A calligraphy practice evaluation system and method based on template character multidimensional feature comparison

Abstract

This invention discloses a calligraphy practice evaluation system and method based on template character multidimensional feature comparison. The system comprises four parts: a copybook acquisition module, an image preprocessing module, a multidimensional feature extraction and comparison module, and a writing quality evaluation module. This user-friendly system offers advantages in convenience and objectivity. Users only need to practice writing Chinese characters based on standardized copybooks. Through this system, they can quickly obtain feedback on their writing, identify various problems in their handwritten characters, and practice calligraphy anytime, anywhere at low cost. The writing quality evaluation method based on template character multidimensional feature comparison proposed in this invention starts from key elements of character structure such as aspect ratio and center of gravity, and combines deep apparent features extracted by convolutional neural networks to provide specific and global indicator evaluations of the user's Chinese character writing. This facilitates users in improving their Chinese character writing quality and possesses interpretability and objectivity.

Description

Technical Field

[0001] This invention belongs to the fields of image analysis and computer vision, and relates to a Chinese character writing quality evaluation system and method, specifically a writing practice evaluation system and method based on template character multidimensional feature comparison. Background Technology

[0002] Writing Chinese characters in a standardized, neat, and legible manner helps to inherit the long-standing Chinese culture, improve one's language skills, and enhance one's cultural background. It also helps to accurately convey the meaning of the writer's message. Calligraphy practice generally begins with copying standard fonts. By comparing one's work with the standard font, one can identify and improve upon any shortcomings in the shape or structure of the characters. Therefore, systematic and standardized guidance is necessary to ensure significant progress while expending effort.

[0003] Instructing calligraphy practice through manual guidance is subjective, costly, lacks universality, and is inefficient, making it difficult to promote calligraphy practice effectively. Currently, calligraphy evaluation systems using computer vision processing have the following main drawbacks: (1) poor implementation, usually requiring specialized handwriting acquisition equipment, resulting in high implementation costs; (2) poor interpretability, typically using similarity as the evaluation indicator and judging the quality of Chinese character writing solely by providing scores, failing to comprehensively describe the problems and defects in writing Chinese characters. Summary of the Invention

[0004] To address the shortcomings of existing Chinese character evaluation systems, such as poor implementability and interpretability, this invention provides a handwriting evaluation system and method based on multi-dimensional feature comparison of template characters. This invention designs a convenient, operable, and objective handwriting evaluation system that can provide objective evaluation results of handwritten Chinese characters. It enables low-cost objective evaluation of users' handwritten Chinese characters, facilitating targeted improvement of handwriting quality.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A calligraphy practice evaluation system based on multidimensional feature comparison of template characters includes four parts: a model calligraphy acquisition module, an image preprocessing module, a multidimensional feature extraction and comparison module, and a handwriting quality evaluation module.

[0007] The calligraphy copybook acquisition module is responsible for acquiring standardized calligraphy copybook images containing template characters and user-written Chinese characters;

[0008] The image preprocessing module is responsible for extracting all template characters and user-written Chinese characters from the image, and matching the template characters and handwritten Chinese characters.

[0009] The multidimensional feature extraction and comparison module is responsible for extracting key features from template characters and handwritten Chinese characters, and comparing the features extracted from template characters and handwritten Chinese characters in combination with deep appearance features;

[0010] The writing quality evaluation module is responsible for analyzing the feature comparison results, providing specific index evaluations of written Chinese characters based on key elements and global index evaluations based on coefficient weighting, and providing optimization suggestions based on the evaluation results.

[0011] A method for evaluating handwriting practice using the aforementioned handwriting evaluation system includes the following steps:

[0012] Step 1: Collect standardized calligraphy template images containing template characters and user-written Chinese characters using the calligraphy template acquisition module;

[0013] Step 2: Extract all template characters and user-written Chinese characters from the image using the image preprocessing module, and match the template characters and handwritten Chinese characters;

[0014] Step 3: Extract key features in Chinese character evaluation, such as aspect ratio and center of gravity, from the template character and handwritten Chinese character using the multidimensional feature extraction and comparison module. Combine the deep appearance features extracted by the convolutional neural network to compare the features extracted from the template character and handwritten Chinese character.

