JP5713707B2 - Stroke generation device, stroke generation method, stroke generation program, character generation device, character generation method, and character generation program - Google Patents

Description

  The present invention relates to a stroke generation device, a stroke generation method, a stroke generation program, a character generation device, a character generation method, and a character generation program according to the present invention, and more particularly, based on the same character written multiple times by a writer. Stroke generation device, stroke generation method, and stroke generation program capable of generating strokes or characters having “transition” that each stroke has a different shape while having “individuality” indicating that The present invention relates to a character generation device, a character generation method, and a character generation program.

  Today, it is common practice to create text on a computer using word processing software or the like. For this reason, when printing a prepared sentence, a predetermined font is selected by the user from fonts (fonts) prepared in advance, and characters are printed by the selected font.

  However, the typeface used for printing is often a well-balanced and easy-to-read typeface prepared in advance. For this reason, it has been difficult to express the characteristics and atmosphere of the text in a typeface according to the genre of the writer or the novel. In order to meet such demands, many fonts resembling handwritten characters have been proposed today.

  However, even if the typeface is similar to a handwritten character, if the typeface is a typeface prepared in advance, the same character is repeatedly printed as the same typeface (character of the same typeface). On the other hand, in a sentence actually written by a human hand, even if the same character is written multiple times, each character has a common personality that can be recognized as the writer's character, and each character is subtly Often different.

  The similarity that can be recognized that the written characters are written by the same writer is called “individuality”, and even characters with this “individuality” are subtle when compared in detail. The degree of diversity showing different shapes is called “transition”.

  In this way, characters prepared in advance as a typeface do not show “individuality” because the difference between the writers cannot be recognized, and furthermore, the same character becomes the same character (a character of the same typeface), and thus “transition”. It does not have.

  On the other hand, a method of using characters written by a writer with almost no deformation has been proposed (for example, see Non-Patent Documents 1 to 3). These methods are mainly used to focus the connection between characters and strokes, and have high personality, but very low when the number of input character data is small It showed displaceability.

  Further, a method has been proposed in which a character is modeled using a Hidden Markov model (HMM) or a Bayesian network, which is a kind of statistical probability model, to generate a character with the highest probability (see Non-Patent Documents 4 to 6). Characters generated by the Hidden Markov model can generate characters that can be read by humans, and characters generated by the Bayesian network generate characters that have shapes that are written by humans. It was possible to do. Characters generated by these methods can generate characters that have a high probability of being recognized as characters written by the writer, so that they can show an intermediate “individuality”, but “transition” is It was low.

  Furthermore, a method for transforming characters into Global has been proposed (see Non-Patent Documents 7 to 13). Transforming a character into Global means performing a transformation that makes the relationship with other characters common in shape. For example, the case where the character is changed into a slanted character (an oblique character) or an emphasized character and the entire character is deformed in common instead of changing a part of the character is applicable.

  In this method, a character is generated by changing the writing speed of a character, modeling an independent character or a relationship between two characters, and changing parameters. The method of transforming characters into Global was not considered for partial deformation of characters (Local deformation), and was developed only for the purpose of giving Global deformation. Characters generated by using the method of transforming characters into Global have the characteristics of exhibiting intermediate or higher personality as compared to other methods, and exhibiting intermediate transitional properties.

  In addition, a method of generating characters of various shapes using a genetic algorithm has been proposed (see, for example, Non-Patent Document 14). By using a genetic algorithm, it is possible to generate characters of very various shapes. However, characters generated using a genetic algorithm have not only lost personality, but also have a feature that characters that are distorted so as to be difficult to read may be generated.

Isabelle Guyon, "Handwriting synthesis from handwritten glyphs", Proc. IWFHR'96, p.309-312, 1996 C.V.Jawahar, A. Balasubramanian, "Synthesis of online handwriting in Indian languages", Proc. IWFHR'06, p.108-113, 2006 Z.Lin, L.Wan, "Style-preserving English handwriting synthesis", Pattern Recognition, Vol. 40, Issue 7, p.2097-2109, 2007 B.-K.Sin, J.H. Kim, "Network-based approach to Korean handwriting analysis", Int. Journal of Pattern Recognition and Artificial Intelligence, Vol.12, No.2, p.233-249, 1998 H. Choi, S.-J. Cho, J.H. Kim, "Generation of handwritten characters with Bayesian network based on-line handwriting recognizers", Proc. ICDAR'03, vol. 2, p.995-999, 2003 H. Choi, J.H. Kim, "Probabilistic synthesis of personal-style handwriting", IEICE Transactions on Information and Systems, Vol. E92-D No.4, p.653-661, 2009 S. Setlur, V. Govindaraju, "Generating manifold samples from a handwritten word", Pattern Recognition Letters, Vol. 15, Issue 9, p.901-905, September 1994 J. Wang, C. Wu, Y.-Q. Xu, H.-Y. Shum, "Combining shape and physical models for online cursive handwriting synthesis", Int. Journal on Document Analysis and Recognition, Vol. 7, Issue 4 , p.1433-2833, 2005 L. Devroye, M. McDougall, "Random fonts for the simulation of handwriting", Electronic Publishing, Vol. 8, No. 4, p.281-294, 1995 M. Mori, A. Suzuki, A. Shio, S. Ohtsuka, "Generating new samples from handwritten numerals based on point correspondence", Proc. IWFHR'00, p. 281-290, 2000 Y. Zheng, D. Doermann, "Handwriting matching and its application to handwriting synthesis", Proc. ICDAR'05, p.861-865, 2005 R. Plamondon, W. Guerfali, "The generation of handwriting with delta-lognormal synergies", Biological Cybernetics, Vol. 78, p.119-132, 1998 D.-H. Lee, H.-G. Cho, "A new synthesizing method for handwriting Korean scripts", Int. Journal of Pattern Recognition and AI 12 (11), p.45-61, 1998 Y. Wada, K. Ohkawa, K. Sumita, "Generation of diversiform characters using a computational handwriting model and a genetic algorithm", ICANN'01, LNCS. Springer, Hedelberg, Volume 2130, p.1217-1224, 2001

  However, in the conventional handwritten character generation method as described above, several writing styles cannot be expressed by one model. Specifically, in the case of generating characters with “transition” that have different shapes while having “individuality”, the conventional handwritten character generation method can express the transition. There was a problem that a plurality of character models had to be prepared according to the variety of transmutability. For this reason, it is possible to generate a variety of handwritten characters (characters with “transition”) while maintaining the basic style of characters written by the writer (ie, maintaining “individuality”). The conventional handwritten character generation method has a problem that it is difficult.

  The present invention has been made in view of the above-mentioned problems, and a plurality of the same characters written by a writer are used as samples, while having “individuality” that can be recognized as characters written by the same writer, Stroke generating device, stroke generating method, stroke generating program, and character generating device capable of generating more strokes or characters having a “transition” that causes a variety of character shapes than the number of samples An object is to provide a character generation method and a character generation program.

  In order to solve the above-described problems, a stroke generation device according to the present invention includes a stroke decomposing unit that decomposes a character structure for each stroke based on information on the trajectory and writing order of the same character, and the stroke decomposing unit. From each other feature point in the stroke to the one feature point, the feature point extracting means for extracting the feature point characterizing the shape of the stroke for each stroke and the one feature point of the stroke as a reference An influence vector set extracting means for obtaining all the feature points in the stroke, a set of the influence vectors obtained by the influence vector set extracting means, and Of the influence vector at all feature points of all strokes of the same character written Clustering means for classifying the set of effect vectors at the feature points by clustering each feature point of the stroke constituting the same part of the same character written multiple times, and the one feature point The position determination pattern for determining the position of the next feature point is set according to the number of classes of the set of the influence vectors at the one feature point classified by the clustering means, and all the features are set. Model generation means for generating a feature point position determination model based on the point position determination pattern, and a position determination pattern for determining the position of the next feature point according to the feature point position determination model generated by the model generation means By using the set of influence vectors classified by the clustering means as Characterized in that it comprises a stroke generating means for generating a stroke and determine the position of each feature point.

  In the stroke generation method according to the present invention, the stroke decomposition means decomposes the structure of the character for each stroke based on the information on the trajectory and writing order of the same character, and is decomposed in the stroke decomposition step. For each stroke, the feature point extracting means extracts a feature point that characterizes the shape of the stroke, and from all other feature points in the stroke based on one feature point of the stroke, The influence vector set extraction means obtains a set of influence vectors indicating the distance and direction to one feature point for all feature points in the stroke, and is obtained in the influence vector set extraction step. A set of the influence vectors written multiple times By clustering the set of influence vectors at all feature points of all strokes for each same character for each stroke feature point constituting the same part of the same character written multiple times, the clustering means, A clustering step for classifying the set of influence vectors at the feature point, and a position determination pattern for determining a position of the next feature point from the one feature point by the model generation unit is classified at the clustering step. A model generation step of setting several patterns according to the number of classes of the set of the influence vectors at the one feature point, and generating a feature point position determination model based on the position determination patterns of all the feature points; According to the feature point position determination model generated in the generation step, Stroke generation in which stroke generation means determines the position of each feature point and generates a stroke by using the set of influence vectors classified in the clustering step as a position determination pattern for determining the position of the point And a step.

