratios are produced from the AG processor 3. The aforementioned genetic algorithm is well known to those skilled in the art and is discussed, for example, by DE Goldberg 'in Genetic Algorithm in Search, Optimization and Machine Learning ("Genetic Algorithms in Search, Optimation and Learning of Machines ") (Addison-Wesley, 1989) and in Handbook of Genetic Algorithms (" Manual of Genetic Algorithms "), edited by L. Davis (Van Nostrand Reinhold, 1990). The evaluation unit 2 compares the information in the target color fed from the mixture characteristic extractor 1 with the predicted proportion fed from the processor 3 d AG to determine, and then produce a suitable state (or evaluated value) descriptive of the degree of accuracy of the predicted proportion, ie the degree of color correspondence between the target color and the color represented by the predicted proportion. The AG processor 3 makes use of the output value from the evaluation unit 2 to progressively increase the degree of accuracy of the predicted ratio, according to the genetic algorithm. Referring to Figures 2 and 3, the AG processor 3 operates in accordance with the genetic algorithm and the proportion vectors to be predicted by this AG processor 3 are treated as chromosomes of the genetic algorithm, as shown in FIG. Figure 2. Assuming that the number of types of elementary pigments is m, the proportion vectors to be predicted represent a vector of order m. In the case of the proportion vector 1, shown in Figure 2, it means that elementary pigments, ie white, green-1, green-2, yellow and red pigments, are mixed in proportions of 0.23, 0.04, 0.31, 0 and 0.11, respectively. Since the genetic algorithm is thus designed to make a search with the use of plural chromosomes, as shown in Figure 2, the AG processor 3 shares with the corresponding plural proportion vectors and successively increases its prediction accuracy. Figure 3 illustrates a flow chart showing how the prediction accuracy of predicted proportions increases by the AG processor 3. The flow chart, shown in Figure 3, is based on the genetic algorithm. As shown there, in the AG processor 3 the n chromosomes, ie the proportions, are first started. The simplest conventional initialization method is to start them with random values. However, according to a different embodiment of the present invention, the initialization method is designated only as will be described later. The proportions, thus initialized, are supplied to evaluate unit 2. As a matter of course, the prediction accuracy of the proportions made in the first cycle will not be so high and, therefore, evaluation unit 2 may not give high fitness to the n proportion vectors. The m fitness values (evaluated values) are again fed back to the AG processor 3, which then selects a number of predicted proportion vectors, which give a high aptitude according to the m fitness values, to feed the processor 3 of AG. The proportion vectors selected by the AG processor 3, represent the parents that produce offspring proportion vectors of the following generation. According to the genetic algorithm, this process is referred to as a selection. The next step is to cross over two predicted proportion vectors, arbitrarily selected from the selected parent ratio vectors, in order to provide predicted proportion vectors of the following next generation. This process is carried out by the cross operation to combine the two order-of-order vectors m and produce two different order-of-order vectors m. The crossing of a simpler point will now be discussed as an example. Suppose that the two order-of-order vectors m are represented by chromosomes a and b. The crossing at a point is a calculation to produce vectors of proportion of the first offspring from an element of order r of the first half of chromosome a and an element of order (mr) of the last half of chromosome b and produce second vectors of proportion of offspring from the order element r of the first half of chromosome b and an element of order (mr) of the last half or of chromosome a, where 0 <; r < m. This crossing refers to the crossing of a point, since the descendants are produced by changing the elements in a limit between the order r and the next order (r + 1) from the beginning. By repeating this crossing, the n-proportion vectors of the second generation are produced. The calculation performed at the end of the genetic algorithm is referred to as a mutation in which random values are added to the descriptive values of the proportions of the arbitrarily chosen element of the chromosomes chosen arbitrarily. Considering the search, this mutation corresponds to a general search to prevent the proportions from falling into the local minimum. The n chromosomes obtained by the previous process of selection, crossing and mutation represent the proportion vectors of the second generation. Those proportionate vectors of the second generation are again fed back to the evaluation unit 2, thus repeating the process shown in Figure 3. The process flow, shown in Figure 3, ends when the prediction accuracy of the predicted proportions increase to a required accuracy. The multi-part functions of the AG processor 3 components will be described later.

