CN111738141A

CN111738141A - Hard-tipped writing calligraphy work judging method

Info

Publication number: CN111738141A
Application number: CN202010566088.9A
Authority: CN
Inventors: 骆力明; 卫星辰; 刘杰; 张磊; 张凯
Original assignee: Capital Normal University
Current assignee: Capital Normal University
Priority date: 2020-06-19
Filing date: 2020-06-19
Publication date: 2020-10-02
Anticipated expiration: 2040-06-19
Also published as: CN111738141B

Abstract

The disclosure relates to a method for judging hard-tipped writing calligraphy works, which comprises the following steps: acquiring a hard-stroke Chinese character image based on a judgment standard, and establishing a judgment data set based on the hard-stroke Chinese character image; carrying out feature selection on Chinese characters in the evaluation data set based on a feature point set to obtain a plurality of features, wherein the feature point set is a set formed by all feature points forming each Chinese character; automatically labeling the feature points of the plurality of features based on the active shape model to obtain labeled features; comparing and calculating the standard words in the evaluation data set with the words to be evaluated according to the marked features to obtain a plurality of feature similarities; establishing a Chinese character feature library according to the plurality of features, the marked feature points and a feature matching algorithm; and establishing a Chinese character automatic judging model based on the grading label and the feature similarity corresponding to the Chinese characters in the Chinese character feature library. The method can improve the marking accuracy, improve the specialty and the objectivity of judging work, and realize the automatic judgment of the writing attractiveness of the Chinese characters by using the computer.

Description

Hard-tipped writing calligraphy work judging method

Technical Field

The disclosure relates to the technical field of image evaluation, in particular to a method for judging hard-tipped writing works.

Background

When the calligraphy evaluation problem is solved by means of an information technology at present, the calligraphy evaluation problem can be divided into an online calligraphy evaluation and an offline calligraphy evaluation according to different research objects, wherein the online evaluation refers to the calligraphy evaluation by collecting various information in the writing process through input equipment such as a handwriting flat plate and the like; the off-line evaluation means that the calligraphy evaluation is carried out by taking a picture or scanning the words written by the students according to the copybook, converting the words into a static image file which can be processed by a computer, and carrying out feature selection and feature comparison on the image. Because the off-line Chinese character evaluation does not need to use special input equipment during writing, the constraint condition is less, and compared with input equipment such as a handwriting panel and the like, the copy copybook is more suitable for students to learn calligraphy and stroke in the teaching level, the off-line evaluation has more guiding significance in calligraphy teaching compared with the on-line evaluation. However, the off-line Chinese character judgment has the defect that the on-line Chinese character judgment only can utilize image static information and contains richer dynamic characteristic information, and the difficulty is higher when the characteristics are recorded by intelligent equipment along with the input process.

For a computer, information contained in an image consists of all pixel points, the computer is difficult to distinguish which pixel points of a hard-stroke Chinese character image represent strokes, which pixel points represent characteristics such as turning and starting when the Chinese character is written, and the characteristics have important significance for judging the writing quality of the hard-stroke Chinese character. Common calligraphy image features include statistical features and structural features, and the statistical features refer to features formed by using a certain method in the same type of characters through statistical results, such as image gray scale features, entropy information and the like. Although the statistical characteristics do not fully utilize the structural information of the Chinese characters, the Chinese characters have stronger identification capability than other western characters due to the complex structure of the Chinese characters, so the statistical characteristics of the Chinese characters are often applied to the field of Chinese character identification, but the defects of the statistical characteristics also make the statistical characteristics unsuitable for the Chinese character evaluation and research. The structural characteristics refer to the basic elements of the Chinese character, such as strokes, components, characteristic points and the like, and the structural characteristics conform to the process of recognizing and writing the Chinese character by people and reflect the internal structural properties of the Chinese character. The research of the structural characteristics of the Chinese characters follows the research idea of firstly extracting basic units of the Chinese character structures, such as strokes, parts and the like, and then forming the characteristics of the Chinese characters by the basic structural units, but the extraction of the parts and the strokes of the Chinese characters is difficult, the topological relation among the structures is complex, and good effects are difficult to achieve. Therefore, how to automatically and accurately label the effective characteristic information in the hard-tipped pen Chinese character image and calculate through the characteristic information to judge the writing aesthetic degree of the Chinese characters becomes a difficulty of Chinese character evaluation.

The above drawbacks are expected to be overcome by those skilled in the art.

Disclosure of Invention

Technical problem to be solved

In order to solve the above problems in the prior art, the present disclosure provides a method for evaluating a hard-pen calligraphy work, which is used to solve the problem of how to accurately evaluate the beauty of writing Chinese characters based on effective characteristic information in a hard-pen Chinese character image.

(II) technical scheme

In order to achieve the above purpose, the present disclosure adopts a main technical solution including:

an embodiment of the present disclosure provides a method for evaluating a hard-tipped writing work, including:

s1, acquiring a hard-stroke Chinese character image according to the judgment standard, and establishing a judgment data set based on the hard-stroke Chinese character image;

s2, selecting characteristics of the Chinese characters in the judging data set based on the characteristic point set to obtain a plurality of characteristics, wherein the characteristic point set is a set formed by all characteristic points forming each Chinese character;

s3, automatically labeling the feature points of the multiple features based on the active shape model to obtain labeled features;

s4, comparing and calculating the standard words in the evaluation data set with the words to be evaluated according to the labeled features to obtain a plurality of feature similarities;

s5, establishing a Chinese character feature library according to the plurality of features, the marked feature points and the feature matching algorithm;

s6, establishing an automatic Chinese character judging model based on the corresponding grading label and the feature similarity of the Chinese characters in the Chinese character feature library.

In an embodiment of the present disclosure, step S1 is preceded by:

s0, constructing judgment criteria, wherein the judgment criteria comprise evaluation contents and text description of the evaluation contents;

step S0 includes:

s01, dividing the evaluation content into an integral evaluation level, a component evaluation level and a stroke evaluation level according to the shape structure of the Chinese character;

s02, each evaluation level comprises a plurality of evaluation elements, and the evaluation elements in the overall evaluation level at least comprise stroke length proportion, stroke inclination, stroke type, stroke interval relation and similar stroke parallel relation; the evaluation elements in the component evaluation hierarchy at least comprise component forms, component density distances and component proportions; the evaluation elements in the stroke evaluation hierarchy at least comprise the form, position and size of the whole character;

s03, solving the text description of the evaluation content according to the specific evaluation element corresponding to the evaluation element of each evaluation level;

and S04, converting the evaluation hierarchy, the evaluation elements and the detailed solutions of the evaluation elements from natural language into computer language to obtain the evaluation standard with quantitative evaluation indexes.