  Further, the stroke generation program according to the present invention is capable of generating a stroke similar to the stroke constituting the character by using information on the trajectory and writing order of the same character written a plurality of times. And, based on the information on the trajectory and writing order of the same character, the computer decomposes the structure of the character for each stroke, and for each stroke decomposed by the stroke decomposition function, A feature point extraction function for extracting a feature point that characterizes a shape, and an influence vector that indicates the distance and direction from all other feature points in the stroke to the one feature point based on one feature point of the stroke Is obtained for all feature points in the stroke. An influence vector set extraction function, and a set of the influence vectors obtained by the influence vector set extraction function, wherein the influence vectors at all feature points of all strokes of the same character written a plurality of times A clustering function for classifying the set of the influence vectors at the feature point by clustering the set for each feature point of the stroke constituting the same part of the same character written a plurality of times, and the one feature point The position determination pattern for determining the position of the next feature point is set according to the number of classes of the set of the effect vectors at the one feature point classified by the clustering function, and all the features are set. A model generation function that generates a feature point position determination model based on a point position determination pattern; By using the set of influence vectors classified by the clustering function as a position determination pattern for determining the position of the next feature point according to the feature point position determination model generated by the model generation function, It is a program for realizing a stroke generation function for generating a stroke for determining the position of a point.

  Note that the characters described above are not limited to general characters such as kanji and hiragana, but include symbols and marks. A stroke corresponds to a point or a line constituting a character. In addition, as a feature point position determination model, for example, a model represented by a structure similar to a tree structure by a set of position determination patterns, or the relationship between each feature point is indicated according to the number of position determination pattern patterns for each feature point. An example is a model made up of a table.

  As described above, in the stroke generating device, the stroke generating method, and the stroke generating program according to the present invention, after decomposing the character structure for each stroke, the feature points characterizing the shape of the stroke are extracted, and one of the corresponding strokes is extracted. A set of influence vectors from other feature points indicating the position of the feature point is obtained. Here, since the decomposed stroke is obtained for each same character written a plurality of times, a plurality of influence vector sets are obtained for each feature point of the stroke constituting the same part of the same character written a plurality of times. Can do. For this reason, in the stroke generation device, the stroke generation method, and the stroke generation program according to the present invention, by clustering a set of a plurality of influence vectors for each feature point of the stroke constituting the same part of the same character written a plurality of times The feature point position determination model for determining the position of the feature point can be obtained based on the position determination pattern.

  In this feature point position determination model, several patterns of branches (position determination patterns) are formed for each feature point, so that it is possible to generate more types of strokes than the number of the same characters written by the writer. Furthermore, since the strokes generated in this way are obtained based on information on the trajectory and writing order of characters actually written by the writer, they have “individuality” indicating that they are the characters of the writer. Furthermore, while having “individuality”, it is possible to have “transition” in which each stroke has a different shape.

  Further, the stroke generation device described above includes an influence point extraction unit that obtains, as an influence point, another feature point located in a trajectory before the one feature point based on one feature point of the stroke, The influence vector set extraction unit may obtain an influence vector indicating a distance and a direction from the influence point to the one feature point at all the influence points in the one feature point.

  Further, in the stroke generation method described above, the influence point extraction unit determines, as an influence point, another feature point located on a trajectory before the one feature point based on one feature point of the stroke. An extraction step, wherein in the influence vector set extraction step, the influence vector set extraction means displays an influence vector indicating a distance and a direction from the influence point to the one feature point for all of the one feature point. You may obtain | require in an influence point.

  Furthermore, the stroke generation program described above causes the computer to determine, as an influence point, another feature point located in a locus before the one feature point with reference to one feature point of the stroke. In the influence vector set extraction function, an influence vector indicating the distance and direction from the influence point to the one feature point is obtained at all influence points in the one feature point. May be.

  When a writer actually writes a character, there is a tendency to write the character by determining the shape, size, and balance of the character and stroke in consideration of the stroke trajectory just before the writing portion. For this reason, the position of the feature point that characterizes the shape of the stroke can also be greatly influenced by the feature point located on the locus before the feature point.

  In the stroke generation device, the stroke generation method, and the stroke generation program according to the present invention, as described above, with reference to one feature point of the stroke, other feature points located in the locus before the one feature point are obtained. As an influence point, an influence vector indicating the distance and direction from the influence point to one feature point is obtained at all the influence points in the one feature point. In this way, by obtaining an influence vector that determines the position of a feature point with reference to another feature point located on a trajectory before one feature point, the feature point determined based on the set of effect vectors The position is determined in a state that reflects the characteristics of the actual writer writing the character.

  Further, the stroke generation device described above is another feature point located on a trajectory before the one feature point on the basis of one feature point of the stroke, and the one feature point and the other feature point An influence point extraction unit that obtains another feature point that is not obstructed by a trajectory of any stroke as an influence point is provided, and the influence vector set extraction unit includes the one feature from the influence point. You may obtain | require the influence vector which shows the distance and direction to a point in all the influence points in the said one feature point.

  Further, the stroke generation method is characterized in that the influence point extracting means is another feature point located in a locus before the one feature point, with the one feature point of the stroke as a reference. An influence point extracting step of obtaining, as an influence point, another feature point that is not obstructed by a trajectory of any stroke of a line connecting the point and the other feature point, and in the influence vector set extraction step, the influence vector The set extraction unit may obtain an influence vector indicating a distance and a direction from the influence point to the one feature point at all the influence points in the one feature point.

  Further, the stroke generation program described above may cause the computer to have another feature point located on a locus before the one feature point with respect to the one feature point of the stroke, and the one feature point. And an influence point extraction function that obtains other feature points that are not obstructed by the trajectory of any stroke as an influence point. In the influence vector set extraction function, An influence vector indicating a distance and a direction from a point to the one feature point may be obtained at all the influence points in the one feature point.

  When a writer actually writes a character, there is a tendency to write the character by determining the shape, size, and balance of the character and stroke in consideration of the stroke trajectory just before the writing portion. In addition, when a writer writes a character, there is a tendency to judge the shape, size, and balance of the character greatly influenced by the locus of the portion close to the portion being written. There is a tendency that it is not easily affected by strokes (such as character dots and lines).

  For this reason, in the stroke generation device, the stroke generation method, and the stroke generation program according to the present invention, as described above, with reference to one feature point of the stroke, the other feature located in the locus before the one feature point The distance from the influence point to the one feature point is obtained as another feature point that is a point and the line connecting the one feature point and the other feature point is not obstructed by the trajectory of any stroke. And the influence vector indicating the direction are obtained at all the influence points in one feature point. In this way, by determining the influence vector, the position of the feature point determined based on the set of influence vectors is determined in a state reflecting the characteristics when the actual writer writes the character. .

  In addition, when the number of feature points differs for each of a plurality of strokes constituting the same part of the same character written a plurality of times, the stroke generation device described above performs a feature point addition process for each stroke, and the feature points It may be provided with a feature point adding means for making the number of the same.

  Further, in the above-described stroke generation method, when the number of feature points is different for each of a plurality of strokes constituting the same part of the same character written a plurality of times, the feature point adding unit adds the feature points for each stroke. You may provide the feature point addition step which processes and makes the number of feature points the same.

  In addition, the stroke generation program described above performs addition processing of feature points for each stroke when the number of feature points is different for each of a plurality of strokes constituting the same part of the same character written multiple times on the computer. It is also possible to realize a feature point addition function for performing the same number of feature points.

  The position of the feature point of the stroke that constitutes the same part of the same character written multiple times by the same writer is often common to each written character, but the written character is not a copy of the exact same character each time Therefore, the number of feature points may be different for each stroke.

  In this way, when the number of feature points of the strokes constituting the same part of the same character written multiple times is different, it is difficult to generate a feature point position determination model by a set of clustered effect vectors. Become. For this reason, in the stroke generation device, the stroke generation method, and the stroke generation program according to the present invention, when the number of feature points differs for each of a plurality of strokes constituting the same portion of the same character written a plurality of times, By performing the feature point addition process, it is possible to perform the process of making the number of feature points the same, and to easily generate the feature point position determination model.

  In addition, the character generation device according to the present invention is based on the stroke generation device described above and the strokes constituting the different parts of the same character, which are decomposed by the stroke decomposition means, as a reference, A set of inter-stroke vectors indicating distances and directions from all feature points of other strokes different from the one stroke to the feature points of the one stroke are all strokes for the same character written multiple times. An inter-stroke vector set extracting means to be obtained for every feature point in the above, and an inter-stroke vector of another stroke constituting the different parts of the same character with the size of one stroke generated by the stroke generating means. And a set of influence vectors in the same stroke that is the same character and constitutes the same part The size adjustment means for adjusting based on the above and the position of the stroke whose size is adjusted by the size adjustment means is set to a set of vectors between strokes in other strokes which are the same character and constitute different parts. And a position adjusting means for adjusting based on this.

  Further, the character generation method according to the present invention is based on the stroke generation method described above and the stroke constituting the different parts of the same character, which is decomposed in the stroke decomposition step, as a reference, An inter-stroke vector set extraction means is written a plurality of times as a set of inter-stroke vectors indicating distances and directions from all feature points of other strokes different from the one stroke to the feature points of the one stroke. In addition, the size adjustment means determines that the size of one stroke generated in the stroke vector set extraction step for each feature point of all the strokes of the same character for each feature point and the stroke generation step is the same character. The same character as the set of inter-stroke vectors of other strokes that make up different parts. The position adjustment means adjusts the size adjustment step for adjusting based on the set of influence vectors in the same stroke constituting the same part, and the position of the stroke subjected to the size adjustment by the size adjustment step, And a position adjustment step of adjusting based on a set of vectors between strokes in other strokes that are the same character and constitute different parts.

  In addition, the above-described character generation program has one stroke between a function realized by the above-described stroke generation program and a stroke constituting a different part of the same character decomposed by the stroke decomposition function. A set of inter-stroke vectors indicating the distance and direction from all feature points of other strokes different from the one stroke to the feature point of the one stroke, with the feature point as a reference, and written multiple times The inter-stroke vector set extraction function for obtaining all the feature points of all the strokes for each character and the size of one stroke generated by the stroke generation function are the same character and constitute different parts The same character and the same part as the set of inter-stroke vectors of other strokes The size adjustment function to be adjusted based on the set of influence vectors in one stroke, and the position of the stroke whose size has been adjusted by the size adjustment function are the same character and constitute different parts It is a character generation program for realizing the position adjustment function to be adjusted based on the set of vectors between strokes in the strokes.