Referring now to Figure 4, there is shown a method for preparing a mixture, according to a preferred embodiment of the present invention. In Figure 4, the reference number 4 represents a ratio predictor and the reference number 5 represents a mixer. The forecaster 4 is comprised of the extractor 1 of the mixing characteristic, the evaluation unit 2 and the AG processor 3, all shown in and described with reference to Figure 1, and has an identical function with that of the prediction system of proportions shown in Figure 1. Mixer 5 operates to mix elementary pigments m in the proportion determined by forecaster 4 of proportions. The functions of the various components will be described in detail later. Figure 5 illustrates an embodiment of the evaluation unit 2 employed in the practice of the present invention. The evaluation unit 2 comprises an evaluation unit 6 for mixing elements, an evaluation unit 7 for the characteristics of the mixture, and a fitness integrator 8. The evaluation unit 7 of the mixing elements, shown there, performs the evaluation from two points of view. This evaluation unit 6 of the mixing element includes a section 61 for evaluating the element number, and an unnecessary evaluation section 63 for the element number. The evaluation unit 6 of the mixing element evaluates which elementary pigments are to be mixed together. In other words, this evaluation unit 6 of the mixing elements serves to convert the predicted ratio by the AG processor 3 into a binary information (ACTIVE ("ON") or NOT ACTIVE ("OFF")), indicative of whether the pigments of the element are going to mix or if they are not going to mix and compare these with the physical characteristics of the target color. In a preferred embodiment of the present invention, a circuit structure, which includes the evaluation section 61 of the element number and the non-necessary evaluation section 63 of the element number, performs a specific evaluation. On the other hand, the evaluation unit 7 of the characteristics of the mixture evaluates the value of the proportion predicted by the AG processor 3. The fitness integrator 8 serves to apply a weight in each fitness plurality and then adds them together, so a fitness can be produced from it as an output from the evaluation unit 2. The simplest weighing method is to apply a weight same. The evaluation section 61 of the element number, shown in Figure 5, includes a selector 620 of the number of the mixing element, a forecaster 611 of the mixing element and a calculator 612 of the distance of the element number. The evaluation section 61 of the element number of the construction described above works in the following manner. The selector 620 of the mixing element receives the proportions, as shown in Figure 2, which have been predicted by the AG processor 3 and converted into one of the binary digits consisting of the information "OFF". , which represents that the elementary pigments in a smaller proportion than the threshold value do not mix, and the information ACTIVE ("ON"), which represents that the elementary pigments in any other proportion are mixed together. The aforementioned threshold value is determined in consideration of the visual perceptibility of whether one can perceive a difference when the elementary pigments are mixed, and the prediction accuracy of the AG processor 3. On the other hand, the predictor 611 of the mixing elements makes use of a color spectrum obtained from the extractor 1 of the characteristic of the mixture, in order to directly predict which elementary pigments are to be mixed together. Figure 6 illustrates an example in which the forecaster 611 of the mixing element is performed by a neural network. In Figure 6, the reference number 6111 represents a converter of the characteristic element and the reference number 6112 represents a threshold processor. The converter 6111 of the characteristic element number is in the form of a neural network of the front-feed type, of three layers, and is adapted to receive the spectrum of color of order n, obtained by the extractor 1 of the characteristic of the mixture, so that any of an ACTIVATION ("ON") produced for the mixture of the elementary pigments of order or a DISPLACEMENT ("OFF") produced for the non-mixing of the elementary pigments of order m, can be produced from them. The neural network for the 6111 converter of the characteristic element number acquires knowledge of a set of available drag data. The elementary pigments are previously mixed together to produce a representative color paint, which is subsequently subjected to colorimetric measurement to determine the color spectrum of the resulting paint. In such a case, since the elementary pigments that have been mixed together are known, an exact relation between the color spectrum and the information of order m of ACTIVE or INACTIVE, can be obtained. This information is used as a signal from the teacher to allow the neural network to acquire knowledge. Several learning algorithms for the neural network have been suggested so far and, if, for example, the widely used subsequent propagation learning rule is used, the 6111 converter of the characteristic element number can be easily carried out. However, in practice, even if the neural network is taught it produces one of the binary digits, 0 or 1, it will not produce a full 0 or 1 and rather a 0.001 or 0.998, since an output layer of the neural network makes use of a continuous function, such as a sigmoid function. The threshold processor 6112 serves to digitize those values. The threshold processor 6112 compares the characteristic element number converter 6111 with the predetermined threshold set in e ', to forcefully convert the output of the neural network into one of the ACTIVE and NON-ACTIVE signals. As an example, the signals of ACTIVE and NOT ACTIVE are produced from the threshold processor 6112 if the value is greater than 0.5 and less than 0.5, respectively. The calculator 612 of the item number operates to make a comparison between the information from which the elementary pigments are to be mixed according to the color spectrum of the objective paint obtained from the predictor 611 of the mixing element and the information of which the elementary pigments are to be mixed according to the predicted proportion obtained from the selector 620 of the number of the mixing element and then to supply an output indicative of the degree of difference between these pieces of information. The calculator 612 of distance of the element number determines two binary vector distances of order m. The simplest method to achieve this is to calculate the number of bits that are different. This distance is then fed to the fitness integrator 8.

