Chinese character font generation method and terminal

Abstract

The invention discloses a Chinese character font generation method and a terminal, and the method comprises the steps: obtaining a font file, and extracting a Chinese character image set from the font file, the Chinese character image set comprising a first source font image set and a first target font image set; training a font generation model according to the first source font image set and the first target font image set, and inputting a second source font image set into the trained font generation model to obtain a second target font image set; and converting the second target font image set into a vector image set, and creating a new font file according to the vector image set. Through the trained font generation model, the second target font image set is generated according to the second source font image set, automatic image generation and image visual style conversion are realized, the generated bitmap image is converted into the vector image, the font is created according to the vector image, the font design efficiency is improved, and the font design efficiency is improved. And diversified design requirements and application scenes are met.

Description

A method and terminal for generating Chinese character fonts Technical Field

[0001] This invention relates to the field of computer application technology, and in particular to a method and terminal for generating Chinese character fonts. Background Technology

[0002] With the development of mobile internet, people's demand for personalized and customized fonts is constantly increasing. However, existing methods for creating Chinese font libraries mainly rely on human experience and design, with a low degree of automation. Designers need to design each Chinese character individually when creating a new font library, which is a time-consuming and expensive process for a large number of Chinese characters.

[0003] In recent years, deep learning has been widely applied in fields such as image generation, object detection, and natural language processing. In image generation, deep learning can combine an ordinary photograph with an artist's painting to generate a photograph with the style of the painting. However, compared to the 26 letters of the English alphabet, Chinese characters have 3755 commonly used characters, the GB2312 standard has 6763 simplified characters, and the CJK (Chinese-Japanese-Korean Unified Ideographs) has a staggering 20902 characters, many with complex strokes and structures. Therefore, research using deep learning for Chinese character font transfer is relatively limited. Currently, there are two main methods for Chinese character font transfer: 1. Stroke-based decomposition and reconstruction: This method decomposes the characters of the source and target fonts into components such as radicals and strokes, and then matches these components using a model. However, this method relies heavily on the quality of the decomposition results, making it difficult to properly decompose complex characters, while simple characters do not require decomposition. Furthermore, the decomposition-based method requires significant time for preprocessing and relies heavily on prior knowledge.

[0004] 2. Deep Learning-Based Approach: Most publicly available solutions currently use one-to-one pairing of source and target font image sets. Here, "one-to-one pairing" means that the content is the same but the fonts are different. Specifically, this approach requires that the source and target font image sets have an equal number of images containing the same characters. For Chinese characters, which have a large number of characters, this inevitably leads to a large training set size, and also requires that the target character be represented completely and accurately. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a Chinese character font generation method and terminal that can automatically generate and convert font styles, improve design efficiency, have no specific requirements on the number of target fonts and do not need to be matched with image content one by one, ensure that the font style conversion process is flexible and efficient, and help meet diverse design needs and application scenarios.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: a Chinese character font generation method, comprising the following steps: S1, obtaining a font file and extracting a set of Chinese character images from the font file, wherein the set of Chinese character images includes a first source font image set and a first target font image set; S2, training a font generation model based on the first source font image set and the first target font image set, and inputting a second source font image set into the trained font generation model to obtain a second target font image set; S3, converting the second target font image set into a vector graphic set, and creating a new font file based on the vector graphic set.

[0007] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: a Chinese character font generation terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the various steps of the above-mentioned Chinese character font generation method.