  As described above, in the character generation device, the character generation method, and the character generation program according to the present invention, an inter-stroke vector indicating a positional relationship between different strokes is obtained. This inter-stroke vector is a vector indicating the distance and direction from all feature points of other strokes different from the one stroke to the feature point of the one stroke, with reference to the feature point of one stroke. . By obtaining an inter-stroke vector for every feature point, it is possible to obtain a relative positional relationship based on feature points from other strokes at the feature point level of the stroke.

  In the character generation device, the character generation method, and the character generation program according to the present invention, when adjusting the stroke size, the size adjustment is performed based on the set of inter-stroke vectors and the set of influence vectors. This is because when adjusting the size of the stroke, not only the relative relationship from other strokes but also the influence on the feature points in the same stroke is affected by the size determination.

  On the other hand, in the character generation device, the character generation method, and the character generation program according to the present invention, when adjusting the position of the stroke, the position is adjusted based on a set of vectors between strokes. When adjusting the position, the relative relationship with other strokes has a significant effect, and the relationship between feature points within the same stroke is less affected than the relative relationship with other strokes. It is a thing.

  Further, the character generation device described above is based on the feature point of one stroke between strokes constituting different parts of the same character decomposed by the stroke decomposing means, and is more in order of writing than the one stroke. The inter-stroke influence point extraction means for obtaining the characteristic points of the other strokes positioned before as the influence point between strokes, and the inter-stroke vector set extraction means includes all the stroke influence point extraction means obtained by the inter-stroke influence point extraction means. A set of inter-stroke vectors indicating the distance and direction from the inter-stroke influence point to the feature point of the one stroke is obtained for every feature point in all strokes of the same character written multiple times. May be.

  Further, in the character generation method described above, the inter-stroke influence point extracting means uses the feature point of one stroke as a reference between the strokes constituting different parts of the same character decomposed in the stroke decomposition step. An inter-stroke influence point extracting step for obtaining a feature point of another stroke positioned in a stroke order before one stroke as an inter-stroke influence point, wherein the inter-stroke vector set extracting step includes: The extraction means is the same in which a set of inter-stroke vectors indicating the distances and directions from all the inter-stroke influence points obtained in the inter-stroke influence point extracting step to the feature points of the one stroke is written a plurality of times. Even for every feature point of every stroke of every character There.

  Further, the character generation program described above is based on the feature point of one stroke between strokes constituting different parts of the same character decomposed by the stroke decomposition function in the computer. Also realizes an inter-stroke influence point extraction function for obtaining the characteristic points of other strokes positioned in the previous stroke order as inter-stroke influence points, and the inter-stroke vector point extraction function includes the inter-stroke influence point extraction function. A set of inter-stroke vectors indicating the distance and direction from all the inter-stroke influence points determined by the above to the feature point of the one stroke, all feature points of all strokes of the same character written multiple times It may be determined every time.

  When a writer actually writes a character, there is a tendency to write the character by determining the shape, size, and balance of the character and stroke in consideration of the stroke trajectory just before the writing portion. For this reason, the inter-stroke influence point of the inter-stroke vector indicating the positional relationship between different strokes can also be greatly influenced by the feature points of other strokes positioned in the stroke order before one stroke.

  In the character generation device, the character generation method, and the character generation program according to the present invention, as described above, between the strokes constituting different parts of the same character, based on the feature point of one stroke, Also, the feature points of other strokes located in the previous stroke order are obtained as inter-stroke influence points, and a set of inter-stroke vectors indicating the distance and direction from the inter-stroke influence point to the one stroke feature point, It is calculated for every feature point of every stroke written for the same character written multiple times. In this way, a set of inter-stroke vectors is obtained by obtaining an inter-stroke vector for adjusting the size and position of the stroke with reference to the feature points of other strokes positioned in the stroke order before the one stroke. The size and position of the stroke determined based on the above are determined in a state reflecting the characteristics of the actual writer writing the character.

  The character generation device described above uses a feature point of one stroke as a reference between strokes constituting different parts of the same character decomposed by the stroke decomposing means, and precedes the stroke in order of writing. Any one of the above-mentioned feature points, wherein a line connecting any one of the feature points of the other stroke and the feature point of the one stroke is not obstructed by the locus of any stroke. Is obtained as an inter-stroke influence point extraction means, and the inter-stroke vector set extraction means is characterized by the one stroke from all the inter-stroke influence points obtained by the inter-stroke influence point extraction means. A set of inter-stroke vectors that indicate the distance and direction to the point is all strokes for the same character written multiple times. Or it may be determined for each of all feature points in over click.

  Further, in the character generation method described above, the inter-stroke influence point extracting means uses the feature point of one stroke as a reference between the strokes constituting different parts of the same character decomposed in the stroke decomposition step. Another stroke located in the stroke order before one stroke, and a line connecting any feature point of the other stroke and the feature point of the one stroke is the locus of any stroke An inter-stroke influence point extracting step for obtaining any one of the feature points that is not obstructed by the inter-stroke influence point, wherein the inter-stroke vector set extracting means includes the inter-stroke effect set in the inter-stroke vector set extracting step. The one stroke is calculated from all the inter-stroke influence points obtained in the extraction step. A set of strokes between vector indicating the distance and direction to the feature points over click, or may be determined for each of all the feature points of all strokes for each of a plurality of times written same character.

  Further, the character generation program described above is based on the feature point of one stroke between the strokes constituting different parts of the same character decomposed by the stroke decomposition function in the computer. Also, other strokes that are positioned forward in the stroke order, and the line connecting any one of the feature points of the other stroke and the feature point of the one stroke is not obstructed by the trajectory of any stroke. An inter-stroke influence point extraction function for obtaining any one of the feature points as an inter-stroke influence point is realized, and all the inter-stroke influences obtained by the inter-stroke influence point extraction function in the inter-stroke vector set extraction function An inter-stroke vector indicating the distance and direction from the point to the feature point of the one stroke Assemble, it may be one cause determined for each all the feature points of all strokes for each of a plurality of times written same character.

  When a writer actually writes a character, there is a tendency to write the character by determining the shape, size, and balance of the character and stroke in consideration of the stroke trajectory just before the writing portion. In addition, when a writer writes a character, it tends to judge the shape, size, and balance of the character by being greatly influenced by other strokes close to the writing location, and is obstructed by the trajectory of the other strokes. There is a tendency that it is not easily affected by the stroke of a remote place.

  For this reason, in the character generation device, the character generation method, and the character generation program according to the present invention, as described above, the feature point of one stroke is used as a reference, and the character generation device, the character generation method, and the character generation program are positioned before the one stroke in the order of writing. The stroke influence point between strokes is a feature point that is not obstructed by the trajectory of any stroke. A set of inter-stroke vectors indicating the distance and direction from the inter-stroke influence point to the feature point of one stroke is obtained for every feature point of all strokes of the same character written multiple times. Yes. In this way, the feature points of other strokes located in the stroke order before one stroke, and the line connecting any feature point of another stroke and the feature point of one stroke The stroke size determined on the basis of the set of inter-stroke vectors by determining the stroke size and the stroke-to-stroke vector based on any feature point that is not obstructed by the stroke trajectory. The height and position are determined in a state reflecting the characteristics of the actual writer writing the character.

  As described above, the stroke generation device, the stroke generation method, the stroke generation program, the character generation device, the character generation method, and the character generation program according to the present invention characterize the shape of the stroke after decomposing the character structure for each stroke. A feature point is extracted, and a set of influence vectors from other feature points indicating the position of one feature point of the corresponding stroke is obtained. Since the decomposed stroke is obtained for each same character written a plurality of times, a set of a plurality of influence vectors can be obtained for each feature point of the stroke constituting the same part of the same character written a plurality of times. For this reason, a feature point position determination model for determining the position of a feature point is obtained by clustering a set of a plurality of effect vectors for each feature point of a stroke constituting the same part of the same character written multiple times. It becomes possible to obtain based on the position determination pattern.

  In this feature point position determination model, a branch point is formed for each feature point, so that it is possible to generate more types of strokes than the number of the same characters written by the writer. Furthermore, since the strokes generated in this way are obtained based on information on the trajectory and writing order of characters actually written by the writer, they have “individuality” indicating that they are the characters of the writer. Furthermore, while having “individuality”, it is possible to have “transition” in which each stroke has a different shape.

It is the block diagram which showed schematic structure of the handwritten character production | generation apparatus which concerns on embodiment. It is the flowchart which showed the process sequence of the production | generation process of the handwritten character performed by CPU which concerns on embodiment. It is explanatory drawing which showed typically the processing content in the production | generation process of a handwritten character. It is a figure for demonstrating the process which matches the number of feature points by comparing two strokes. It is a figure for demonstrating the process which compares the some stroke and matches the number of feature points. (A) is the figure which showed the feature point in the character of "2" comprised by one stroke with seven feature points (P0-P6), (b) is the feature point of P6 to P2. It is the figure shown with the influence vector from the feature point from to P5. (A)-(c) is the figure which showed the clustering result (class 0-class 2) of the influence vector in the "P6" point of the stroke shown in FIG. (A) is the figure which showed an example of the model which shows the position determination pattern of the feature point which comprises a stroke, (b) is a table instead of the model shown to (a), and shows the position determination pattern of the feature point in a table | surface. FIG. (A)-(c) is the case where the writer wrote the same character composed of four strokes three times, and between the strokes from the inter-stroke influence point of each stroke W1 to W3 to the feature point of the stroke W4. It is the figure which showed the vector. (A)-(c) is a figure for demonstrating the process in which the position of the last feature point "P6" of the number "2" is determined. (A)-(d) is a figure for demonstrating the magnitude | size adjustment process of a stroke. (A)-(d) is a figure for demonstrating the position adjustment process of a stroke. (A) shows a vertically-written character string consisting of “2 gmx four-way snow charming right stone stone” written by a writer five times, and (b) shows 10 columns generated based on the character string of (a). The minute string is shown.