Another specific example of the evaluation unit 6 of the mixing element is the evaluation section 63 of the unnecessary item number. This evaluation section 63 of the unnecessary element number, shown in Figure 5, includes the selector 620 of the mix element number, a knowledge base 631 and a penalty calculator 632. The evaluation section 63 of the unnecessary element number of the construction, described above, operates in the following manner. The penalty calculator 632 is adapted to receive from the selector 620 of the mixing element number the information in which the elementary pigments are to be mixed with each other and has access to the knowledge base 631 to determine whether a combination of elementary pigments input from selector 620 of the mixing element number is unacceptable or unnatural. Knowledge base 631 stores knowledge associated with the color combination therein, such as, for example: Rule 1: Avoid the use of complementary color pigments, for example red and green pigments; Rule 2: Avoid the use of pigments of the same color, for example pigments green-1 and green-2; and Rule 3: Maintains the proportion in approximately 100% The penalty calculator 632 generates evaluated values that depend on the extent that the predicted proportion obeys these rules A portion of the evaluation unit 2 in which a third important evaluation is carried out is the evaluation unit 7 of the mixing characteristic. As shown therein, this mixing feature evaluation unit 7 includes a forecaster 72 of the distance of the mixing characteristic operating to interpret, based on the predicted ratio by the AG processor 3, the perceived attributes of the predicted color. , for example the coordinates of the values L * -a * -b * in the chroma-ticity diagram (x, y) CIÉ of 1976. It will be noted that L *, a * and b * represent luminosity, hue and intensity of color . This calculator 71 of the characteristic distance of the mixture receives the target coordinates of L * -a * -b * from the mixing characteristic extractor 1 and calculates the color distance, Euclidean distance, between the two vectors of the mixing ratio. order m in the space L * -a * -b *. The values L *, a * and b * are calculated from the surface spectral reflectance. It is difficult to obtain high accuracy in mapping the proportions to L * -a * -b *. Figure 7 shows a method of acquiring the mapping by means of a learning function of the neural network. The training data for this purpose can be obtained in a manner similar to the training data for the neural network shown in Figure 6. The elementary pigments are mixed in advance together to supply a representative color paint which is submitted subsequently to the colorimetric measurement to determine the color spectrum of the resulting paint. Once the color spectrum has been analyzed, the conversion of the color space coordinates is easy to achieve. Therefore, the ratio between the proportion of the elementary pigments and the coordinates of the color space is available in advance and this is used as a training data for the neural network of the forecaster 72 characteristic of the mixture. The color of the paint obtained when the elemental pigments have been mixed, according to the ratio predicted by the AG processor 3 and the color of the objective paint are converted into the distance in the color space that is subsequently fed to the integrator of aptitude 8. In this way, the evaluation unit 2 evaluates the accuracy of the predicted ratio by the AG processor 3. The evaluation is fed back to the AG processor 3. In the above description, the evaluation unit 2 has three components, namely the evaluation section 61 of the element number, the evaluation section 63 of the unnecessary element number and the evaluation unit. 7 evaluation of the mix characteristic, to carry out the evaluation, the use of only one of them is sufficient for evaluation unit 2 to carry out this evaluation. However, the use of the three components for the evaluation unit 2 is advantageous in that greater accuracy can be expected, but the use of these components in combination is not always essential. (See the following Table.) When preparing a paint by mixing the elemental pigments, one can not see what color will be presented, unless they really blend together. However, the occasional preparation of a mixture of the elemental pigments so that the resulting paint can represent a color corresponding to that of the objective paint requires a substantial amount of time and also involves a problem associated with economy. Three evaluation methods carried out by the evaluation unit 2 provide a solution to this problem and, specifically, predicting the color that would be given by mixing the elementary pigments, the color correspondence simulated by the computer can be achieved stably with the objective painting. Figure 8 illustrates the structure of the AG processor 3 according to a preferred embodiment of the present invention. Referring now to Figure 8, the AG processor 3 includes a unit 31 that determines the initial value and a unit 32 that determines the dynamic AG.