In an embodiment of the present disclosure, step S1 includes:

s11, obtaining a whole copybook image obtained after a writer writes according to the standard copybook;

s12, dividing the whole copybook image into single Chinese character images;

s13, preprocessing a single Chinese character image to obtain a processed image, wherein the preprocessing comprises denoising, binarization and image resolution size adjustment;

and S14, scoring and marking the processed image according to the evaluation standard, wherein the marked data form an evaluation data set.

In an embodiment of the present disclosure, the plurality of features in step S2 include: the characteristics of the overall evaluation level and the component evaluation level comprise a distance, a direction, a mass center and a circumscribed rectangle; the characteristics of the stroke evaluation hierarchy include stroke length, stroke slope, stroke spacing, and stroke curvature.

In an embodiment of the present disclosure, the plurality of features include a centroid feature, a distance feature, a direction feature, an external rectangle feature, a stroke length feature, a stroke inclination feature, a stroke distance relationship feature, and a stroke curvature feature, and the selecting features based on the feature point set in step S2 to obtain the plurality of features includes:

s21, obtaining all feature points covering a single Chinese character framework in a completion mode according to the distribution characteristics of the feature points, calculating an average value according to the abscissa of all feature points covering the single Chinese character framework to obtain the centroid in the abscissa direction, calculating the centroid in the ordinate direction according to the ordinate of all feature points covering the single Chinese character framework to obtain the centroid characteristics;

s22, traversing all feature points of the whole Chinese character or a certain part in the Chinese character, recording the number as m, forming m feature vectors by taking the mass center as a starting point and the feature points as an end point, respectively calculating the length of each feature vector, forming a distance feature number sequence containing m items, and acquiring distance features;

s23, traversing all feature points of the whole Chinese character or a part in the Chinese character, recording the number as m, forming m feature vectors by taking the mass center as a starting point and the feature points as an end point, respectively calculating the direction cosine of each feature vector, forming a direction feature sequence containing m items, and acquiring direction features;

s24, finding the minimum value x of the abscissas of all the feature points in the feature coordinate system_minAnd the maximum value x of the abscissa_maxAnd the minimum value y of the ordinate_minAnd the maximum value y of the ordinate_maxAccording to x_min、 x_max、y_min、y_maxFour coordinate positions (x) determined_min,y_min)，(x_min,y_max)，(x_max,y_min)， (x_max,y_max) Determining the length, width and area of the external rectangle according to the four corner positions of the external rectangle to obtain the characteristics of the external rectangle;

s25, calculating the linear distance between the starting feature point and the ending feature point in each stroke, wherein the obtained linear distance is the stroke length, and acquiring the stroke length feature;

s26, forming a feature vector by traversing all feature points of a certain stroke and concentrating two feature points of any continuous feature point, calculating the cosine value of an included angle between each feature vector and a preset direction, and obtaining the stroke gradient feature by taking the average value obtained based on the cosine values of a plurality of included angles as the stroke gradient;

s27, extracting strokes according to a preset writing sequence, numbering, selecting starting points and ending points in adjacent strokes according to the numbers, forming a plurality of groups of corresponding points through one-to-one correspondence, respectively calculating distances between the corresponding points, averaging, and obtaining stroke distance relation characteristics;

s28, judging whether the three feature points of the starting point, the final point and the equal division point of each stroke are collinear, if the three feature points are not collinear, determining a circumscribed circle according to a triangle formed by connecting the three feature points, calculating the radius of the circumscribed circle as the curvature of the feature points, and acquiring the curvature features of the strokes.

In an embodiment of the present disclosure, step S3 includes:

s31, selecting feature points according to the shape features of the Chinese characters according to a preset standard;

s32, selecting a preset number of single Chinese character images, and labeling the feature points according to the positions of the feature points;

s33, respectively aligning the shapes of the single Chinese character images with the average shape in a preset number in an affine transformation mode;

s34, performing principal component analysis on the aligned shape to obtain a preliminary shape model;

s35, adjusting the characteristic points of the preliminary shape model to the target image to obtain a local gray scale model of the characteristic points;

and S36, performing handwritten shape search on the processed image based on the preliminary shape model and the local gray scale model of the feature points to obtain the labeling features.

In an embodiment of the present disclosure, step S36 includes:

taking the processed image as a test set, covering the processed image with a primary shape model, and recording the coordinates of the feature points as vectors

Performing handwritten character shape search on the processed image according to the preliminary shape model and the local gray level model of the feature points, and determining the optimal matching position of each feature point of the processed image;

all the feature points of the processed image are adjusted to the best matching positions, and the new coordinates of the feature points are the new shape vectors

Will vectorPerforming affine transformation with the new shape vector

The closest requirement is achieved, and an updated vector is obtained

With updated vector

Repeating the handwritten character shape search until reaching the condition of stopping the search, and performing the final vector

The position of the feature point is marked to obtain a marked feature.

In an embodiment of the present disclosure, step S4 includes:

comparing and calculating the characteristics of the standard word and the word to be evaluated according to a corresponding characteristic matching algorithm from the evaluation angles of different characteristics to obtain the characteristic similarity of the standard word and the word with the evaluation;

the feature matching algorithm comprises a Pearson correlation coefficient, cosine similarity, Euclidean distance and a proportional relation;

the feature similarity comprises distance feature similarity obtained through calculation based on a Pearson correlation coefficient, direction feature similarity and stroke gradient feature similarity obtained through calculation based on cosine similarity, centroid feature similarity obtained through calculation based on an Euclidean distance, circumscribed rectangle similarity, stroke length feature similarity, stroke interval feature similarity and curvature feature similarity obtained through calculation based on a proportional relation.

In an embodiment of the present disclosure, step S5 includes:

s51, adding the characteristics obtained in the step S2, the marked characteristic points obtained in the step S3 and the characteristic matching algorithm in the step S4 into a Chinese character characteristic library, wherein the Chinese character characteristic library comprises a characteristic point library and a characteristic matching algorithm library;

and S52, extracting feature points of each new Chinese character to be evaluated based on the Chinese character feature library by using the feature point library through the established active shape model, and calculating the feature similarity of each feature by using the feature comparison algorithm model in the feature matching algorithm library.

In an embodiment of the present disclosure, the automatic chinese character evaluation model in step S6 includes a level evaluation model obtained by using a threshold classification method, a support vector machine and K neighbors, and a percentage evaluation model obtained by using linear regression, where the level evaluation model is a three-level evaluation model.

(III) advantageous effects

The beneficial effects of this disclosure are: the evaluation indexes are quantified by establishing an evaluation standard, effective information of a plurality of characteristics is selected to compare calligraphy works of the character to be evaluated and the standard character from multiple angles, and the characteristic points are extracted and labeled by using the active shape model, so that the labeling accuracy can be improved, the specialty and the objectivity of evaluation work are improved, and the automatic evaluation of the writing attractiveness of the hard-tipped pen Chinese characters is realized by using a computer.