  Hereinafter, as an example of a character generation device according to the present invention, a handwritten character generation device will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram illustrating a schematic configuration of a handwritten character generation apparatus according to an embodiment.

  As shown in FIG. 1, the handwritten character generation device 1 includes a display unit 11, an input operation unit 12, and a main body unit 13.

  The display unit 11 has a role of displaying the processing content in the main body unit 13 so as to be visible to the user. A general display device such as a liquid crystal display or a CRT display is used for the display unit 11.

  The input operation unit 12 is input means operated by a user when inputting data or the like to the main body unit 13. In the main body unit 13 according to the present embodiment, a tablet for inputting handwritten characters is used as an example. The tablet includes an electronic pen 12a and a screen unit 12b that reads the coordinate position of the pen tip written by the electronic pen 12a.

  The user writes characters on the screen unit 12b with the pen tip of the electronic pen 12a as if writing the characters with a writing instrument, so that each stroke of the characters (dots and lines constituting the characters are applicable). In the handwritten character generating apparatus 1 according to the embodiment, after the pen tip of the electronic pen 12a is brought into contact with the screen unit 12b, the pen point moves on the screen unit 12b until the pen tip moves away from the screen unit 12b. The coordinate of the pen tip movement state obtained by a series of operations is defined as one stroke). The sample character for generating the handwritten character by the handwritten character generation device 1 according to the present embodiment is based on the movement state of the electronic pen 12a operated by the user via the screen unit 12b of the input operation unit 12. To be acquired.

  In FIG. 1, only the tablet is shown as the input operation unit 12, but the input operation unit 12 is not limited to the tablet, and includes devices such as a general keyboard, mouse, and trackball. Can do.

  The main body unit 13 includes a control unit 20 and a data recording unit 21.

  The data recording unit 21 includes a general auxiliary storage device such as a hard disk or an SSD (Solid State Drive). The data recording unit 21 stores a program related to a handwritten character generation process executed by the control unit 20. The data recording unit 21 records sample characters acquired by the input operation unit 12. Further, the data recording unit 21 records various information recorded as necessary in the process of generating handwritten characters (for example, information on a statistical model (feature point position determination model) described later). It is possible.

  The control unit 20 has a role of performing handwritten character generation processing, which will be described later, in accordance with a program recorded in the data recording unit 21. Specifically, the control unit 20 is a CPU (Central Processing Unit: stroke decomposing unit, stroke generating unit, inter-stroke vector set extracting unit, inter-stroke influence point extracting unit, influence point extracting unit, feature point adding unit, feature point extracting unit. Means, influence vector set extraction means, clustering means, model generation means, size adjustment means, position adjustment means) 30 and RAM (Random Access Memory) 31. The CPU 30 has a role of substantially executing handwritten character generation processing in the control unit 20. The RAM 31 is used as a work area used for processing by the CPU 30.

  Next, generation of handwritten characters by the CPU 30 of the control unit 20 will be described. The CPU 30 executes handwritten character generation processing according to the program recorded in the data recording unit 21.

  FIG. 2 is a flowchart showing a processing procedure of handwritten character generation processing executed by the CPU 30, and FIG. 3 is an explanatory diagram schematically showing processing contents of handwritten character generation processing.

  First, the CPU 30 acquires sample information of characters to be generated (step S.1 in FIG. 2 and step S.21 in FIG. 3). Specifically, first, a sample character or character string (for example, a sentence or the like) is repeatedly written to the writer using the input operation unit 12 (the electronic pen 12a and the screen unit 12b) only a predetermined number of times, for example, five times. do that for me. The CPU 30 records the character information acquired via the input operation unit 12 in the data recording unit 21 as sample information. This sample information includes information on the trajectory of the writing tip from the beginning of writing to the end of writing of each stroke in the character written by the user, temporal information from the writing start coordinate to the writing end coordinate, and the position of the pen tip ( It is composed of (coordinate) progress information. Further, the sample information includes information on the positional relationship between a plurality of strokes in the same character, information indicating the order in which each stroke is written, and the like.

  In addition, in the handwritten character generation apparatus 1 according to the present embodiment, an input operation unit 12 including an electronic pen 12a and a screen unit 12b is provided. By using the input operation unit 12, characters written by a writer A configuration is adopted in which information is acquired as sample information. However, the character sample information is not necessarily limited to the configuration that is read by the input operation unit 12. For example, the sample information acquired in advance using another input operation unit or the like without the input operation unit 12 is provided. The structure recorded on the data recording part 21 may be sufficient. When the sample information is recorded in the data recording unit 21 in advance, the CPU 30 performs step S.1. As the first process, a process of reading sample information recorded in the data recording unit 21 is performed.

  Next, the CPU 30 individually extracts each stroke of each character based on the acquired sample information (step S.2, step S.22, stroke decomposition step), and characterizes the shape of one stroke. A point (hereinafter referred to as “feature point”) is obtained (feature point extraction step), and a process of sharing the number of feature points in a plurality of strokes indicating the same stroke is performed (steps S.3 and S). .23, feature point adding step). The process of sharing the number of characteristic points in each stroke will be described later in detail.

  Next, CPU30 calculates | requires the relationship of each feature point in the same stroke (step S.4, step S.23). Specifically, for each feature point, the CPU 30 obtains another feature point within the same stroke that can affect the one feature point, and the relative direction and distance from the other feature point to the one feature point. The influence vector is obtained based on Then, the CPU 30 obtains a set of influence vectors from a plurality of other feature points in one feature point for each feature point (effect vector set extraction step), and performs clustering processing for each set of effect vectors (step S.5, step S.23, clustering step). By this clustering process, the relative directional relationship of each feature point in each stroke and the distance relationship between adjacent feature points can be obtained by classifying.

  Thereafter, the CPU 30 branches the position of the feature point characterizing the shape of the stroke for each feature point according to the class of the influence vector set based on the class of the clustered effect vector set and the sample information ( A process of generating a statistical model (divided into position determination patterns) is performed (step S.6, step S.23, model generation step). Details thereof will be described later.

  Next, the CPU 30 obtains a mutual relationship with a plurality of strokes constituting the character based on the relationship with the feature points in each stroke (step S.7, step S.24). Specifically, for each feature point, the CPU 30 obtains another influence point (feature point) in another stroke that can affect the one feature point, and relative to the one feature point from the other influence point. A stroke-to-stroke vector is obtained based on the correct positional relationship and its distance. Then, the CPU 30 obtains a set of inter-stroke vectors from another feature point in another stroke to one feature point for each feature point (step S.8, step S.24, inter-stroke vector set extraction step). .

  Step S. described so far. 1-step S.E. 8 (steps S.21 to S.24), the CPU 30 generates a handwritten character having “personality” and “transition” based on the acquired sample information. You will get the pattern. This step S.I. 1-step S.E. By using the model pattern generated based on the step 8, the following step S. 9 and step S. Through 10 processes, a handwritten character having “individuality” and “transition” is generated.

  First, the CPU 30 executes step S.1. 6 and the statistical model obtained in step 6; Each stroke is generated based on the influence vector obtained in step 5 (step S.9, step S.25, stroke generation step). Although a detailed description of the generation of each stroke will be described later, in principle, one feature point is influenced by other feature points in the same stroke before the feature point is drawn (such a feature point Its location) is determined. Furthermore, based on the idea that feature points that can be affected are greatly affected by a visible influence point that is not obstructed by a stroke trajectory from one feature point (this influence point is called a visible influence point). A stroke is generated.

  Next, the CPU 30 executes step S.1. Based on the inter-stroke vector of each feature point acquired in step 8, the size and position of each stroke are adjusted (steps S.10 and S.25). Specifically, the CPU 30 adjusts the relative size (a size adjustment step) between a plurality of strokes formed by one character and the relative position adjustment (position) between the strokes based on the vector between strokes. Adjust step).

  Thus, step S.E. 9 for generating a stroke, and step S. By repeatedly adjusting the position and size of each stroke in step 10, handwritten characters with “individuality” and “transition” are more kinds than the number of characters written by the writer as sample characters. It can be generated (step S.26).

  Next, details of each processing content executed in the above-described program will be described.

[Process for sharing the number of characteristic points constituting one stroke in a plurality of strokes indicating the same stroke (step S.3 in FIG. 2)]
Step S. of FIG. The “process for obtaining a plurality of feature points constituting one stroke” shown in FIG.

  As described above, step S.E. 2, the CPU 30 extracts each stroke of each character individually based on the sample information. Here, the sample information is information for each character related to the trajectory and the writing order of the same character written multiple times by the writer. For example, if the writer has input the character “I” by handwriting only five times, as sample information, “I” composed of two strokes of “NO” and “|” The sample information is extracted by 5 characters.

  The CPU 30 executes step S.1. 2, for example, the stroke of the “no” portion is extracted by 5 characters, In step 3, a characteristic point constituting the stroke of the “no” portion is extracted. Specifically, the CPU 30 may indicate the shape of the stroke by connecting the points that are characteristic points that form the shape of the stroke, that is, by connecting the adjacent feature points with a straight line. Extract feature points in possible strokes. For example, since “|” is a straight line, if one feature point is set for each of the straight lines, the shape of “|” can be represented by a connection between the points. However, “No” has a curved shape, so if you do not set multiple feature points corresponding to the curved state between them, not just the feature points at both ends, the curved state is connected to the points. Can no longer be shown. Therefore, the CPU 30 extracts more feature points than “|” by the CPU 30.