The AG processor 3 operates in the following manner, according to the genetic algorithm. Since the genetic algorithm is widely used in art, the terms "population size", "chromosome", "gene", etc., which are used in association with the genetic algorithm, are used here without giving a definition of they, as it is believed will be understood by experts in the field. The chromosomes produced from the AG processor 3 are shown in the diagram. Proportions, that is to say the mixing ratios, of the elementary m pigments, combined in the chromosomes, correspond respectively to m genes. According to the genetic algorithm, n such chromosomes are prepared, in which n represents the size of the population. At first, the genes of each of the n chromosomes are initialized by the unit 31 which determines the initial value, which causes the AG processor 3 to produce an initial output. The unit 31 that determines the initial value operates only once during the start. The values of the chromosomes evaluated by the evaluation unit 2 are returned to the AG processor 3. The unit 32 that determines the dynamic AG performs, based on aptitude, genetic operations (selection, crossing and mutation), according to the genetic algorithm, to produce n chromosomes of the next generation. The chromosomes (ie, the predicted proportions) thus produced come from the AG processor 3. The unit 32 that determines the dynamic AG repeats this procedure. Thus, this unit 32 which determines the dynamic AG plays a very important role of the AG processor 3. However, although the AG 3 processor operates in accordance with the same genetic algorithm, the performance of predicting the ratio made by the AG processor 3 depends on the initial value determined by the unit 31 that determines the initial value. Next, the structure and operation of the unit 31 that determines the initial value will be discussed. Referring to Figure 8, the unit 31 that determines the initial value includes a counter 311, a control unit 312, a determiner 313 of the initial ratio, a generator 314 of multiple optimal selections, a feature extractor 316 of the space of color, a classifier 317 of the color space and a corrector 318 of the random initial value. The unit 31 that determines the initial value of the construction, described above, performs an efficient initialization using three methods, such as those carried out hitherto in the genetic algorithm. The first initialization is carried out in the initial proportion determiner 313, which is adapted to receive from the extractor 1 of the blend characteristic information about the color spectrum of the objective paint and produce the ratio, ie the ratio of mixing, of the elemental pigments. The determiner 313 of initial proportion by itself can be said to be an apparatus that predicts the proportions and can be realized by a neural network. Such a neural network is reported in a document of Bishop et al. , discussed herein in relation to the prior art and which can be carried out in a manner similar to the neural network shown in any of Figures 6 and 7. As shown by the experimental data, which will be discussed below, the determiner 313 of the initial proportion alone does not provide a sufficient forecast. However, the output of the unit 313 for determining the value of the initial ratio can be expected to be close to the proportion that will be finally predicted by the proportions prediction apparatus, as compared to the random value used for the initialization during execution of the standard genetic algorithm. Therefore, if the predicted proportion, obtained from the initial proportion determiner 313, is used in the proportion prediction system of the present invention as the initial value, the performance of the prediction can be increased. The second initialization is performed on the determiner 313 of the initial ratio, base 631 of knowledge and generator 314 of multiple optimal selections. With the determiner 313 of the initial ratio, a simple initial value of a good "birth" can be obtained. However, the determiner 313 of the initial ratio fails to take into account a relationship between the complementary color and the same color and the predicted predicted proportion of how and what elementary pigments are to be mixed and often contains unnecessary and / or useless combinations of the elementary pigments. The generator 314 of multiple optimal selections is adapted to receive the predicted ratio from the determiner 313 of the initial ratio and produces a predicted proportion of a good birth, different from the output from the determiner 313 of the initial ratio making access to knowledge in the color combination described in base 631 of knowledge, so that the proportion of the elementary pigments can be made to be zero (ie, these elementary pigments should not be mixed together) in the event that the predicted proportion contains an unnecessary combination and / or useless of elementary pigments. The generator 314 of multiple optimal selections serves to provide better starting points for the genetic search. The third initialization is carried out in the generator 315 of the random initial value, the extractor 316 of the color space characteristic, the classifier 317 of the