Drawings

FIG. 1 is a flow chart of a method for evaluating a hard-tipped writing work according to an embodiment of the present disclosure;

FIG. 2 is a flowchart of step S0 in one embodiment of the present disclosure;

FIG. 3 is a detailed evaluation criteria and evaluation index chart according to an embodiment of the present disclosure;

FIG. 4 is a flowchart of step S1 in one embodiment of the present disclosure;

FIG. 5 is a standard copybook designed in an embodiment of the present disclosure;

FIG. 6 is a diagram of the original data of a single Chinese character after segmentation in an embodiment of the present disclosure;

FIG. 7 is a graph of a grayscale histogram of a single Chinese character calligraphic image according to an embodiment of the present disclosure;

FIG. 8 is a comparison diagram before and after binarization processing of a calligraphic image in an embodiment of the disclosure;

fig. 9 is a drawing of a pen-script calligraphy evaluation tool in accordance with an embodiment of the present disclosure;

FIG. 10 is a flowchart of step S3 in one embodiment of the present disclosure;

FIG. 11 is a schematic diagram illustrating feature point movement according to an embodiment of the present disclosure;

FIG. 12 is a graph illustrating a trend of local texture size changes according to an embodiment of the present disclosure;

fig. 13 is a diagram of a method for searching a matching position with a best feature point according to an embodiment of the disclosure.

Detailed Description

For the purpose of better explaining the present disclosure, and to facilitate understanding thereof, the present disclosure will be described in detail below by way of specific embodiments with reference to the accompanying drawings.

All technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein in the description of the disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

Based on the above, the present disclosure provides a hard-tipped writing work judging method based on a field expert visual angle, which includes: judging standard making and judging model building, wherein the judging standard making mainly comprises the steps of arranging the judging standard according to the field expert visual angle and quantizing the judging index; the establishment of the judgment model comprises the establishment of a judgment data set, the selection of characteristics, the automatic marking of the characteristics, the comparison and calculation of the characteristics, the establishment of a Chinese character characteristic library and the establishment of an automatic judgment model.

Fig. 1 is a flowchart of a method for evaluating a hard-tipped writing work according to an embodiment of the present disclosure, as shown in fig. 1, including the following steps:

as shown in fig. 1, in step S1, a hard-pen chinese character image is obtained based on the evaluation criterion, and an evaluation data set is established based on the hard-pen chinese character image;

as shown in fig. 1, in step S2, feature selection is performed on the chinese characters in the evaluation data set based on a feature point set to obtain a plurality of features, where the feature point set is a set composed of all feature points constituting each chinese character;

as shown in fig. 1, in step S3, automatically labeling feature points of a plurality of features based on the active shape model to obtain labeled features;

as shown in fig. 1, in step S4, comparing and calculating the standard word in the evaluation data set with the word to be evaluated according to the labeled features to obtain a plurality of feature similarities;

as shown in fig. 1, in step S5, a chinese character feature library is established according to the plurality of features, the labeled feature points and the feature matching algorithm;

as shown in fig. 1, in step S6, an automatic Chinese character evaluation model is established based on the corresponding scoring labels and feature similarities of the Chinese characters in the Chinese character feature library.

Based on the judging method, the characteristics of each Chinese character judging standard are selected, different algorithms are selected from multiple angles to establish a calligraphy work judging model, and Chinese characters with different structures and different evaluating angles have better judging accuracy.

The specific implementation of the steps of the embodiment shown in fig. 1 is described in detail below:

in step S1, a hard-tipped Chinese character image is obtained based on the evaluation criterion, and an evaluation data set is established based on the hard-tipped Chinese character image.

In an embodiment of the present disclosure, step S1 is preceded by:

s0, constructing judgment criteria, wherein the judgment criteria comprise the evaluation content and the text description of the evaluation content.

Fig. 2 is a flowchart of step S0 in an embodiment of the disclosure, and as shown in fig. 2, step S0 specifically includes the following steps:

in step S01, the evaluation content is divided into three evaluation levels of whole, part and stroke according to the structure of the Chinese character.

In step S02, each evaluation level includes a plurality of evaluation elements, and the evaluation elements in the overall evaluation level at least include stroke length ratio, stroke inclination, stroke type, stroke distance relationship, and similar stroke parallel relationship; the evaluation elements in the component evaluation hierarchy at least comprise component forms, component density distances and component proportions; the evaluation elements in the stroke evaluation hierarchy at least comprise the form, position and size of the whole character.

In step S03, the text description of the evaluation content is detailed in such a manner that the evaluation elements in each evaluation level correspond to specific evaluation elements.

In this embodiment, in the process of making the evaluation criterion, firstly, the evaluation criterion is discussed with experts in the calligraphy field, referring to the "outline of education for calligraphy in middle and primary schools", and the experts summarize the following overall evaluation criterion for chinese characters from the teaching perspective for hard-tipped pen calligraphy in primary schools (taking a hard-tipped pen script as an example in this embodiment), as shown in table 1:

TABLE 1 Total evaluation of script calligraphy in hard-tipped pen

Secondly, according to the evaluation criteria shown in the table 1, combining the focus of experts in the evaluation work, splitting the evaluation criteria according to the three levels of the whole body, the part and the strokes in the Chinese character body structure, and combing the evaluation elements in the evaluation work based on the teaching level, as shown in the table 2:

TABLE 2 evaluation factors for hard-tipped pen regular script calligraphy

In step S04, the evaluation hierarchy, the evaluation elements, and the details of the evaluation elements are all converted from natural language to computer language, and the evaluation criteria having quantitative evaluation indexes are obtained.

The evaluation criteria are refined, detailed evaluation rules are formulated, each evaluation rule forms a quantitative index, for example, a calligraphy field expert can be invited to study, evaluation is respectively carried out from three levels of whole characters, parts and strokes according to the professional perspective of the expert, and the evaluation rules are determined. And then converting the evaluation rules into a computer language for expression to form quantifiable evaluation indexes, wherein fig. 3 is an evaluation standard rule and an evaluation index graph in one embodiment of the disclosure, and as shown in fig. 3, the stroke level comprises the evaluation indexes of a single stroke and the evaluation indexes among strokes.

After the step S0, the expert view is used to make the judgment index and the quantitative judgment, the judgment data set is established in the step S1, and the data category and the establishment direction are determined based on the judgment criteria. In order to ensure that the hard-stroke Chinese character images which are abundant and diverse enough can be collected finally to cover all the structure types specified in the evaluation standard, the copybook can be designed by self, the Chinese characters required to be copied by students are specified as the guide direction, and the collected data is preprocessed and labeled.

Fig. 4 is a flowchart of step S1 in an embodiment of the disclosure, and as shown in fig. 4, step S1 specifically includes the following steps:

in step S11, an entire signature image obtained after the writer writes in accordance with the standard signature is acquired.