  If the writer acquires the sample information by writing the same character multiple times, the number of feature points may differ depending on the shape of the stroke, even if the stroke is the same character. is there. In this case, it is necessary to match the number of feature points.

  As a technique for comparing the number of feature points by comparing two strokes, a technique called Dynamic time warping with segment division is known. This method is an extension of the existing dynamic time warping (L. Rabiner, B.-H. Juang, "Fundamentals of speech recognition", Prentice Hall, USA, p.209-213, 1993). For example, let us consider a case where there is a stroke having at least four feature points indicated by a solid line and a stroke having at least two feature points indicated by a broken line as shown in FIG. 4A, when the characteristic number of the stroke of the solid line and the stroke of the broken line are matched, first, the corresponding feature point is obtained.In the stroke shown in FIG. Therefore, the CPU 30 determines that D1 and Q1 correspond to at least a corresponding point, and D4 and Q2 correspond to the stroke D1 and the stroke Q4 of the broken line. And the corresponding point To do.

  Next, since there is no corresponding point, the CPU 30 performs a process of adding a new feature point for a stroke with a small number of feature points. As shown in FIG. 4A, points corresponding to D2 and D3 in the solid stroke do not exist in the broken stroke. Therefore, the CPU 30 performs a process of adding points corresponding to D2 and D3 between Q1 and Q2 in the stroke of the broken line.

  FIG. 4B is an enlarged view showing a partial stroke from Q1 to Q2 in the stroke of the broken line and a partial stroke from D1 to D4 in the stroke of the solid line. As shown in FIG. 4 (b), the CPU 30 determines the ratio of the length from D1 to D2, the length from D2 to D3, and the length from D3 to D4 with respect to the overall length from D1 to D4. Ask for. Then, Qa point corresponding to D2 and Qb point corresponding to D3 are added between Q1 and Q2 in the stroke of the broken line so as to correspond to this ratio. In this way, by adding feature points, the number of feature points can be adjusted between different strokes.

  However, the method of Dynamic time warping with segment division shown in FIG. 4 is effective only for two strokes. On the other hand, in the handwritten character generation device 1 according to the present embodiment, it is necessary to apply it to feature points of three or more strokes. For this reason, the handwritten character generation device 1 according to the present embodiment employs a technique called “Matching through template among many strokes”. This method newly defines a stroke for which a feature point is obtained based on two strokes as a template, and each time a new stroke is added, a new feature point is calculated based on the newly added stroke and the template. In search, a new template with new feature points added is defined. In addition, a template is newly defined, and a process of adding a new feature point corresponding to an already added stroke is repeatedly performed. By repeating such processing, it is possible to adjust the feature points so that the number of feature points is the same for all strokes.

  FIG. 5A shows a stroke A and a stroke B in which the feature points have been adjusted. In stroke A and stroke B, white circles and black circles indicate feature points, black circles indicate feature points in the stroke before adjustment, and white circles are newly added based on adjustment processing (Dynamic time warping with segment division). The characteristic points are shown. Further, the template indicates a newly defined template.

  As shown in FIG. 5A, the template defined first is defined in consideration of the feature points generated based on the stroke A feature points and the stroke B feature points. Next, when a new stroke C is added, as shown in FIG. 5B, a new feature is created based on the template defined by FIG. 5A and the newly added stroke C. Points are generated and a new template is generated. Further, when a new template is generated, the stroke A and the stroke B whose feature points have already been adjusted in the previous stage are newly generated by comparing the new stroke C with the template. A point is added.

  As described above, a template is generated based on the result of generating a new feature point based on the two templates, and the newly added template is defined as a feature point of the generated template. A new feature point is added by comparison, and a new template is generated in consideration of the added feature point. Also, a feature point corresponding to a newly generated feature point is added to a template that has already been compared for template generation. By repeating this process, the number of feature points formed in all strokes can be made the same.

[Process for obtaining relationship between feature points in the same stroke (Step S.4: Influencing Model in FIG. 2)]
Next, the “process for obtaining the relationship between the feature points in the same stroke (step S.4 in FIG. 2)” will be described. In the handwritten character generation apparatus 1 according to the present embodiment, “the position of an arbitrary point in the same stroke is determined by being influenced by the positions of many other points in the same stroke”. Based on the assumption, the relationship between each feature point in the same stroke is obtained. Step S. described above. Among the feature points whose number is shared by 3, other feature points that affect the position of a specific feature point are defined as influencing points at that feature point. Then, out of many feature points defined as influence points, Visible influence points are defined based on the influence points written before (in time) the specific feature points in the same stroke. Ask. Here, a visible is a stroke trajectory (line) that is written until a line connecting a specific feature point and each feature point written before that feature point reaches the specific feature point. means not blocked by segment).

  FIGS. 6A and 6B show the feature points in the character “2” composed of one stroke as seven feature points (P0 to P6). FIGS. 6A and 6B show a case where “2” of four characters is acquired as a sample character. Thus, when seven feature points are obtained from the character “2”, the line connecting “P6” and “P0” and the line connecting “P6” and “P1” are “P2”. Will be blocked by the trajectory of the stroke from “P3” to “P3”. Therefore, for “P6”, “P0” and “P1” do not correspond to visible influence points.

  Further, the CPU 30 performs processing for obtaining a RIP (Representative Influencing Point). Here, RIP does not correspond to a visible influence point in all strokes, but since there are some feature points detected as visible influence points, the same determination as the visible influence point is performed. Means a feature point.

  For example, in the fourth stroke (right end of the screen) of the number “2” shown in FIG. 6B, the line connecting “P6” and “P4” reaches from “P2” to “P3”. It is blocked by the stroke trajectory. For this reason, if the fourth “2” in FIG. 6B is determined, “P4” does not correspond to the visible influence point in “P6”.

  However, in the first to third “2” in FIG. 6B, the line connecting “P6” and “P4” is not obstructed by the trajectory of the stroke from “P2” to “P3”. . Therefore, when judging from the first to third “2” in FIG. 6B, “P4” corresponds to a visible influence point in “P6”. In this way, even if the same character is written by the same writer, there is a case where one influence point may or may not correspond to a visible influence point due to a delicate stroke trajectory. To do. With respect to such influence points, based on the number of characters corresponding to visible influence points (in the case shown in FIG. 6B, “P4” corresponds to a visible influence point for three of the four characters). Thus, the ratio of the number of samples that can be judged as a visible influence point is obtained, and when this ratio is equal to or greater than a certain reference value, the RIP is recognized. In the present embodiment, for example, when a visible influence point can be determined at a ratio of 40% or more, the influence point is recognized as RIP.

[Clustering Processing for a Set of Influence Vectors Obtained for Each Feature Point (Step S.5: Vector Clustering in FIG. 2)]
Next, the “clustering process (step S.5 in FIG. 2) for obtaining a set of influence vectors from a plurality of influence points in one feature point for each feature point and performing each influence vector set” will be described. . Visible influence points at each feature point are obtained by the above-described “process for obtaining the relationship between the feature points within the same stroke (step S.4 in FIG. 2)”. This visible influence point indicates a feature point that is greatly affected in the same stroke at a specific feature point. That is, the specific feature point is determined to be greatly influenced by each vector (influence vector indicating distance and direction) from the plurality of visible influence points to the specific feature point. Therefore, each feature point in the stroke is determined based on the influence vector from the visible influence point obtained at each feature point.

  The CPU 30 determines the combination (pattern) of the influence vectors from the visible influence points for each feature point of the stroke, and performs clustering on the combination of the influence vectors. In the handwritten character generation apparatus 1 according to the present embodiment, “Simplified Adaptive Resonance Theory 2 (B.-H. Kim, K.-M. Koo, Y.-M. Park, E.-Y Cha, “A study on quantization method using ART2 for contents-based image retrieval (in Korean)”, Proc. 22th Conf. Korea Information Processing Society, Vol. 11, No. 2, 2004 ”is used. 7 (a) to (c) show the clustering results (class 0 to class 2) of the influence vectors at the point “P6” of the stroke shown in FIG.

  More specifically, with reference to the “P6” point shown in FIG. 6, the “P2” point, the “P3” point, the “P4” point, and the “P5” point to the “P6” point that are visible to the “P6” point. A point influence vector is obtained, and clustering is performed based on the feature of each influence vector. The clustering shown in FIG. 7 shows the clustering result when grouping is performed based on the result of calculating the distance between the four influence vectors. From this clustering result, for example, it is shown that the position determination pattern of the “P6” point may be the number of classes clustered with reference to the “P5” point of the same stroke (three as shown in FIG. 7). ing. That is, the position of the “P6” point obtained according to each clustering result has an influence corresponding to the class depending on whether any class (class 0 to class 2) is selected with reference to the “P5” point. It is determined by the direction and distance of the vector.

[Process of constructing a statistical model indicating the position determination pattern of feature points (step S.6 in FIG. 2: Statistical Model)]
Next, the “process for forming a statistical model indicating the position determination pattern of feature points” will be described. A method for determining the position of the next feature point from the previous feature point is obtained by the above-described clustering process of the influence vectors (step S.5 in FIG. 2). For example, the position determination pattern of the next feature point “P6” point based on the “P5” point shown in FIG. 6 is determined based on the set of influence vectors in any of the three patterns shown in FIG. The Accordingly, the “P6” point is determined at three different positions (three classes) with respect to the “P5” point.

  In this way, by determining a set of influence vectors at all feature points, a position determination pattern for the next feature point is obtained from the previous feature point. Furthermore, the ratio of which pattern the next feature point is determined according to which pattern can be calculated based on the sample information. That is, by determining which class in the clustered influence vector set the sample information for n times constituting the same stroke corresponds to the total number n, the ratio based on the corresponding number of each class can be obtained. .

  In other words, the position of the feature point indicating the feature of the stroke is determined based on the class of the set of influence vectors with reference to the immediately preceding feature point in the same stroke, and the determination ratio of the class is determined for each class in the sample information. It will be decided according to the corresponding ratio.