color space, the corrector 318 of the random initial value and the knowledge base 631. The generator 315 of the random initial value randomly produces proportions in a manner similar to the random initial value generator used for the initialization of the standard genetic algorithm. The extractor 316 of the color space characteristic performs the calculation to determine the coordinates of the color space, such as the space L * -a * -b * of the target color spectrum. The color space classifier 317 divides or classifies the color space into representative color areas and provides a descriptive output of the extent to which the coordinates of the color space, obtained from the color space feature extractor 316, belong to which representative color area divided by such a classifier 317. Figure 9 illustrates an example in which a two-dimensional space a * -b * is projected onto the arc-tangent hue angle (b * / a *) and the angle of Tint is classified confusing in five representative color areas, namely, red (R), yellow (Y), green (G), blue (B) and violet (V). The triangles in Figure 9 define the respective degrees of membership in the correspondence with the hue angles from red (R) to violet (V). The color varies continuously and the degree of ownership moves progressively from a certain color area representative to the next, with a change in the hue angle. Figure 9 also illustrates a generated output of the color space classifier 317, when a coordinate (a * j_, b * ±) of the color space, from the extractor 316 of the color space feature. The color space classifier 317 provides an output that indicates that, in the case of the arc-tangent hue angle (b * -_ / a * i) for these coordinates, 0.75 and 0.25, which belong to the color areas red and yellow, respectively, but not to any other color area. The corrector 318 of the random initial value receives this degree of membership and also the random initial value for the proportion of the elementary pigments from the generator 315 of the random initial value. Then, the corrector 318 of the random initial value refers to the degree of belonging to the representative color obtained from the classifier 317 of the color space and the knowledge of the color combination, obtained from the base 631 of the knowledge, in order to modify the proportions generated randomly. As an example, when the color predominantly represents red, but moves to a yellow area to some extent, the corrector 318 of the random initial value forces the proportions of green to be zero, due to the red-green complementary color relationship, and make any of the red-1 or red-2 ratio zero. The above three initializations are controlled by the counter 311 and the control unit 312. The counter 311 is initialized to zero. The control unit 312 operates only when the counter count 311 is zero. In other words, the population is initialized by the determiner 313 of the initial ratio, the generator 314 of multiple optimal selections and the corrector 318 of the random initial value. Since the generator 315 of the random initial value can initialize any number, some are initialized at the beginning by the initial ratio determiner 313 and then the outputs from the determiner 313 of the initial ratio are modified by the generator 314 of multiple optimal selections, according to some rules in the knowledge base 631, while the rest is initialized by the corrector 318 of the random initial value. While, as a routine, the ratio prediction system and the method of obtaining mixtures of the present invention work satisfactorily even if the generator outputs 315 of the random initial value are only used for direct initialization as in the case of the standard genetic algorithm, the performance can be expected to be improved, due to the multiple optimal selections generated by the corrector 318 of the random initial value. The initial population, thus formed, in the manner described above, is fed to the unit 32 for determining the dynamic AG. When the output from the AG processor 3 is returned to the evaluation unit 2 and the value determined by the evaluation unit 2 is fed back to the AG processor 3, the count of the counter 311 is increased by one. Since the count of the counter 311 is no longer zero, the initialization of the initial value determination unit 31 takes place only once at the beginning. It will be noted that in the above description, the unit 31 that determines the initial value has been described as constructed and executes the three initialization methods, each method can be expected from place to the value of a good birth, compared to the initialization with the random value, as executed in the algorithm of the prior art. Therefore, although the three initialization methods are not performed simultaneously, it is expected that the forecasting performance and manufacturing performance will be high compared to the prior art proportions prediction system and the method of obtaining mixtures. . Although, also, the color space classifier 317 used in the embodiment of the present invention has been shown to be operable for the unclear classification of the color space, similar effects can be obtained even if a division of curled type is carried out, since there is