[0015] Step 4: Analyze the feature comparison results through the writing quality evaluation module, provide specific index evaluations of written Chinese characters based on key elements and global index evaluations based on coefficient weighting, and provide optimization suggestions based on the evaluation results.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] (1) This invention proposes a calligraphy evaluation system based on multi-dimensional feature comparison of template characters. This system is user-friendly and has the advantages of convenience and objectivity. Users only need to practice writing Chinese characters based on standardized calligraphy templates. Through this system, they can quickly obtain feedback on their writing and identify various problems in handwritten Chinese characters, making it convenient for users to practice calligraphy anytime and anywhere at low cost.

[0018] (2) This invention proposes a writing quality evaluation method based on template character multidimensional feature comparison. Starting from the key elements of character structure such as aspect ratio and center position, this method combines the deep appearance features extracted by convolutional neural network to give users specific index evaluation and global index evaluation of Chinese character writing, which helps users improve the quality of Chinese character writing and has interpretability and objectivity. Attached Figure Description

[0019] Figure 1 A diagram showing the components of a calligraphy practice evaluation system based on multi-dimensional feature comparison of template characters;

[0020] Figure 2 This is a standardized calligraphy template example.

[0021] Figure 3 The system implements the optimal and worst Chinese character selection results.

[0022] Figure 4 To evaluate the specific results of handwritten Chinese characters, evaluation examples and... Figure 3 The same method is used. Detailed Implementation

[0023] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0024] A calligraphy evaluation system based on multi-dimensional feature comparison of template characters, such as Figure 1 As shown, the calligraphy practice evaluation system comprises four parts: a calligraphy copybook acquisition module, an image preprocessing module, a multi-dimensional feature extraction and comparison module, and a writing quality evaluation module.

[0025] The calligraphy copybook acquisition module is responsible for acquiring standardized calligraphy copybook images containing template characters and user-written Chinese characters;

[0026] The image preprocessing module is responsible for extracting all template characters and user-written Chinese characters from the image, and matching the template characters and handwritten Chinese characters.

[0027] The multidimensional feature extraction and comparison module is responsible for extracting key features in Chinese character evaluation, such as aspect ratio and center of gravity, from template characters and handwritten Chinese characters. It then compares the features extracted from template characters and handwritten Chinese characters with the deep appearance features extracted by the convolutional neural network.

[0028] The writing quality evaluation module is responsible for analyzing the feature comparison results, providing specific index evaluations of written Chinese characters based on key elements and global index evaluations based on coefficient weighting, and providing optimization suggestions based on the evaluation results.

[0029] This invention provides a method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters. The method analyzes the structure of written Chinese characters based on the user's copying results from standardized calligraphy models, combining key elements in Chinese character evaluation such as aspect ratio and center of gravity, with deep appearance features extracted by convolutional neural networks. It then provides evaluation and optimization suggestions for the user's handwriting, specifically including the following steps:

[0030] Step 1: Collect standardized calligraphy template images containing template characters and user-written Chinese characters using the calligraphy template acquisition module. This module can directly acquire images using the user's mobile phone, without the need for professional handwriting capture equipment. The specific requirements for the standardized calligraphy template are as follows:

[0031] Standardized calligraphy practice books are used for Chinese character writing practice. Each line consists of an equal number of identical, aligned square boxes. The first square in each line contains a template character for practice, and the remaining squares are blank square boxes for users to trace the template character. An example of a standardized calligraphy practice book is shown below. Figure 2 As shown.

[0032] Step 2: Extract all template characters and user-written Chinese characters from the image using the image preprocessing module, and then match the template characters and handwritten Chinese characters. The specific steps are as follows:

[0033] Step 2-1: Grayscale processing: Convert the collected standardized calligraphy template three-channel RGB image into a single-channel grayscale image.