The standard copybook is a standard Chinese character copybook, the design process is that through the discussion with calligraphy experts, in 3500 Chinese characters commonly used by students in middle and primary schools, sample characters of all 5 structures including single-body characters, left (middle) and right structures, upper (middle) and lower structures, semi-surrounding structures and full-surrounding structures are selected, the total number of the sample characters is 960, and the sample characters are manufactured into copybooks, the total number of the copybooks is 32 pages, each page has 30 standard Chinese characters, each Chinese character has a standard Chinese character demonstration, and the calligraphy experts write the reference when the students copy and copy. And writing areas with grid and blank grid attached to simulate the common situation of primary and secondary school students when writing, so that they can copy and copy, and fig. 5 is a standard copybook designed in an embodiment of the present disclosure.

And then, delivering the designed standard copybook to a publishing company for publishing and printing, distributing the copybook to students in middle and primary schools through two schools in a certain place, guiding the students by a teacher in a calligraphy class to finish writing the copybook, covering the students from three to nine grades by a writer, collecting all the copybooks back to a laboratory after the writing is finished, and manually converting the written copybook into an electronic image for storage by using a scanner.

In step S12, the entire signature image is divided into single chinese character images. In order to facilitate the analysis of the writing condition of each Chinese character, the entire copybook image needs to be divided into single Chinese character images, and fig. 6 is an original data diagram of a single Chinese character after division in an embodiment of the present disclosure.

In step S13, the single kanji image is preprocessed to obtain a processed image.

In an embodiment of the present disclosure, the preprocessing includes denoising, binarization, and image resolution resizing, among others. The denoising process uses median filtering, the binarization process adopts a method of independently calculating a threshold value for each picture, the specific flow is to count the gray value of each Chinese character image to form a gray level histogram, and fig. 7 is a calligraphy image gray level histogram of a single Chinese character according to an embodiment of the disclosure. As shown in fig. 7, the whole graph will exhibit two peak values, one of which is concentrated on the main background color and the other one is concentrated on the handwriting color, the median of the two peak values is taken as the threshold value to perform binarization processing, the pixel value of the pixel point greater than the threshold value is set to 255, and the pixel value of the pixel point less than the threshold value is set to 0. Finally, the size of all the images is adjusted to 70 × 70 resolution, and fig. 8 is a comparison graph before and after the binarization processing of the calligraphic image in the embodiment of the disclosure.

In step S14, the processed image is labeled according to the evaluation criterion, and the labeled data constitutes an evaluation data set.

In one embodiment of the present disclosure, in order to implement computer evaluation, score labels (i.e., scoring labels) required for evaluation need to be given to all data to be sent to a training model as training labels for learning. In order to facilitate the marking work, the method realizes the evaluation tool of the regular script calligraphy in the hard-tipped pen based on the evaluation standard, the tool is internally provided with all standard character samples and the evaluation standard, and corresponding explanation is given to each standard rule. The overall scoring of the Chinese characters is divided into two scoring modes of 5-level system and percentage system, the other evaluation standards are 5-level system, 1 level represents the worst, 5 level represents the best, and fig. 9 is a tool diagram for evaluating the hard-tipped regular script calligraphy in one embodiment of the disclosure.

In step S2, a plurality of features are obtained by performing feature selection on the chinese characters in the evaluation data set based on the feature point set.

The characteristic point set is a set formed by all the characteristic points forming each Chinese character, the end, the fold, the divergence and the intersection points on the strokes of the Chinese character and the key background points on the background of the Chinese character are usually called Chinese character characteristic points, reflect the essential characteristics of the structure of the Chinese character and centralize the main structure information of the Chinese character. The plurality of features in step S2 of the present embodiment include a centroid feature, a distance feature, a direction feature, an external rectangle feature, a stroke length feature, a stroke inclination feature, a stroke distance relationship feature, and a stroke curvature feature.

The method designs 12 mathematical features which can be directly used for calculation, wherein the whole level and the component level respectively comprise four features of distance, direction, mass center and circumscribed rectangle, and the stroke level comprises four features of stroke length, stroke gradient, stroke distance and stroke curvature. In order to extract the features from the image, the present disclosure further provides a feature representation method based on a feature point set, which is essentially to perform an operation by using coordinates of all relevant feature point sets constituting the features to obtain a sequence or a vector representing the features. In step S2, the description and operation method for each feature in the process of obtaining multiple features by performing feature selection based on the feature point set is as follows:

(1) obtaining all feature points covering a single Chinese character framework in a completion mode according to the distribution characteristics of the feature points, calculating an average value according to the abscissa of all the feature points covering the single Chinese character framework to obtain the centroid in the abscissa direction, calculating the centroid in the ordinate direction according to the ordinate of all the feature points covering the single Chinese character framework to obtain the centroid characteristics. The centroid of the image is also called as the gravity center of the image, exists in the whole level and the component level, and directly reflects the distribution condition of strokes of the Chinese characters and the position of the gravity center of the Chinese characters in the grid of the Chinese characters.

The following method is designed for extracting the centroid characteristics of the image: and (1-1) calculating all points covering the Chinese character skeleton through the characteristic points. According to the characteristics of the characteristic points selected by the Chinese characters, all the points in the Chinese character skeleton can be obtained by complementing all the points between every two points, and because the hard-tipped Chinese characters are only images formed by line strokes in visual observation, all the points in the Chinese characters are represented by the Chinese character skeleton. And (1-2) respectively calculating the mass centers of the x direction and the y direction of the abscissa, wherein the mass center of the x direction is the average value of the abscissa of all the points, and the mass center of the y direction is the average value of the ordinate of all the points.

(2) Traversing all feature points of the whole Chinese character or a certain part in the Chinese character, recording the number as m, forming m feature vectors by taking the mass center as a starting point and the feature points as an end point, respectively calculating the length of each feature vector, forming a distance feature number sequence containing m items, and obtaining distance features.

The distance features exist in the whole level and the component level, and are mainly vector length features of a vector group formed by a feature point set. The feature extraction strategy is to form feature vectors for all feature points and the mass center and calculate the length | A of any vector_IB_IAnd | forming a set of distance feature series.

The following method is designed for extracting the distance features to the calligraphy image: and (2-1) traversing all feature points (the number is recorded as m) of the whole or a part. (2-2) taking the centroid feature point as a starting point (A) and the rest feature points as an end point (B) to show that each feature point vector is AB_iForming a vector group X containing m vectors: AB₀，AB₁，……，AB_m. (2-3) calculating the length of each vector | AB₀I, forming a sequence of m terms.

(3) Traversing all feature points of the whole Chinese character or a part in the Chinese character, recording the number as m, forming m feature vectors by taking the mass center as a starting point and the feature points as an end point, respectively calculating the direction cosine of each feature vector, forming a direction feature number sequence containing m items, and obtaining direction features.