a significance in that the initial value of a good birth is determined by the use of the knowledge of each area of color divided. To demonstrate quantitatively the effects of the present invention and the validity of the three components 61, 63 and 7 in the evaluation unit 2, the following five apparatuses (a) to (e) were compared experimentally. (a) The determiner 313 of the initial ratio in the form of a neural network. (The proportions forecasting apparatus is operable to predict the proportion with the use of the neural network, discussed previously in relation to the prior art.) (B) The proportions prediction apparatus in which the evaluation unit 2 includes only the evaluation unit 7 of the mix characteristic. (c) The proportions prediction apparatus, in which the evaluation unit 2 includes the unit 7 for evaluating the characteristic of the mixture and the section 63 for evaluating the unnecessary element number 63. (d) The prediction apparatus of proportions, in which the evaluation unit 2 includes the evaluation unit 7 of the characteristic of the mixture and the evaluation section 61 of the number of the element, (e) The proportions prediction apparatus, in which the unit of evaluation 2 includes all three components: the evaluation unit 7 of the characteristic of the mixture, the evaluation section 63 of the unnecessary element number and the evaluation section 61 of the element number.

During the experiments, 92 data randomly selected from the 302 data of the objective paints of the known formulations used as the test data were used and each of the apparatuses (a) to (e) were operated to predict the proportion of the pigments. elementary The prediction errors of the proportions are shown in the following table Thus, it is clear that the present invention has had superior effects, when the evaluation unit 2 has the three components. Although the above description of the present invention has been described as being applied to the production of a color paint, which generally corresponds to the color of the objective paint, prepared by mixing the elemental pigments, the present invention is not limited to the production of such physical mixture and can be equally applicable to the mixture of waves, such as color, light and sound. Likewise, although the described modalities are directed to physical mixtures, the present invention is equally applicable to mixtures of anything that the human being can perceive. As an example, in the cases of the color formulation based on the visually perceptible mixture, such as in a color printer, the proportions prediction apparatus of the present invention can operate to predict the proportion of surface areas of elementary colors spatially distributed. Also, in the case of a color formulation based on the visually perceptible mixture, such as a color display, the proportions prediction apparatus of the present invention can operate to predict the mixing ratio of the light intensities of the color elements spatially distributed. Similarly, the method for obtaining the mixture of the present invention can have a number of applications. INDUSTRIAL APPLICABILITY As described hereinabove, according to the present invention, the proportions in which the elements are to be mixed to produce the objective of a predetermined specification, can be predicted with high accuracy, without the need to make a real mixture of the elements. Consequently, a mixture that corresponds substantially to the purpose of the predefined specification can be easily prepared. This leads to sales-jas that the period of time required to make the forecast and manufacturing and the associated cost, can be reduced. In that way, the present invention has a number of industrial application forms.

Claims (19)

1. CLAIMS 1. A system of prediction of proportions, to forecast the elements to be mixed and their proportions in order to achieve a given objective, this system includes: an extractor of the characteristics of the objective, to extract a characteristic from a target predetermined; an evaluation element, which is operable by receiving proportion vectors, which are represented by the characteristics of the predetermined objective and the quantity of each of the elements, respectively, to determine the suitability of the proportion vectors, based on the characteristic extracted from the predetermined objective; and a genetic algorithm processor, to predict the proportion vectors based on their fitness, according to the genetic algorithm, in which the quantity of each of the elements and each of the proportion vectors are represented by a gene and one chromosome, respectively; these proportion vectors, which have been predicted by the processor of the genetic algorithm, are input to the evaluation element to cause this evaluation element to repeat the evaluation of the proportion vectors, to determine the optimal proportion vectors.
2. 2. The system, as claimed in claim 1, wherein the evaluation element is an evaluation element of the mixing element, which receives the proportions predicted by the processor of the genetic algorithm, to evaluate the types of elements whose proportions do not they are zero, based on the input values of the characteristic.