  FIG. 8A is a diagram showing a model (statistical model) of the position determination pattern of a series of feature points obtained in this way. In FIG. 8A, order indicates the order of feature points constituting the stroke, and order 0 indicates the first feature point.

When the value of order is increased by 1, that is, when the next feature point is determined, a set of clustered influence vectors based on the previous feature point, that is, the order before the value is increased by 1. There are as many position determination patterns as feature points branch. For example, feature points order1 has two types of positioning pattern for the feature point C ₀₀ of Order0, one feature point C _10, the other characteristic point is in the C _11. Similarly, for the feature point of order2, there are two types of position determination patterns for the feature point of order1. Therefore, feature points order2 _{is, C 20, C 21, C} 22, C 23, and has a four. Similarly order3 feature point _has a four _{C 30, C 31, C 32} , C 33.

Further, the ratio of which of the branched position determination patterns is selected is determined according to the ratio of the sample pattern corresponding to which position determination pattern among n pieces of sample information. For example, when there is information about 10 strokes as sample information, the determination ratio of the position determination pattern is determined depending on which of the 10 strokes corresponds to the clustered class. . In FIG. 8 (a), the it is shown that the percentage of branching to the feature point _{to C 10} feature points _{C 00} 0.6, the ratio of branched to the characteristic point _{C 11} from the feature point _{C 00} has become 0.4 Has been. That is, when the number of samples at the feature point C ₀₀ is 10, 6 samples indicate the position determination pattern belonging to the feature point C ₁₀ , and 4 samples indicate the position determination pattern belonging to the feature point C ₁₁ . It will be.

  In this way, a statistical model as shown in FIG. 8A is obtained by obtaining the correspondence relationship with the previous feature point in accordance with the number of classes and the ratio of all feature points. be able to. By completing the stroke according to this model, it is possible to show a stroke having various characteristics.

  Since each feature point obtained according to the model is determined based on a set of influence vectors clustered based on the writer's features, each stroke formed based on the feature points is a shape that represents the writer's features. Will be shown. For this reason, the strokes obtained using the model are strokes having “individuality” that can be recognized as characters written by the same writer, and the “transition” that causes diversity in the shape of each character. It has the feature of showing a stroke with “character”.

  In addition, the stroke required using the model can be shown in more types of shapes (combination of feature points) than the number of sample information, which is more than the number of characters actually written by the writer as a sample. A stroke model can be generated.

  When building a model, as described above, it is necessary to make the number of feature points included in the same stroke (constituting a stroke) the same. For this reason, when the number of feature points in the stroke is not the same, it is necessary to preferentially perform the process of making the number of feature points the same by performing the process described with reference to FIGS. Arise.

  Further, the model is not limited to a model having a shape similar to a tree structure as shown in FIG. FIG. 8B shows a position determination pattern of feature points using a table instead of the model shown in FIG. As described above, the specific model configuration has a shape similar to a tree structure as long as the determination ratio of the position determination pattern is obtained depending on which of the clustered classes it corresponds to. May be specified by a table, or may be established by a different configuration.

[Process for obtaining a mutual relationship in a plurality of strokes based on a relationship with a feature point in each stroke (step S.7: Interstroke Model in FIG. 2)
Next, “a process for obtaining a mutual relationship among a plurality of strokes constituting a character based on a relationship with a feature point in each stroke” will be described. Through the model configuration process described above, the stroke shape having “individuality” and “transition” can be obtained. In this process, a process for generating a character having “personality” and “transition” is performed by combining strokes having “personality” and “transition”.

  In order to have “individuality” and “transition” in the generated characters, not only each stroke has “individuality” and “transition”, but also the relationship between the strokes forming the character However, it must have the characteristics required based on the sample information. For this reason, a mutual relationship between one stroke and another stroke constituting the character is obtained based on the direction and distance from the feature point in the other stroke to the feature point in the one stroke. Therefore, in this process, based on each sample information, the process of obtaining the position of the feature point of one stroke by a vector from the feature point of another stroke (hereinafter referred to as an inter-stroke vector) is performed.

  It is assumed that the feature point in one stroke is affected by all feature points of other different strokes between strokes constituting different parts of the same character. When a writer actually writes a character, determine the position and size of the stroke to be drawn next, taking care to correspond to the position and size of the stroke already drawn (for example, a character dot or line).・ Tends to adjust. Therefore, also in this process, an inter-stroke vector is obtained from the feature points of other strokes already drawn. It is possible to determine and adjust the position and size of the target stroke by determining the direction and distance from the stroke feature point that has already been drawn based on the inter-stroke vector thus determined. It becomes.

  Also, when determining an inter-stroke vector, it is determined in consideration of visible influence points. In other words, the stroke-to-stroke vector is based on the point of influence where a line connecting a specific feature point in one stroke and each feature point drawn before that feature point is not obstructed by any stroke. Ask for.

  9A to 9C show a case where the writer has written the same character composed of four strokes three times. Each character is composed of four strokes: a stroke W1 consisting of a left vertical line, a stroke W2 consisting of an upper horizontal line, a stroke W3 consisting of a right vertical line, and W4 consisting of a central horizontal line. The stroke order is W1, W2, W3, W4. The strokes W1 to W3 are composed of two feature points located at the respective end portions, and the stroke W4 is composed of three feature points arranged in the horizontal direction.

  In FIGS. 9A to 9C, the inter-stroke vector at the leftmost feature point of the stroke W4 is indicated by a vector notation (arrow symbol). More specifically, FIGS. 9A to 9C show strokes W1 to W3 written before the stroke W4 among the stroke characteristic points different from the stroke W4, and each stroke W1. A line between each feature point of W3 to the leftmost feature point of the stroke W4 is drawn as an inter-stroke vector of feature points (visible feature points) that are not obstructed by the presence of the strokes W1 to W3 It is. Based on the inter-stroke vector thus determined, it is possible to adjust the size and position between strokes.

[Stroke Generation Processing Based on Model and Influence Vector (Step S.9 in FIG. 2)]
Next, the CPU 30 executes step S.1. 6 and the statistical model obtained in step 6; A process for generating each stroke based on the influence vector obtained in step 5 will be described.

  As described above, the position of a feature point within the same stroke can be obtained from an influence vector from another feature point. In addition, the set of influence vectors used when determining each feature point is one of the classes of clustered effect vectors, and which set of effect vectors is adopted depends on the feature points of the model ( Determined for each order). Therefore, by determining the position of each feature point along the route from the first order 0 to the last order according to the branch of the model, multiple strokes with “personality” and “transition” are generated. It becomes possible to do.

  FIGS. 10A to 10C show a process of determining the position of the last feature point “P6” of the number “2” character. FIG. 10A shows a state where the position up to the feature point “P5” is determined in the numeral “2”, and FIG. 10B shows an influence vector used to determine the “P6” point. A set of FIG. 10C shows the influence vectors at the feature points “P2” to “P5”, which are the visible influence points shown in FIG. 10A, using the influence vectors shown in FIG. FIG. As shown in FIG. 10C, when the position of “P6” is specified using a clustered set of influence vectors, the tip of each influence vector does not necessarily coincide with one point. As shown in FIG. 10C, when the tip position of each influence vector does not coincide with one point, the position obtained by the weighted average in each influence vector is determined as the feature point “P6”. In this way, strokes having various shapes can be generated by determining each feature point in the stroke using a set of any of the influence vectors.

[Process of adjusting the size and position of each stroke based on the stroke vector (step S.10 in FIG. 2)]
Next, a method for adjusting the size and position of each stroke based on the inter-stroke vector of each feature point will be described. First, when adjusting the size of the stroke, for all feature points of one stroke, the influence vector from the influence point in the same stroke and the inter-stroke vector from the influence point in the other stroke are set. Use to calculate the virtual position in one stroke. In the calculation of the virtual position in the stroke, it is possible to calculate the virtual position in consideration of the relative positional relationship between the strokes by using the inter-stroke vector in the other strokes. Furthermore, by calculating the virtual position using the influence vector from the influence point in the same stroke, it is possible to obtain the virtual position in consideration of the relative positional relationship of each feature point in the same stroke.

  Then, the CPU 30 calculates the size (height dimension and width dimension) of the virtual stroke in one stroke from the calculated virtual position, and calculates the size of the existing stroke. Adjust according to the height.

  FIGS. 11A to 11D are diagrams showing a process of adjusting the stroke size. The strokes W1 to W4 in FIG. 11A are the same as the strokes shown in FIG. When adjusting the size of the stroke W4 in FIG. 11 (a), the CPU 30 selects the influence vector at the influence point in the stroke W4 and the inter-stroke vector at the influence points of the other strokes W1 to W3. The virtual position of the stroke W4 indicated by the white circle in (b) is calculated. Then, the CPU 30 calculates a virtual stroke W4a based on the calculated virtual position as shown in FIG. 11C, and the existing stroke W4 as shown in FIG. 11D. The size is adjusted according to the calculated size of the virtual stroke W4a.

  Next, the CPU 30 performs position adjustment processing of the stroke whose size has been adjusted. In the position adjustment process, the virtual position of the feature point in one stroke is calculated using only the inter-stroke vector at the influence point of another stroke. The position adjustment processing is different from the size adjustment processing in that only the inter-stroke vector for other strokes is used and the influence vector in the same stroke is not used. The size adjustment in one stroke has already been performed, and the size and shape of the stroke are not changed by this position adjustment processing, and the position adjustment in one stroke has a mutual position with other strokes. This is because the relationship is required to generate characters with “personality” and “transition”.

  Then, the position of the stroke is adjusted so that the error between the calculated virtual position of the feature point in the stroke and the actual position of the feature point in the stroke whose size has been adjusted is minimized.