The direction features exist in the whole level and the component level, and are mainly vector direction cosine features of a vector group formed by a feature point set. The feature extraction strategy is to form feature vectors for all feature points and the mass center, calculate the direction cosine of any vector and form a group of direction feature number sequences.

The following method is designed for extracting the direction features of the calligraphy image: and (3-1) traversing all feature points (the number is recorded as m) of the whole or a part. (3-2) taking the centroid feature point as a starting point (A) and the rest feature points as an end point (B) to show that each feature point vector is AB_iForming a vector group X containing m vectors: AB₀，AB₁，……，AB_m. (3-3) calculating each vector separatelyThe direction cosine forms a direction value sequence containing m terms.

(4) The character outline rectangle feature is an important feature for describing and evaluating the shape and appearance of the Chinese character and the size of the character lattice occupied by the Chinese character. The following method is designed for extracting the external rectangle characteristics of the Chinese characters: (4-1) finding the minimum value x of the abscissas of all the feature points in the feature coordinate system_minAnd the maximum value x of the abscissa_maxAnd the minimum value y of the ordinate_minAnd the maximum value y of the ordinate_maxAccording to x_min、x_max、y_min、y_maxFour coordinate positions (x) determined_min,y_min)，(x_min,y_max)，(x_max,y_min)，(x_max,y_max) As the four corner positions of the external rectangle. And (4-2) determining the length, width and area of the external rectangle according to the four corners of the external rectangle to obtain the characteristics of the external rectangle.

(5) The stroke length characteristic is the key point of stroke evaluation, and the characteristic extraction strategy and the specific method are as follows: and calculating the linear distance between the initial stroke characteristic point (A) and the final stroke characteristic point (B) in each stroke, wherein the obtained linear distance is the stroke length, and acquiring the stroke length characteristic.

(6) The stroke gradient characteristic is that transverse drawing is taken as a main characteristic, partial strokes in a Chinese character usually have proper inclination angles, and the following method is designed for extracting the stroke gradient characteristic of the Chinese character: (6-1) traversing all feature points of a certain stroke, and forming a feature vector A for two feature points in any continuous feature point set_iB_i. (6-2) calculating each feature vector A_iB_iAnd acquiring stroke gradient characteristics by taking an average value obtained based on the plurality of included angle cosine values as stroke gradient with the included angle cosine value of the preset direction (the direction that y is 0).

(7) The stroke space relation is expressed between the adjacent strokes, and the uniform white distribution and the uniform white retention in the Chinese character evaluation standard can be expressed by calculating the space relation of the adjacent strokes. The stroke distance relation feature extraction method comprises the following steps: (7-1) extracting strokes according to a preset writing sequence and numbering 0 and 1 … …; (7-2) selecting phases based on the numbersThe skeleton points of the strokes with the same number of starting points (A) and ending points (B) in the adjacent strokes are recorded as Q_iIf the distance between the two corresponding points is calculated, the two corresponding points are required to be taken from the approximate relative positions of the two adjacent strokes to calculate the distance, so the points taken from the two strokes are the same, namely the stroke skeleton points with the same quantity; and (7-3) forming a plurality of groups of corresponding points by corresponding the starting point, the final point and the like of two adjacent strokes one by one, respectively calculating the distances between the corresponding points and averaging to obtain the stroke distance relation characteristic.

(8) The curvature of the curve is the rotation rate of the tangent direction angle to the arc length of a certain point on the curve, and indicates the degree of deviation of the curve from a straight line, and the larger the curvature is, the larger the bending degree of the curve is. The stroke curvature characteristics can describe the bending degree of strokes such as left falling, right falling and the like, and the stroke curvature characteristic extraction method comprises the following steps: (8-1) selecting three characteristic points of the stroke, namely a double-end point and an equant point; (8-2) judging whether the three feature points of the starting point, the ending point and the equant point of each stroke are collinear; (8-3) if the three feature points are not collinear, determining a circumscribed circle according to a triangle formed by connecting the three feature points, calculating the radius of the circumscribed circle as the curvature of the feature points, and acquiring the stroke curvature feature.

In step S2, the characteristics of the chinese character image are selected based on the quantized evaluation indexes, the evaluation indexes with higher coupling degree are subjected to manual dimensionality reduction, and low-dimensionality and weak-correlation characteristics that can cover and represent all indexes are summarized to replace the indexes for calculation.

In step S3, feature points of the plurality of features are automatically labeled based on the active shape model, and labeled features are obtained.

Fig. 10 is a flowchart of step S3 in an embodiment of the present disclosure, which specifically includes the following steps:

in step S31, feature points are selected for the shape features of the chinese characters according to preset criteria. The selected positions are classified into a contour method and a skeleton method. The outline method is that the characteristic points are uniformly selected along the outline of the stroke, the characteristic points are basically positioned at the outer edge of the stroke, and the complete form of the stroke can be shown; the skeleton method is that all feature points are in the middle of the stroke, and only the skeleton form of the stroke can be shown. Different methods are selected according to the stroke characteristics and the overall structure of the Chinese character, for example, the horizontal drawing considers the problem of stroke skeletons in the evaluation process so as to select a skeleton method, and the vertical drawing needs to distinguish the stroke forms in the evaluation process so as to select a contour method capable of expressing form information in multiple ways.

According to the three characteristics and labeling methods that must be provided for feature point selection, the more detailed and specific feature point selection principle is summarized as shown in table 3:

TABLE 3 Chinese character feature point selection principle

In step S32, a preset number of single chinese character images are selected, and feature points are labeled according to the positions of the feature points.

In the step, 50 single Chinese character images are selected, the number of Chinese character feature points and the positions of all image feature points are determined according to a table 3 and are stored in a text file, the suffix of the file is pts, the first line in the file represents the version number of a label format, the second line represents the number of a label point set, in the following { }, according to the sequence from top to bottom, the coordinate position of one feature point in each line, the first number represents an abscissa value, and the second number represents an ordinate value. In order to standardize the feature point representation method, when the feature points are recorded and stored, the horizontal coordinates and the vertical coordinates of the feature points are respectively described by the feature points (x, y), the number of the feature points of the Chinese characters is represented by n, and the feature points of the sample set pictures are labeled according to the labeling sequence strictly so as to ensure that the sequence of the labeled feature points is consistent.

In step S33, a preset number of single kanji images are shape-aligned with the average shape, respectively, by way of affine transformation.

Align 50 images with the average shape, each time the new shape is aligned with the model average shapeIn the process of searching for a reasonable position, affine transformation is carried out through a following formula, so that the Euclidean distance of each feature point of the current shape is minimum. Wherein T is the shape characteristic point matrix needing to be transformed currently, and the aim is to make the characteristic point set of the current transformation and the average characteristic point set closest. Wherein x_t、y_tThe offset in the x-axis and y-axis directions, r is the scaling factor, and θ is the rotation angle.