3. 3. The system, as claimed in claim 2, wherein the evaluation element of the mixing element includes a selector of the number of the proportion element, which receives the proportion vectors predicted by the genetic algorithm processor, and operates to digitize them, comparing them with a predetermined threshold value, in order to select the elements whose proportions are greater than that predetermined threshold value, a forecaster of the mixing element, which receives the characteristic extracted by the extractor of the objective characteristic, and a calculator of the distance of the element number, operable to compare an element number, obtained from the selector of the number of the mixing element, with an element number, obtained from the forecaster of the mixing element.
4. 4. The system, as claimed in claim 2, in which the evaluation element of the mixing element includes a knowledge base, in which unnecessary combinations for mixing the elements are described in the form of knowledge pieces. , and a penalty element, which receives the number of the element that will be used in the mix in the number selector of the mixing element and which refers to the knowledge base to decrease the fitness of one of the elements , in the case of the presence of the unnecessary combination of element numbers.
5. 5. The system, as claimed in claim 1, in which the evaluation element is an element for evaluating the characteristic of the mixture, operable to evaluate a characteristic of a mixture, which is formed by mixing the elements in accordance with the ratio predicted by the processor of the genetic algorithm, comparing the characteristic of the predetermined goal entered.
6. The system, as claimed in claim 5, in which the element of evaluation of the characteristic of the mixture includes a predictor characteristic of the mixture, operable upon receiving a vector of proportion, predicted by the processor of the genetic algorithm, for predict the characteristic of the mixture formed by mixing the elements according to the proportion given by the proportion vectors, and a distance calculator of the characteristic of the mixture, to compare the characteristic extracted by the extractor of the characteristic of the mixture with the characteristic of the mixture obtained from the forecaster of the characteristic of the mixture and produce a similarity as an aptitude.
7. 7. The system, as claimed in claim 1, in which the processor of the genetic algorithm includes a section for determining the initial value of this genetic algorithm, to determine an initial value by the proportion of each of the elements to be predicted. , and a dynamic processor of the genetic algorithm, operable on the basis of this genetic algorithm to determine the proportions in sequence, and where the section for determining the initial value of the genetic algorithm includes at least one element for determining the initial proportion value, adapted for receive the characteristic of the objective and operable to adjust the proportion of the elements that form the objective to be an initial value.
8. The system, as claimed in claim 1, wherein the genetic algorithm processor includes a section for determining the initial value of the genetic algorithm to determine an initial value for the proportion of each of the elements to be predicted. , and a dynamic processor of the genetic algorithm, operable on the basis of this genetic algorithm, to determine in sequence the proportions, and in which the section for determining the initial value of the genetic algorithm includes at least one element for determining the initial proportion value, adapted to receive the characteristic of the objective and operable to adjust the proportion of the elements that form the target that is going to be an initial value, a knowledge base, in which the combinations related to the knowledge of the elements are described, and a generator of multiple optimal selections, to compare an output of the value determination section of initial proportion with the knowledge in the knowledge base and to adjust new initial values for the processor of the genetic algorithm, replacing the proportion of a candidate of unnecessary element with zero.
9. 9. The system, as claimed in claim 1, wherein the target is a color and where the processor of the genetic algorithm includes a section for determining the initial value of the genetic algorithm, to determine an initial value for the proportions of the elements to be predicted and a dynamic processor of the genetic algorithm, operable on a base of this genetic algorithm to determine in sequence the proportions, and where the section of determination of the initial value of the genetic algorithm includes an extractor of the space characteristic of color to produce coordinates in a color space, such as L * -a * -b *, of the color of the lens, a color space sorter, adapted to receive the coordinates in the color space obtained by the extractor of the color space characteristic, to determine which color area representative of the objective is included, a knowledge base, in which the combinations Labeled with the color knowledge of the elements are described, a random initial value generator, to produce random proportions, and a random initial value corrector, adapted to receive the ratio generated by the random initial value generator and the information in the representative color area determined by the color space classifier and operable to do a search in the knowledge base, to determine if an inappropriate color combination in the color area is included in a random value of the proportion obtained from the value initial random and to correct the inappropriate random proportion.
10. The system, as claimed in claim 9, wherein the color space classifier is operable for the indistinct classification of the color space, to provide a plurality of outputs representative of the degrees of membership of one or or more representative color areas, to which the color of the target material and the color of the objective belong.