  12A to 12C are diagrams showing a process of adjusting the position of the stroke. FIG. 12A shows the strokes W1 to W4 after the size of the stroke W4 is adjusted by adjusting the size of the stroke. As shown in FIG. 12B, the CPU 30 uses only the inter-stroke vectors at the influence points of the other strokes W1 to W4, and uses the virtual positions of the feature points in the stroke W4 (white dots in FIG. 12B). Is calculated. Then, as shown in FIG. 12C, the CPU 30 calculates the virtual position (the white circle point in FIG. 12B) of the calculated feature point of the stroke W4 and the actual feature point in the stroke whose size has been adjusted. The position of the stroke W4 is adjusted so that the error with respect to the position is minimized.

  In this way, strokes of various patterns are generated according to the branch of the model, and by adjusting the size and position of the generated strokes as described above, the writer who wrote the character as a sample “ Characters having “individuality” and “transition” can be generated in various types more than the number of characters written as samples.

  FIG. 13A shows a handwritten character string written as a sample by a writer only five times in a vertically written character string made up of “2 gmx Shijo Yuki Mizushiishi”. On the other hand, in FIG. 13B, the character string of the sample in FIG. 13A is read by the input operation unit 12, and the “personality” and “transition” of the sample character string written by the writer based on the processing described above. An example is shown in which only 10 character strings having a character are generated. Comparing FIG. 13 (a) and FIG. 13 (b), each character is not the same character, but shows a common shape. In this way, by generating a character using the handwritten character generation device 1, it is possible to generate a character reflecting “individuality” indicating characteristics that can be recognized as a character string written by the writer. Thus, it is possible to generate a character having “transition”, which is a character having a characteristic / difference as a different character while each character has a common feature. In addition, the generated character string has five or more different “transitions” characters, even though the character string written as a sample is five times, that is, there are only five characters per character sample. A column can be generated.

  As described above, by using the handwritten character generation device 1 according to the present embodiment, the position of the feature point in the stroke constituting the character is obtained from the set of influence vectors in each feature point in the same stroke. Therefore, it is possible to generate a stroke having “individuality” and “transition” of a writer who wrote a character used as a sample.

  Furthermore, by clustering and classifying a set of influence vectors at each feature point, it is possible to select multiple (points of model branches) for determining the position of the feature point in the stroke. More types of strokes than the number of times can be generated.

  In addition, the size and position of each stroke can be determined using a set of influence vectors and a set of vectors between strokes, so that the position and size between strokes can be adjusted according to the character characteristics of the sample. As a result, it is possible to generate more characters with “personality” and “transition” of the writer than the number of times the sample has been written.

  Although the character generation device according to the present invention has been described in detail with reference to the drawings, the character generation device according to the present invention is not limited to the above-described embodiments. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

  For example, average character data of all characters to be generated is recorded in advance in the first database as a standard pattern, and a predetermined character having characteristics of characters to be generated is described by a certain writer. This makes it possible to generate a character having specific characteristics different from those written by the writer.

  Specifically, in the standard pattern recorded in the first database, all characters are separated in units of strokes, and a large number of groups are generated by clustering the separated strokes. Then, the relationship between the strokes having a large number of clustered groups and the characters to be generated is obtained and recorded in the third database. On the other hand, sample information obtained by sample characters written by the writer is recorded in the second database.

  In this way, a standard pattern is recorded in the first database, sample information of characters written by the writer is recorded in the second database, and a large number of stroke groups clustered in the third database are generated. By recording information indicating the relationship with the character to be tried, a set of strokes relating to the characters that can be generated recorded in the second database can be obtained from the third database, and a large number of obtained strokes can be obtained. A new character can be generated by using the information. Existing research methods (for example, Jungpil Shin, Kazunori Suzuki, Atsushi Hasegawa, "Handwritten Chinese Character Font Generation Based on Stroke Correspondence", International Journal of Computer Processing of Oriental Languages, Vol.18, Issue 3, p.211-226, 2005), it is possible to generate about 3000 types of Chinese characters different from the input Chinese characters only by having the writer input about 100 types of Chinese characters.

  Also, as sample information, rather than using only the characters written by the same writer, by collecting sample information by mixing different characters written by different writers, the character characteristics of each writer can be obtained. It becomes possible to generate characters with new “individuality” and “transition” mixed. In particular, by recording the standard pattern described above in a database in advance, it is possible to generate a character different from the character written by the writer. For this reason, by constructing a sufficient standard pattern, even if the number of characters input by the user is small, it is possible to generate many types of characters similar to the characters of the actual writer.

  Kanji has many types of characters, and each character is composed of a set of various common strokes called “radicals”. For this reason, the relationship between a plurality of “radicals” is expressed as “a stroke corresponding to a set of inter-stroke vectors based on feature points of other strokes” in the handwritten character generation apparatus 1 according to the present embodiment. By modeling based on the “processing method obtained for each feature point”, it is possible to generate other characters using a common radical.

  Furthermore, by applying the character generation method according to the present invention, it is possible to generate a cursive character based on sample characters written in a typeface. Note that the cursive type has a characteristic that it has a stroke-like shape in which strokes of characters are connected differently from other characters. For this reason, in order to generate a cursive font, general connection information between strokes forming the cursive font is required, and such connection information is recorded in the fourth database.

  When generating a cursive typeface from a typeface, a lot of typeface strokes are generated based on the sample information of the typeface. Then, the generated strokes of the typeface are connected based on the connection information recorded in the fourth database to generate a character to which a part of the stroke is connected. After that, the generated character information is converted into the frequency domain using Fourier transform, and after the high frequency information in the converted frequency domain is removed, the data is converted into character information (planar information) using inverse Fourier transform. Convert. After removing the high-frequency components after performing the Fourier transform and then converting it back to text information using the inverse Fourier transform, a cursive font with a smooth and aesthetically deformed (aesthetically collapsed) shape is generated. It becomes possible to do.

1 ... handwritten character generation device (stroke generation device, character generation device)
DESCRIPTION OF SYMBOLS 11 ... Display part 12 ... Input operation part 12a ... Electronic pen 12b ... Screen part 13 ... Main-body part 20 ... Control part 21 ... Data recording part 30 ... CPU (stroke decomposition | disassembly means, stroke generation means, inter-stroke vector set extraction means, stroke Inter-effect point extraction means, influence point extraction means, feature point addition means, feature point extraction means, influence vector set extraction means, clustering means, model generation means, size adjustment means, position adjustment means)
31 ... RAM

Claims (21)