[0034] Step 2-2: Template character and handwritten character extraction: After binarizing the grayscale image, the contour detection algorithm is used to obtain the contours of all Chinese characters in the image, and then the minimum bounding rectangle of all Chinese characters is extracted.

[0035] Steps 2-3: Matching Template Characters and Written Characters: If there are m template characters in the standardized template, and each template character requires n characters to be copied, then among all the smallest bounding rectangles, the first m characters with their center x-coordinates arranged from smallest to largest are the template characters. This process is used to extract all template character image slices. Let the center y-coordinate of the i-th template character be y. i The y-coordinate of the j-th handwritten Chinese character to be matched is y. j Calculate the vertical coordinate offset Δy ij =|y i -y j |,Δy ij After arranging the values ​​from smallest to largest, the first n numbers are the corresponding handwritten Chinese characters, and all handwritten character image slices are extracted in this way.

[0036] Step 3: Extract key features for Chinese character evaluation, such as aspect ratio and center of gravity, from the template character and handwritten Chinese characters using a multi-dimensional feature extraction and comparison module. Combine these with deep appearance features extracted by a convolutional neural network, and compare the features extracted from the template character and handwritten Chinese characters. The specific steps are as follows:

[0037] Step 3-1: Compare the similarity between the template character and the corresponding handwritten Chinese character to obtain a similarity score. Extract feature vectors from the template character and the handwritten Chinese character using feature extraction methods such as convolutional neural networks. Calculate the cosine of the angle between the feature vectors to measure the similarity between them. The result is used as the similarity score. The specific steps are as follows:

[0038] Step 3-1-1: For the i-th template character and the j-th matched handwritten Chinese character, extract the feature vectors l of the template character and the handwritten Chinese character respectively using a convolutional neural network. i and l j The convolutional neural network can be, but is not limited to, typical feature extraction networks such as ResNet and VGG. To reduce computational complexity, instead of using a convolutional neural network for feature extraction, the j-th handwritten Chinese character can be stretched to the same size as the i-th template character, and then the image can be flattened to obtain the feature vectors l of the template character and the handwritten Chinese character. i and l j .

[0039] Step 3-1-2: Calculate the eigenvector l i and l j The cosine of the included angle measures the similarity between feature vectors, yielding a similarity score. The closer the similarity is to 1, the higher the similarity. The formula is written as:

[0040]

[0041] Step 3-2: Compare the key elements of the template character with the corresponding handwritten Chinese character to obtain the key element score. The elements include, but are not limited to, key features related to the evaluation of the Chinese character shape and structure, such as aspect ratio and center of gravity. The specific steps are as follows:

[0042] Step 3-2-1: Compare the aspect ratios of the template character and the corresponding handwritten Chinese character to obtain the aspect ratio score. For the i-th template character and the j-th matched handwritten Chinese character, the ratio of the vertical axis to the horizontal axis is counted as the aspect ratio, denoted as hw respectively. i and hw j Calculate the relative aspect ratio difference Δhw ij The method is the ratio of the difference in aspect ratio between the template character and the handwritten Chinese character to the aspect ratio of the template character. The formula is written as:

[0043]

[0044] The difference is mapped to the range of 0 to 1 using the normal distribution function, and is used as the aspect ratio score. The closer the aspect ratio is to 1, the closer it is to 1. The formula can be written as:

[0045]

[0046] Step 3-2-2: Compare the center-of-gravity positions of the template character and the corresponding handwritten Chinese character to obtain the center-of-gravity position score. For the i-th template character and the j-th matched handwritten Chinese character, the center-of-gravity positions are (x...). i ,y i ) and (x j ,y j ), calculate the difference Δx between the centroid positions of the horizontal and vertical axes respectively. ij and Δy ij Then calculate the distance ΔD from the center of gravity. ij Formula writing:

[0047]