In step S34, principal component analysis is performed on the aligned shape to obtain a preliminary shape model.

By aligning the training samples, 50 shapes are obtained, denoted as { F }₁，F₂，…，F₅₀Their average shape }

Can be expressed as:

from the covariance formula:

then, singular value decomposition is carried out on the covariance matrix to obtain an eigenvalue lambda and an eigenvector P, after the eigenvalue lambda is arranged from large to small, 50 lambdas are selected, wherein [ lambda ] is₁，λ₂，λ₃，...，λ₅₀]Satisfies the following conditions:

where η is a self-settable weight threshold, set here at 94%.

The eigenvector corresponding to these n eigenvalues is P ═ P₁，p₂，…，p₅₀]And constructing a statistical model of the training set by P:

where S is a statistical model of the training set constructed from P,

is the average of the statistical model from S.

At this time, a new vector of the chinese character shape in the hard-tipped pen can be obtained each time the value of the weighting coefficient vector b is changed. Meanwhile, in order to prevent the final shape vector change from being excessively deformed and maintain the original basic shape, the ASM model also has certain constraint on the value of the weighting coefficient. Carrying out statistical analysis on the probability distribution of the parameters to obtain the value range of the parameter b,

wherein λ is_iIs the ith eigenvalue of Cov.

In step S35, the feature points of the preliminary shape model are adjusted to the target image, and a local gray scale model of the feature points is obtained.

And slightly adjusting the primary shape obtained in the last step to move each point to a better position, and establishing a corresponding local gray model for each point. FIG. 11 is a schematic diagram illustrating feature point movement according to an embodiment of the disclosure, and as shown in FIG. 11, let the ith feature point of the jth picture be X_iAt this time, cross X_iMaking a straight line l, l perpendicular to X_i-1And X_i+1On the straight line, X_iMoving within a certain range along l in order to adjust the feature points towards the edges of the target image for a better fit of the image. FIG. 12 is a graph of the variation trend of local texture size, which is derived from the gray values included in the vector during the moving process to obtain a local texture g_ijThe trend of the magnitude of the change is shown in FIG. 12, g_ijThe maximum value of (2) represents the position of the edge of the target image, and 50 trainings can be obtained by performing the same operation on all training samplesTraining the texture characteristics of the ith point of the sample, and averaging:

variance:

thereby obtaining a local gray model of the feature point.

In step S36, a handwritten shape search is performed on the processed image based on the preliminary shape model and the local grayscale model of the feature point, so as to obtain an annotation feature.

The method comprises the following specific steps: (36-1) taking the processed image as a test set, covering the processed image with a preliminary shape model, and recording the coordinates of the feature points as vectors

And (36-2) performing handwritten character shape search on the processed image according to the preliminary shape model and the local gray level model of the feature points, and determining the best matching position of each feature point of the processed image. For the shape search, the essence is to find the best matching position of each feature point in a new picture, fig. 13 is a method diagram for searching the best matching position of the feature points in an embodiment of the present disclosure, as shown in fig. 13, k points are taken from the left and right of the ith feature point as the center to form 2k +1 candidate points, and each candidate point takes its p points adjacent to the left and right as the sub-local features of the candidate point, so that the sub-local features of 2(k-p) +1 candidate points are obtained. In fig. 13, the circular points are 2k +1 candidate points, and the square points represent 5 points included in the sub-local feature of the 8 th candidate point when p is 2. Suppose the sub-local feature of the j-th feature point is represented as g'_ξξ ═ 1, 2,.., 2(k-p) +1, and the mahalanobis distance D is calculated according to the following formula_ξ，

(36-3) after traversing each candidate point, selecting the point with the minimum Mahalanobis distance as the best matching position of the feature point. All the feature points of the processed image are adjusted to the best matching positions, and the new coordinates of the feature points are the new shape vectors

(36-4) vector

Performing affine transformation with the new shape vector

The closest requirement is reached (wherein the closest can be obtained by continuously adjusting the rotation angle, the scaling size and the displacement in the affine transformation process)

And vector

Determined by the minimum of the inter-euclidean distances), resulting in updated vectors

With updated vector

The handwritten character shape search described above is repeated until a condition for stopping the search is reached (until the condition for stopping the search is reached)

No longer change or maximum iteration search stop is reached), the last vector is searched

The position of the feature point is marked to obtain a marked feature. At this moment, findThe positions of the 21 feature points are the positions of the feature points extracted for this image.

In order to solve the problem that the image characteristics of the hard-tipped pen Chinese characters are difficult to label in the step S3 in the process of evaluating the writing of the hard-tipped pen Chinese characters, an active shape model based on point distribution is adopted for training, and the training comprises two parts of establishing an ASM model and utilizing the established ASM model for characteristic selection.

In step S4, the standard word in the evaluation data set is compared with the word to be evaluated according to the labeled features and calculated, so as to obtain a plurality of feature similarities.

After image features capable of reflecting all evaluation indexes are extracted, comparing and calculating features of the standard word and the word to be evaluated according to corresponding feature matching algorithms from evaluation angles of different features to obtain feature similarity of the standard word and the word with the evaluation index;

the feature matching algorithm comprises a Pearson correlation coefficient, cosine similarity, Euclidean distance and a proportional relation; the feature similarity comprises distance feature similarity obtained through calculation based on a Pearson correlation coefficient, direction feature similarity and stroke gradient feature similarity obtained through calculation based on cosine similarity, centroid feature similarity obtained through calculation based on an Euclidean distance, circumscribed rectangle similarity obtained through calculation based on a proportional relation, stroke length feature similarity, stroke distance feature similarity and curvature feature similarity.

Step S4, designing different similarity comparison methods according to the evaluation angles of different characteristics and the characteristics of the calligraphy image, so as to calculate the similarity of the standard word and the word to be evaluated on the characteristics, wherein the specific implementation method is as follows:

the pearson correlation coefficient, which is a quantity that studies the degree of linear correlation between two variables, has a value between-1 and 1, is often denoted as r, and is defined by the formula:

where Cov (x, y) is the covariance of x and y, Var [ x ] is the variance of x, and Var [ y ] is the variance of y. The larger the correlation coefficient result, the better the correlation degree of x and y is generally indicated; the smaller the correlation coefficient result, the worse the correlation degree of x and y is generally indicated; when the correlation coefficient calculation result is 1, it is proved that a linear relationship exists between x and y.

The specific implementation method of the distance feature similarity based on the Pearson correlation coefficient comprises the following steps: (1) respectively calculating m length characteristic sequences of the word to be evaluated and m length characteristic sequences of the standard word; (2) and calculating a correlation coefficient of the corresponding sequence of the word to be evaluated and the standard word to obtain the distance characteristic similarity.