11. 11. The system, as claimed in claim 9, wherein the color space classifier is operable for the indistinct classification of the color space, to provide a plurality of outputs representative of the degrees of membership to one or two or more representative color areas, to which the objective color belongs, and where the correct initial random value is operable to determine the number of candidates of the proportions of the initial value given randomly, according to the degrees of belonging.
12. 12. A method for preparing a predetermined target by mixing a predetermined number of elements in a predetermined proportion, this method comprises the steps of: extracting a characteristic from the predetermined target; evaluate the proportion vectors having vector elements represented by respective amounts of the elements, based on the characteristic extracted from the objective; forecast the proportion vectors, based on fitness according to a genetic algorithm in which the quantity of each of the elements and each of the proportion vectors are represented by a gene and a chromosome, respectively; and determine the optimal proportions required to prepare the objective, repeating the evaluation and prediction stages.
13. 13. The method, as claimed in claim 12, in which the evaluation step is a step of evaluating the mixing element, which receives the proportions predicted by the processor of the genetic algorithm, to evaluate the type of the elements whose proportions are not zero, on the basis of the values of the input characteristics.
14. 14. The method, as claimed in claim 13, wherein the step of evaluating the mixing element includes a step of selecting a number of the proportion element, which receives the proportioned vectors predicted during the step of prediction and operates to digitalize them by comparing them with a predetermined threshold value, so as to select the elements whose proportions are greater than the predetermined threshold value, a prediction step of the mixing element, which receives the extracted characteristic, and a calculation step. of the distance of the element number for comparing an element number obtained from the selector of the mixing element number with an element number obtained during the prediction step of the mixing element.
15. 15. The method, as claimed in claim 13, wherein the step of evaluating the mixing element includes a knowledge base, in which unnecessary combinations of mixing of the elements are described in the form of knowledge pieces, and a step of penalization, which receives the number of the element to be used, mixed in the stage of selection of the number of the element of mixture and which refers to the knowledge base to decrease the fitness of one of the elements in the case of the presence of the unnecessary combination of element numbers.
16. 16. The method, as claimed in claim 12, wherein the evaluation step is a step of evaluating the characteristic of the mixture, to evaluate a characteristic of a mixture, which is formed by mixing the elements according to the Proportioned proportion during the prediction stage, comparing the characteristic of the predetermined goal entered.
17. 17. The method, as claimed in claim 16, wherein the step of evaluating the characteristic of the mixture includes a step of predicting the characteristic of the mixture, operable upon receiving the predicted proportion vector during the step of prediction, to forecast the characteristic of the mixture formed by mixing the elements according to the proportion given by the proportion vectors, and a step of calculating the distance of the characteristic of the mixture, to compare the characteristics extracted during the stage of extraction of the characteristic of the mixture, with the characteristic of the mixture obtained from the forecaster of the characteristic of the mixture and to produce a similarity as an aptitude.
18. 18. The method, as claimed in claim 12, wherein the prediction step includes a step of determining the initial value of the genetic algorithm, to determine an initial value for the proportion of each of the elements to be predicted. , and a step of dynamic process of the genetic algorithm, operable on the basis of this genetic algorithm, to determine in sequence the proportions and in which the step of determining the initial value of the genetic algorithm includes at least a step of determining the initial proportion value , adapted to receive the characteristic of the objective and operable to adjust the proportion of the elements that form the target that is going to be an initial value. The method, as claimed in claim 12, wherein the step of calculating the genetic algorithm includes a step of determining the initial value of the genetic algorithm, to determine an initial value for the proportion of each of the elements that is they will predict, and a stage of dynamic process of the genetic algorithm, operable on a base of this genetic algorithm, to determine in sequence the proportions, and where this stage of determination of the initial value of the genetic algorithm includes at least one stage of determination of the initial proportion value, adapted to receive the characteristic of the objective and operable to adjust the proportion of the elements that form the target that is going to be the initial value, and a stage of generation of multiple optimal selections, to compare an output of the stage of determining the value of the initial proportion with knowledge in the knowledge base, in which the combination The items related to the knowledge of the elements are described, and to adjust new initial values for the calculation stage of the genetic algorithm, replacing the proportion of a candidate of unnecessary element with zero.