[0008] The beneficial effects of this invention are as follows: The Chinese character font generation method and terminal provided by this invention extracts a set of Chinese character images from a font file, including a first source font image set and a first target font image set, to provide training data for a subsequent font generation model. The font generation model is trained based on the first source font image set and the first target font image set. The trained font generation model has no specific requirement for the number of target fonts and does not need to be matched one-to-one with the image content, making the font generation process more flexible and adaptable to diverse design needs and application scenarios. The second source font image set is input into the trained font generation model to obtain the second target font image set, which is then converted into a vector atlas. A new font file is created based on the vector atlas, realizing the conversion of font style without matching samples, and converting the source font style into the target font style. This greatly simplifies the tedious manual operations in traditional font design methods and improves the efficiency and diversity of font design. Attached Figure Description

[0009] Figure 1 is a flowchart of a Chinese character font generation method according to an embodiment of the present invention; Figure 2 is a schematic diagram of a Chinese character font generation terminal according to an embodiment of the present invention; Figure 3 is a flowchart of the specific steps of a Chinese character font generation method according to an embodiment of the present invention; Figure 4 is a result display diagram of a Chinese character font generation method according to an embodiment of the present invention; Reference numerals: 1, a Chinese character font generation terminal; 2, memory; 3, processor. Detailed Implementation

[0010] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0011] Referring to Figure 1, this embodiment of the invention provides a method for generating Chinese character fonts, including the following steps: S1, obtaining a font file and extracting a set of Chinese character images from the font file, wherein the set of Chinese character images includes a first source font image set and a first target font image set; S2, training a font generation model based on the first source font image set and the first target font image set, and inputting a second source font image set into the trained font generation model to obtain a second target font image set; S3, converting the second target font image set into a vector graphic set, and creating a new font file based on the vector graphic set.

[0012] As can be seen from the above description, the beneficial effects of the present invention are as follows: by extracting a set of Chinese character images from a font file, including a first source font image set and a first target font image set, training data is provided for the subsequent font generation model. The font generation model is trained based on the first source font image set and the first target font image set. The trained font generation model has no specific requirements on the number of target fonts and does not need to be matched one-to-one with the image content, so that the font generation process is more flexible and can adapt to diverse design needs and application scenarios. The second source font image set is input into the trained font generation model to obtain the second target font image set, and the second target font image set is converted into a vector atlas. A new font file is created based on the vector atlas, realizing the conversion of font style without matching samples, and converting the source font style into the target font style. This greatly simplifies the tedious manual operations in traditional font design methods and improves the efficiency and diversity of font design.

[0013] Further, training the font generation model based on the first source font image set and the first target font image set includes: training a generator model and a discriminator model. The generator model includes a first generator model and a second generator model, and the discriminator model includes a first discriminator model and a second discriminator model. The generator model structure includes an upsampling region, a downsampling region, and a densely connected convolutional network. The discriminator model structure includes a convolutional layer, an instance normalization layer, and a leakage correction linear unit. The first generator model converts the first source font image set into a fake first target font image set. The first discriminator model compares the fake first target font image set with the first target font image set to obtain a first adversarial loss value. The second generator model converts the first target font image set into a fake first source font image set. The second discriminator model compares the fake first source font image set with the first source font image set to obtain a second adversarial loss value. The overall adversarial loss value is obtained by weighting the first adversarial loss value and the second adversarial loss value.

[0014] As described above, the generator model and discriminator model are trained through an adversarial training mechanism. The first generator model generates images that are infinitely close to the style of the first target font image set based on the first source font image set. The first discriminator model further distinguishes between generated images and real images to adversarially challenge the first generator model. The second generator model generates images that are infinitely close to the style of the first source font image set based on the first target font image set. The second discriminator model further distinguishes between generated images and real images to adversarial challenge the second generator model. This allows the generator model to learn the distribution of real data and generate diverse and creative samples. Furthermore, the overall adversarial loss value is obtained by weighting the first adversarial loss value and the second adversarial loss value, thereby improving the representativeness of the data and facilitating subsequent adjustment of training parameters based on the overall adversarial loss value.

[0015] Furthermore, the step of training the font generation model based on the first source font image set and the first target font image set further includes: adjusting the training parameters of the generator model and the discriminator model according to the overall adversarial loss value, repeatedly training the generator model and the discriminator model, and stopping the training of the generator model and the discriminator model when the number of repeated training rounds reaches a preset value.