The cosine similarity measure can measure the vector similarity between any dimensions, and is especially more applied to high-dimensional space. Cosine similarity measures the similarity between two vector inner product space included angles by measuring cosine values of the two vector inner product space included angles. The closer the cosine value is to 1, the closer the directions of the vectors are, i.e. the more similar the two vectors are, the more similar the cosine similarity is calculated by the following formula:

the specific implementation method of the direction feature similarity based on the cosine similarity comprises the following steps: (1) respectively calculating all feature vectors of the word to be evaluated and all feature vectors of the standard word; (2) and calculating cosine similarity corresponding to the feature vector of the word to be evaluated and the standard word, wherein the obtained cosine similarity mean value is the direction feature similarity.

The specific implementation method of the stroke gradient characteristic similarity based on the cosine similarity comprises the following steps: (1) acquiring vector representation of strokes of the word to be evaluated and strokes of the corresponding standard word; (2) and calculating cosine similarity between the two vectors, wherein the obtained result is the stroke gradient characteristic similarity.

The Euclidean distance is a most common distance calculation formula, can measure the absolute distance of two points in a multi-dimensional space, and is suitable for very dense and continuous data. Assuming that x, y are two points in an n-dimensional space, the euclidean distance calculation formula is:

the specific implementation method of the centroid feature similarity based on the Euclidean distance comprises the following steps: (1) respectively acquiring centroid characteristics of the character to be evaluated and the standard character in the x direction and the y direction; (2) and calculating the Euclidean distance between the character to be evaluated and the standard character centroid, wherein the obtained result is the centroid feature similarity.

The proportion relation is the proportion of the character to be evaluated and the relevant characteristics of the standard character:

firstly), the similarity of the circumscribed rectangles is based on the area proportional relation between the word to be evaluated and the circumscribed rectangles of the standard word, and the specific method is as follows: (1) respectively calculating the areas s1 and s2 of the circumscribed rectangles of the word to be evaluated and the standard word; (2) and (5) calculating the area proportion relation s-s 1/s2 to obtain the similarity of the circumscribed rectangle features.

Secondly), the stroke length is based on the proportional relation between the stroke length of the character to be evaluated and the stroke length of the standard character, and the specific method comprises the following steps: (1) respectively calculating stroke lengths l1 and l2 corresponding to the strokes of the character to be evaluated and the standard character; (2) and (5) calculating the length proportional relation r ═ l1/l2, and obtaining the similarity of the stroke length characteristics.

Thirdly), the stroke distance is based on the proportional relation between the stroke distance of the character to be evaluated and the stroke distance of the standard character, and the specific method is as follows: (1) and respectively calculating stroke distance values d1 and d2 corresponding to the strokes of the word to be evaluated and the standard word. (2) And (4) calculating the proportional relation d-d 1/d2, and obtaining the stroke pitch feature similarity.

Fourthly) the curvature characteristic similarity is based on the proportional relation between the stroke curvature R1 of the character to be evaluated and the stroke curvature R2 of the standard character, and the specific implementation method is as follows: (1) and respectively calculating the curvatures R1 and R2 of the strokes of the word to be evaluated and the strokes corresponding to the standard word. (2) And (5) calculating the curvature proportional relation R-R1/R2, and obtaining the curvature characteristic similarity.

In step S5, a chinese character feature library is created according to the plurality of features, the labeled feature points, and the feature matching algorithm.

Firstly, adding the features obtained in S2, the labeled feature points obtained in S3 and the feature matching algorithm in S4 into a Chinese character feature library, wherein the Chinese character feature library comprises a calligraphy image feature point library and a feature matching algorithm library which comprise definitions of each Chinese character feature.

Secondly, extracting feature points of each new Chinese character to be evaluated based on a Chinese character feature library through the established active shape model, and calculating the feature similarity of each feature by using a feature comparison algorithm model in a feature matching algorithm library.

In step S6, an automatic Chinese character evaluation model is established based on the corresponding scoring labels and feature similarities of the Chinese characters in the Chinese character feature library.

The Chinese character automatic judging model is divided into grade judgment and percentile judgment according to the judging form. And taking the feature similarity and the grading marks of the Chinese characters as the input of an automatic judging model, and training and adjusting the model by setting and modifying parameters. The evaluation model based on grade evaluation adopts three methods, namely a threshold value grading method, a support vector machine and K neighbor, respectively. The evaluation model based on the percentage evaluation adopts a linear regression evaluation model. The final evaluation result is divided into three levels of strokes, components and the whole, wherein each level gives two scores, namely a level evaluation score and a percentile evaluation score. In the process of establishing the classifier model, repeated discussion is carried out with experts and experimental results are combined, so that the three-level rating system is found to have similar instruction function from the teaching perspective compared with the five-level system, but the experimental results are more accurate, and therefore, the model based on the level evaluation in the invention adopts the three-level evaluation system.

1) The method comprises the steps of threshold classification, wherein the Chinese character grade evaluation can be divided into a good grade, a middle grade and a poor grade according to the similarity between the student calligraphy characters and the standard characters, the higher the similarity between the student calligraphy characters and the standard characters is, the better the student calligraphy characters are, the higher the evaluation is, and the quality of the Chinese characters is continuously reduced along with the continuous reduction of the similarity. The threshold classification is the most intuitive result of classification, and has the characteristics of simplicity, high efficiency, easy calculation and the like. The threshold value can be determined by analyzing probability density distribution of different types of data, such as a nuclear density graph.

Experimental data from the established data set, 200 pieces of image data were selected for each chinese character. And respectively assigning a weight to each similarity characteristic value from the three levels of the whole, the parts and the strokes, sequentially increasing or decreasing the weight of a certain characteristic value from equal weight until the optimal weight of different Chinese characters is obtained, and obtaining the final characteristic value through weighting calculation. And (4) respectively calculating a denucleation density curve according to three evaluated grades by using the characteristic values:

the balance points between the three levels are taken as threshold segmentation points.

It should be understood that the nuclear density curve method is only for convenience of describing the embodiment of the present invention, and does not indicate or imply that the referred threshold determination method is a nuclear density estimation method, and thus, is not to be construed as limiting the present invention.

2) The support vector machine is a two-classification model, and the method is based on a VC (virtual C) dimension theory and a structure risk minimization principle of a statistical learning theory, and seeks an optimal compromise between the complexity and the learning capability of the model according to limited sample information so as to obtain the best popularization capability.