Stroke disassembling means for disassembling the structure of the character by stroke based on information on the trajectory and stroke order of the same character;
Feature point extracting means for extracting a feature point characterizing the shape of the stroke for each stroke decomposed by the stroke decomposing means;
Influence of obtaining a set of influence vectors indicating distances and directions from all other feature points in the stroke to the one feature point with respect to one feature point of the stroke for all feature points in the stroke Vector set extraction means;
A set of the influence vectors obtained by the influence vector set extraction means, wherein the set of the influence vectors at all the feature points of all the strokes of the same character written a plurality of times is written a plurality of times. Clustering means for classifying the set of influence vectors at the feature points by clustering for each feature point of the stroke that constitutes the same part of the same character;
A position determination pattern for determining the position of the next feature point from the one feature point is set according to the number of classes of the set of the influence vectors in the one feature point classified by the clustering means. Model generation means for generating a feature point position determination model based on the position determination pattern of all feature points;
By using the set of influence vectors classified by the clustering means as a position determination pattern for determining the position of the next feature point according to the feature point position determination model generated by the model generation means, A stroke generation device comprising: stroke generation means for determining a position of a point and generating a stroke. An influence point extracting means for obtaining, as an influence point, another feature point located on a trajectory before the one feature point with respect to one feature point of the stroke;
The said influence vector set extraction means calculates | requires the influence vector which shows the distance and direction from the said influence point to the said one feature point in all the influence points in the said one feature point. The stroke generation device described. With reference to one feature point of the stroke, which is another feature point located in the locus before the one feature point, and which is a line connecting the one feature point and the other feature point An influence point extracting means for obtaining other feature points that are not obstructed by the trajectory of the stroke as an influence point;
The said influence vector set extraction means calculates | requires the influence vector which shows the distance and direction from the said influence point to the said one feature point in all the influence points in the said one feature point. The stroke generation device described. When the number of feature points is different for each of multiple strokes that make up the same part of the same character written multiple times, feature point addition processing is performed to make the number of feature points the same for each stroke. The stroke generation device according to any one of claims 1 to 3, further comprising: means. The stroke generation device according to any one of claims 1 to 4,
Between the strokes constituting different parts of the same character decomposed by the stroke disassembling means, the feature point of one stroke is used as a reference, and all the feature points of other strokes different from the one stroke are used to calculate the one stroke. An inter-stroke vector set extracting means for obtaining a set of vectors between strokes indicating a distance and a direction to a feature point of the stroke for each feature point in all strokes of the same character written multiple times;
The size of one stroke generated by the stroke generating means is the same as that of the set of inter-stroke vectors of other strokes that constitute the same character and different parts, and the same character that constitutes the same part. A size adjusting means for adjusting based on a set of influence vectors in the stroke of
Position adjusting means for adjusting the position of the stroke whose size has been adjusted by the size adjusting means, based on a set of vectors between strokes in other strokes which are the same character and constitute different parts;
A character generation device comprising: Between strokes constituting different parts of the same character that have been decomposed by the stroke disassembling means, with reference to the characteristic point of one stroke, other strokes positioned in the order of writing before the one stroke. An inter-stroke influence point extraction means for obtaining a feature point as an inter-stroke influence point is provided.
The inter-stroke vector set extraction means includes a plurality of sets of inter-stroke vectors indicating distances and directions from all the inter-stroke influence points obtained by the inter-stroke influence point extraction means to the feature points of the one stroke. The character generation device according to claim 5, wherein the character generation device is obtained for every feature point in all strokes of the same character written once. Between strokes constituting different parts of the same character that have been decomposed by the stroke disassembling means, with reference to the feature point of one stroke, other strokes that are positioned in the order of writing before the one stroke. And any one of the feature points in which the line connecting any one of the feature points of the other stroke and the feature point of the one stroke is not obstructed by the trajectory of any stroke is defined as an inter-stroke influence point. Equipped with a means for extracting the influence points between strokes,
The inter-stroke vector set extraction means includes a plurality of sets of inter-stroke vectors indicating distances and directions from all the inter-stroke influence points obtained by the inter-stroke influence point extraction means to the feature points of the one stroke. The character generation device according to claim 5, wherein the character generation device is obtained for every feature point in all strokes of the same character written once. A stroke disassembling step in which the stroke disassembling means disassembles the structure of the character by stroke based on the information on the trajectory and the stroke order of the same character;
For each stroke decomposed in the stroke decomposition step, the feature point extraction means extracts a feature point that characterizes the shape of the stroke; and
Based on one feature point of the stroke, an influence vector set extraction unit is configured to obtain a set of influence vectors indicating distances and directions from all other feature points in the stroke to the one feature point. An influence vector set extraction step for all feature points;
A set of the influence vectors obtained in the influence vector set extraction step, wherein the set of the influence vectors at all the feature points of all the strokes of the same character written a plurality of times is written a plurality of times. A clustering step in which the clustering means classifies the set of the influence vectors at the feature points by clustering for each feature point of the stroke constituting the same part of the same character,
The model generation means sets the position determination pattern for determining the position of the next feature point from the one feature point to the number of classes of the set of the influence vectors in the one feature point classified in the clustering step. In response to this, a model generation step for setting several patterns and generating a feature point position determination model based on the position determination patterns of all feature points;
Stroke generation by using the set of influence vectors classified in the clustering step as a position determination pattern for determining the position of the next feature point according to the feature point position determination model generated in the model generation step A means for generating a stroke, characterized in that the means includes a stroke generation step of determining a position of each feature point and generating a stroke. The influence point extracting means includes, as a reference, an influence point extraction step for obtaining, as an influence point, another feature point located on a trajectory before the one feature point based on one feature point of the stroke.
In the influence vector set extraction step, the influence vector set extraction means obtains an influence vector indicating a distance and a direction from the influence point to the one feature point at all the influence points in the one feature point. The stroke generation method according to claim 8. On the basis of one feature point of the stroke, the influence point extracting means is another feature point located in a locus before the one feature point, and the one feature point and the other feature point An influence point extraction step for obtaining, as an influence point, another feature point that is not obstructed by the trajectory of any stroke.
In the influence vector set extraction step, the influence vector set extraction means obtains an influence vector indicating a distance and a direction from the influence point to the one feature point at all the influence points in the one feature point. The stroke generation method according to claim 8. When the number of feature points is different for each of a plurality of strokes constituting the same part of the same character written a plurality of times, the feature point adding means performs a feature point addition process for each stroke to calculate the number of feature points. The stroke generating method according to any one of claims 8 to 10, further comprising the same feature point adding step. A stroke generation method according to any one of claims 8 to 11,
Between strokes constituting different parts of the same character decomposed in the stroke decomposing step, the feature point of one stroke is used as a reference, and all the feature points of other strokes different from the one stroke are used to determine the one stroke. The inter-stroke vector set extraction means that calculates the set of inter-stroke vectors indicating the distance and direction to the feature point of the stroke for each feature point of all the strokes of the same character written multiple times. A vector set extraction step;
The size of the one stroke generated in the stroke generation step is the same character as the set of inter-stroke vectors of other strokes that constitute the different parts of the same character and the size adjusting means. A magnitude adjustment step for adjusting based on a set of influence vectors in the same stroke constituting the same part;
Position adjustment in which the position adjustment means adjusts the position of the stroke whose size has been adjusted by the size adjustment step based on a set of vectors between strokes in other strokes that are the same character and constitute different parts. A character generation method comprising the steps of: Between strokes constituting different parts of the same character decomposed in the stroke decomposition step, the inter-stroke influence point extraction means is more in a stroke order than the one stroke based on the feature points of one stroke. An inter-stroke influence point extraction step for obtaining a feature point of another stroke positioned in front as an inter-stroke influence point is provided,
In the inter-stroke vector set extracting step, the inter-stroke vector set extracting means calculates the distances and directions from all the inter-stroke effect points obtained in the inter-stroke effect point extracting step to the feature points of the one stroke. The character generation method according to claim 12, wherein a set of inter-stroke vectors to be shown is obtained for every feature point of every stroke written for a plurality of times for the same character. Between strokes constituting different parts of the same character decomposed in the stroke decomposition step, the inter-stroke influence point extraction means is more in a stroke order than the one stroke based on the feature points of one stroke. Any one of the above-described features, wherein a line connecting any feature point of the other stroke and the feature point of the one stroke is not obstructed by the trajectory of any stroke. An inter-stroke influence point extraction step for obtaining points as inter-stroke influence points;
In the inter-stroke vector set extracting step, the inter-stroke vector set extracting means calculates the distances and directions from all the inter-stroke effect points obtained in the inter-stroke effect point extracting step to the feature points of the one stroke. The character generation method according to claim 12, wherein a set of inter-stroke vectors to be shown is obtained for every feature point of every stroke written for a plurality of times for the same character. A stroke generation program capable of generating a stroke similar to a stroke constituting the character by using information on a trajectory and a writing order of the same character written a plurality of times,
On the computer,
A stroke disassembly function for disassembling the structure of the character by stroke based on the information on the trajectory and writing order of the same character;
A feature point extraction function for extracting a feature point characterizing the shape of the stroke for each stroke decomposed by the stroke decomposition function;
Based on one feature point of the stroke, a set of influence vectors indicating distances and directions from all other feature points in the stroke to the one feature point are obtained for all feature points in the stroke. An influence vector set extraction function;
A set of the influence vectors obtained by the influence vector set extraction function, wherein the set of the influence vectors at all the feature points of all the strokes of the same character written a plurality of times is written a plurality of times. A clustering function for classifying the set of the influence vectors at the feature points by clustering for each feature point of the stroke constituting the same part of the same character,
A position determination pattern for determining the position of the next feature point from the one feature point is set according to the number of classes of the set of the influence vectors in the one feature point classified by the clustering function. A model generation function for generating a feature point position determination model based on a position determination pattern of all feature points;
By using the set of influence vectors classified by the clustering function as a position determination pattern for determining the position of the next feature point according to the feature point position determination model generated by the model generation function, A stroke generation program for realizing a stroke generation function for generating a stroke for determining a position of a point. In the computer,
Based on one feature point of the stroke, realize an influence point extraction function that obtains another feature point located in a trajectory before the one feature point as an influence point,
16. The influence vector set extraction function allows an influence vector indicating a distance and a direction from the influence point to the one feature point to be obtained at all influence points in the one feature point. Stroke generation program described in 1. In the computer,
With reference to one feature point of the stroke, which is another feature point located in the locus before the one feature point, and which is a line connecting the one feature point and the other feature point Realize the influence point extraction function to obtain other feature points that are not obstructed by the trajectory of the stroke as the influence point,
16. The influence vector set extraction function allows an influence vector indicating a distance and a direction from the influence point to the one feature point to be obtained at all influence points in the one feature point. Stroke generation program described in 1. In the computer,
When the number of feature points is different for each of multiple strokes that make up the same part of the same character written multiple times, additional feature points are added to make the number of feature points the same for each stroke. The stroke generation program according to any one of claims 15 to 17, wherein a function is realized. On the computer,
A function realized by the stroke generation program according to any one of claims 15 to 18,
Between strokes constituting different parts of the same character, which are decomposed by the stroke decomposition function, the feature point of one stroke is used as a reference, and all the feature points of other strokes different from the one stroke are used as the reference point. An inter-stroke vector set extraction function for obtaining a set of vectors between strokes indicating the distance and direction to the feature point of the stroke for each feature point of all strokes of the same character written multiple times;
The size of one stroke generated by the stroke generation function is the same as the set of vectors between strokes of other strokes that are the same character and constitute different parts, and the same character that constitutes the same part A size adjustment function to adjust based on a set of influence vectors in the stroke of
A position adjustment function for adjusting the position of the stroke whose size has been adjusted by the size adjustment function based on a set of vectors between strokes in other strokes that are the same character and form different parts is realized. Character generation program for. In the computer,
Between strokes constituting different parts of the same character, which are decomposed by the stroke decomposition function, with reference to the characteristic point of one stroke, other strokes positioned in the stroke order before the one stroke. Realize the inter-stroke influence point extraction function to obtain the feature points as the inter-stroke influence points,
In the inter-stroke vector set extraction function, a plurality of sets of inter-stroke vectors indicating distances and directions from all the inter-stroke influence points obtained by the inter-stroke influence point extraction function to the feature points of the one stroke The character generation program according to claim 19, wherein the character generation program is obtained for every feature point of every stroke for every same character written once. In the computer,
Between strokes constituting different parts of the same character, which are decomposed by the stroke decomposition function, with reference to the feature point of one stroke, and other strokes positioned in the stroke order before the one stroke. And any one of the feature points in which the line connecting any one of the feature points of the other stroke and the feature point of the one stroke is not obstructed by the trajectory of any stroke is defined as an inter-stroke influence point. Realize the function to extract the influence points between strokes
In the inter-stroke vector set extraction function, a plurality of sets of inter-stroke vectors indicating distances and directions from all the inter-stroke influence points obtained by the inter-stroke influence point extraction function to the feature points of the one stroke The character generation program according to claim 19, wherein the character generation program is obtained for every feature point of every stroke for every same character written once.

Images