[0048] The difference is mapped to the range of 0 to 1 using the normal distribution function, and this is used as the centroid position score. The closer the value is to 1, the closer the center of gravity is to the position of the centroid. The formula can be written as:

[0049]

[0050] Step 4: Analyze the feature comparison results using the handwriting quality evaluation module, providing specific indicator evaluations based on key elements and global indicator evaluations based on coefficient weighting. Based on the evaluation results, provide optimization suggestions. The specific steps are as follows:

[0051] Step 4-1: Weight the similarity score and key element score calculated in Step 3 to obtain the overall score for each handwritten Chinese character. If there are b evaluation scores, the weight assigned to each score is α. k The formula for calculating the total score is:

[0052]

[0053] An "Excellent" grade (S) for Chinese character writing is given based on the overall score. ij ∈[0.9,1.0]), good (S) ij ∈[0.8,0.9)), Pass (S) ij ∈[0.6,0.8)), failing (S) ij The system evaluates characters within the range [0, 0.6) and sorts them by score, marking the best and worst written characters in each line. The system's implementation of selecting the best and worst written characters is shown below. Figure 3 As shown.

[0054] Step 4-2: Based on the specific comparison results calculated in Step 3-2, provide an evaluation of the key elements related to the character structure, using Δhw ij An evaluation can be given regarding whether the overall structure of the Chinese character is too thin, too wide, or moderate, based on Δx. ij and Δy ijIt can provide an evaluation of the center-of-gravity shift of Chinese characters. The evaluation criteria can be expanded according to actual application needs, including but not limited to aspect ratio and center-of-gravity position. The system's evaluation results for specific criteria are as follows: Figure 4 As shown, its evaluation use cases and Figure 3 The same method is used.

Claims

1. A method for evaluating handwriting practice based on multidimensional feature comparison of template characters, characterized in that... The method includes the following steps: Step 1: Collect standardized calligraphy template images containing template characters and user-written Chinese characters using the calligraphy template acquisition module; Step 2: Extract all template characters and user-written Chinese characters from the image using the image preprocessing module, and match the template characters and handwritten Chinese characters; Step 3: Extract key features for Chinese character evaluation, such as aspect ratio and center of gravity, from the template character and handwritten Chinese characters using the multi-dimensional feature extraction and comparison module. Combine this with the deep appearance features extracted by the convolutional neural network, and compare the features extracted from the template character and handwritten Chinese characters. The specific steps are as follows: Step 3-1: Compare the similarity between the template character and the corresponding handwritten Chinese character to obtain a similarity score. Extract the feature vectors of the template character and the handwritten Chinese character respectively using the feature extraction method. Calculate the cosine value of the angle between the feature vectors to measure the similarity between the feature vectors. The calculation result is used as the similarity score. Step 3-2: Compare the key elements of the template character with the corresponding handwritten Chinese character to obtain the key element score. The specific steps are as follows: Step 3-2-1: Compare the aspect ratio of the template character with the corresponding handwritten Chinese character to obtain the aspect ratio score; Step 3-2-2: Compare the center-of-gravity positions of the template character and the corresponding handwritten Chinese character to obtain the center-of-gravity position score. The specific steps are as follows: For the i-th template character and the j-th matched handwritten Chinese character, the centroid positions are respectively and Calculate the difference in centroid position between the horizontal and vertical axes respectively. and Then calculate the distance to the center of gravity. Formula writing: ； The difference is mapped to the range of 0 to 1 using the normal distribution function, and this is used as the centroid position score. , The closer the value is to 1, the closer the center of gravity is to the position of the centroid. The formula can be written as: ； Step 4: Analyze the feature comparison results through the writing quality evaluation module, provide specific index evaluations of written Chinese characters based on key elements and global index evaluations based on coefficient weighting, and provide optimization suggestions based on the evaluation results.

2. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 1, characterized in that... In step 1, the specific requirements for the standardized calligraphy template are as follows: Standardized calligraphy practice books allow users to practice writing Chinese characters. Each line consists of an equal number of identical, aligned square boxes. The first square in each line contains a template character for practice, while the rest of the line consists of blank square boxes where users can copy the template character.

3. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 1, characterized in that... The specific steps of step 2 are as follows: Step 2-1: Grayscale processing: Convert the collected standardized calligraphy template three-channel RGB image into a single-channel grayscale image; Step 2-2: Template character and handwritten character extraction: After binarizing the grayscale image, the contour detection algorithm is used to obtain the contours of all Chinese characters in the image, and then the minimum bounding rectangle of all Chinese characters is extracted; Steps 2-3: Matching Template Characters and Written Characters: If there are m template characters in the standardized template, and each template character requires n characters to be copied, then among all the smallest bounding rectangles, the first m characters with their center x-coordinates arranged from smallest to largest are the template characters. This process is used to extract all template character image slices. Let the center y-coordinate of the i-th template character be y. i The y-coordinate of the j-th handwritten Chinese character to be matched is y. j Calculate the vertical axis offset , After arranging the values ​​from smallest to largest, the first n numbers are the corresponding handwritten Chinese characters, and all handwritten character image slices are extracted in this way.

4. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 1, characterized in that... The specific steps of step 3-1 are as follows: Step 3-1-1: For the i-th template character and the j-th matched handwritten Chinese character, extract the feature vectors of the template character and the handwritten Chinese character respectively using a convolutional neural network. and ; Step 3-1-2: Calculate the feature vector and The cosine of the included angle measures the similarity between feature vectors, yielding a similarity score. , The closer the similarity is to 1, the higher the similarity. The formula is written as: 。 5. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 4, characterized in that... Step 3-1-1 is replaced by: for the i-th template character and the matched j-th handwritten Chinese character, directly stretch the j-th handwritten Chinese character to the same size as the i-th template character, then flatten the image to obtain the feature vectors of the template character and the handwritten Chinese character. and .

6. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 1, characterized in that... The specific steps of step 3-2-1 are as follows: For the i-th template character and the j-th matched handwritten Chinese character, the ratio of the vertical axis to the horizontal axis is called the aspect ratio, denoted as hw respectively. i and hw j Calculate the relative aspect ratio difference Δhw ij Formula writing: ； The difference is mapped to the range of 0 to 1 using the normal distribution function, and is used as the aspect ratio score. , The closer the aspect ratio is to 1, the closer it is to 1. The formula can be written as: 。 7. The method for evaluating handwriting practice based on multi-dimensional feature comparison of template characters according to claim 1, characterized in that... The specific steps of step 4 are as follows: Step 4-1: Weight the similarity score and key element score calculated in Step 3 to obtain the overall score for each handwritten Chinese character. If there are b evaluation scores, the weights assigned to each score are as follows: The formula for calculating the total score is: ； The overall score will be used to indicate the excellence level in Chinese character writing. ,good Pass failing The evaluation process also sorts and marks the best and worst written Chinese characters in each line based on their scores. Step 4-2: Evaluate the key elements of the character structure based on the comparison results.

8. A calligraphy evaluation system based on template character multidimensional feature comparison, wherein the method described in any one of claims 1-7 is characterized in that... The calligraphy practice evaluation system comprises four parts: a copybook acquisition module, an image preprocessing module, a multi-dimensional feature extraction and comparison module, and a writing quality evaluation module. The calligraphy copybook acquisition module is responsible for acquiring standardized calligraphy copybook images containing template characters and user-written Chinese characters; The image preprocessing module is responsible for extracting all template characters and user-written Chinese characters from the image, and matching the template characters and handwritten Chinese characters. The multidimensional feature extraction and comparison module is responsible for extracting key features from template characters and handwritten Chinese characters, and comparing the features extracted from template characters and handwritten Chinese characters in combination with deep appearance features; The writing quality evaluation module is responsible for analyzing the feature comparison results, providing specific index evaluations of written Chinese characters based on key elements and global index evaluations based on coefficient weighting, and providing optimization suggestions based on the evaluation results.

Images