Although SVMs were originally implemented for two classes, SVMs can also be used for multi-class problems, and common construction methods fall into two categories, one-to-one and one-to-many. One-to-one approach, namely constructing k (k-1)/2 classifiers, each of which trains two different classes of data, uses a voting strategy in the classification. Experimental data from the established data set, 200 pieces of image data were selected for each chinese character. Respectively calculating twelve similarity characteristic values from the three levels of the whole body, the components and the strokes, and carrying out normalization processing on each characteristic value to form sample characteristics

Calculating to make the sample set linearly separable hyperplane:

w^Tx+b＝0

and the following conditions are met, and the sample set categories can be correctly classified:

the points of the two types of samples that are closest to the hyperplane and parallel to the training samples on the hyperplane are called support vectors. The separation of the two classes is the sum of the distances of the support vectors to the hyperplane:

and intuitively, the optimal hyperplane is the hyperplane positioned in the positive middle of the positive and negative samples, so the problem of finding the optimal hyperplane can be converted into the problem of finding the maximum interval in the sample space. To maximize the separation of the two classes, i.e.

And (3) minimum, simultaneously satisfying the constraint conditions of the following formula, and establishing an SVM model optimization problem:

the SVM model optimization problem is a convex quadratic programming problem, a Lagrangian function is introduced, and the corresponding Lagrangian function of the problem is as follows:

since the function has strong duality, let L (w, b, a) have a partial derivative of w and b as 0, and solve α to obtain the SVM model:

3) the core idea of the K-Nearest Neighbor (KNN) algorithm is to determine the class of a sample to be classified by judging the class of K samples near a certain sample to be classified, that is, if most of K samples near the certain sample belong to class a, the sample also belongs to class a. Experimental data from the established data set, 200 pieces of image data were selected for each chinese character. Respectively calculating 12 similarity characteristic values from the three levels of the whole body, the parts and the strokes, carrying out normalization processing on each characteristic value, and then sending the characteristic values into a KNN model for training.

4) The linear regression prediction method is used for expressing the relation degree of linear correlation of two variables. The experimental data is from the data set established, each Chinese character selects 200 pieces of image data, four overall feature similarity calculation result values of all the image data are calculated respectively, normalization processing is carried out on each feature value, the feature value is used as an independent variable x, the image evaluation and marking result is used as a dependent variable h (x), a linear relation exists between the feature value and the dependent variable x, and a linear regression model is established as follows:

h_θ(x)＝θ₀+θ₁x

many linear regression models can be fitted by linear regression algorithms. But different models have different fitting capabilities for the data. The final goal of linear regression is to find the best model describing the linear relationship between the data, i.e. the model with the least difference between the prediction model and the actual result, which is called the loss function. Therefore, the problem of finding the optimal linear regression model, that is, finding the model with the minimum loss function corresponding to the model, the cost function formula:

substituting the linear regression model formula into the loss function formula, and applying the parameter theta₀And theta₁As an argument of function J, the core objective of solving the problem is obtained:

the least squares method available to solve the cost function, for J (θ)₀,θ₁) Theta in (1)₀And theta₁Respectively obtaining the following derivatives:

when the two formulas are 0, theta can be obtained₀And theta₁Optimal solution:

it should be understood that the least squares method is only for convenience of describing the embodiments of the present invention, and does not indicate or imply that the indicated method of solving the cost function is the least squares method, and therefore, should not be construed as limiting the present invention.

In summary, the hard-pen calligraphy evaluation method based on the field expert visual angle provided by the embodiment of the disclosure can solve the problems that the evaluation work in the teaching scenes of primary and middle schools is subjective and lacks of professionality, and the existing method is low in accuracy and single in evaluation angle due to complex Chinese character structures, and can achieve professional and objective evaluation of the field expert visual angle on the hard-pen calligraphy works of students by a computer, and has the following advantages:

1) a one-word one-modeling form is adopted, each Chinese character is manually designed according to an evaluation standard, evaluation characteristics are selected, different algorithms are selected from multiple angles to establish a calligraphy work evaluation model, the advantages of the algorithms are fully exerted, and the evaluation accuracy rate is high for the Chinese characters with different structures and different evaluation angles.

2) The active shape model is generally applied to face recognition and medical images, and the application of the whole character feature labeling of the hard-tipped pen Chinese characters is in a blank state.

3) On the basis of summarizing and concluding the evaluation standard of the hard-tipped pen calligraphy, the evaluation standard is quantized to meet the requirement of informatization processing, the possibility that the evaluation of the middle and primary school hard-tipped pen calligraphy is changed from artificial evaluation limited by subjective factors into accurate and scientific image field processing technology is explored, the advantage of image processing is expanded to the calligraphy field, then systematic analysis and evaluation are carried out on the hard-tipped pen calligraphy, the accuracy of calligraphy evaluation is improved, and the calligraphy evaluation is more intelligent and automatic.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for judging hard-tipped calligraphy works is characterized by comprising the following steps:

2. The method for evaluating hard-tipped writing instruments according to claim 1, wherein step S1 is preceded by the steps of:

step S0 includes:

3. The method for evaluating hard-tipped writing instruments according to claim 1, wherein step S1 comprises:

s12, dividing the whole copybook image into single Chinese character images;

4. The method for evaluating a hard-tipped writing work according to claim 2, wherein the plurality of features in step S2 include: the characteristics of the overall evaluation level and the component evaluation level comprise a distance, a direction, a mass center and a circumscribed rectangle; the characteristics of the stroke evaluation hierarchy include stroke length, stroke slope, stroke spacing, and stroke curvature.

5. The method for evaluating hard-tipped calligraphic works according to claim 4, wherein the plurality of features includes a centroid feature, a distance feature, a direction feature, an external rectangle feature, a stroke length feature, a stroke inclination feature, a stroke distance relationship feature and a stroke curvature feature, and the selecting features based on the feature point set in step S2 includes:

s24, finding the minimum value x of the abscissas of all the feature points in the feature coordinate system_minAnd the maximum value x of the abscissa_maxAnd the minimum value y of the ordinate_minAnd the maximum value y of the ordinate_maxAccording to x_min、x_max、y_min、y_maxFour coordinate positions (x) determined_min,y_min)，(x_min,y_max)，(x_max,y_min)，(x_max,y_max) Determining the length, width and area of the external rectangle according to the four corner positions of the external rectangle to obtain the characteristics of the external rectangle;

6. The method for evaluating hard-tipped writing instruments according to claim 3, wherein step S3 comprises:

7. The method for evaluating hard-tipped writing instruments according to claim 6, wherein step S36 comprises:

Will vector

Performing affine transformation with the new shape vector

The closest requirement is achieved, and an updated vector is obtained

With updated vector

The position of the feature point is marked to obtain a marked feature.

8. The method for evaluating hard-tipped writing instruments according to claim 4, wherein the step S4 comprises:

9. The method for evaluating hard-tipped writing instruments according to claim 1, wherein step S5 comprises:

10. The method for evaluating hard-tipped writing instruments according to claim 1, wherein the automatic Chinese character evaluation model in step S6 comprises a grade evaluation model obtained by using a threshold value classification method, a support vector machine and K nearest neighbors and a percentage evaluation model obtained by using linear regression, wherein the grade evaluation model is a three-grade evaluation model.