[0016] As described above, by adjusting the training parameters of the generator and discriminator models through the overall adversarial loss value, and repeatedly training the generator and discriminator models, the training effect of the models is optimized to reach a balance state. The generator model generates a set of images that are realistic enough to be mistaken for real images, while the discriminator model can hardly distinguish between real and fake images. When the number of repeated training rounds reaches a preset value, the training of the generator and discriminator models is stopped, and the stability of the training process is improved by dynamically adjusting the training parameters.

[0017] Furthermore, the step of inputting the second source font image set into the trained font generation model to obtain the second target font image set includes: when the second source font image set is the same as the first source font image set, it is determined that the visual style of the second target font image set is similar to the visual style of the first target font image set; when the second source font image set is different from the first source font image set, it is determined that the second target font image set has a new visual style.

[0018] As described above, the font generation model can not only generate visual styles corresponding to the training data, but also transform existing visual styles into new visual styles, thereby generating new styles of Chinese fonts, improving the diversity of font design. Furthermore, by using the font generation model to generate automatically, it reduces the reliance on professional designers, lowers costs, and makes font design more economical and efficient.

[0019] Further, step S3 specifically involves: using a vector graphics conversion tool to convert the second target font image set into a vector graphic set, and modifying the vector graphic set, including standardizing the view frame size and offset matrix of the vector graphic set; using a script file to batch import the modified vector graphic set into font editing software to create new font files.

[0020] As described above, by converting bitmap images into vector images and standardizing the viewport size and offset matrix of vector atlases, different application design needs can be met, achieving better visual effects and improving the flexibility of font design. Script files can be used to batch import modified vector atlases into font editing software to create new fonts. Multiple vector atlases can be processed simultaneously, reducing repetitive work and increasing font generation speed.

[0021] Referring to Figure 2, another embodiment of the present invention provides a Chinese character font generation terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the above-described Chinese character font generation method.

[0022] The Chinese character font generation method and terminal of the present invention are applicable to automatically generating fonts and converting font styles, thereby improving font design efficiency. The following is a description of specific implementation methods: Please refer to Figures 1 and 3. Embodiment 1 of the present invention is: a Chinese character font generation method, including the following steps: S1, obtaining a font file and extracting a set of Chinese character images from the font file, wherein the set of Chinese character images includes a first source font image set and a first target font image set.

[0023] In this embodiment, by extracting a first source font image set and a first target font image set from the font file, such as source font image A and target font image B, the reliability of the data source is ensured, thereby providing training data for the subsequent font generation model. Furthermore, the font file is parsed into an image format to facilitate the training and processing of the subsequent font generation model.

[0024] S2. Train a font generation model based on the first source font image set and the first target font image set, and input the second source font image set into the trained font generation model to obtain the second target font image set.

[0025] In this embodiment, a font generation model is trained based on a first source font image set and a first target font image set. The trained font generation model has no specific requirements on the number of target fonts and does not need to be matched one-to-one with the image content, so that the font generation process is more flexible and can adapt to diverse design needs and application scenarios. The second source font image set is input into the trained font generation model to obtain the second target font image set, which provides image data for the subsequent generation of new fonts. For example, the second source font image set is the 6763 Chinese character images included in the GB2312 standard in the original font C, and the second target font image is the new font D.

[0026] Further, in this embodiment, training the font generation model based on the first source font image set and the first target font image set includes: training a generator model and a discriminator model. The generator model includes a first generator model and a second generator model, and the discriminator model includes a first discriminator model and a second discriminator model. The first generator model converts the first source font image into a fake first target font image set. The first discriminator model compares the fake first target font image set with the first target font image set to obtain a first adversarial loss value. Specifically, the first generator model converts the source font image A into a fake target font image. The first discriminator compares the fake target font image B' with the target font image B and calculates the first adversarial loss value using a loss function, wherein the loss function is preferably the MSELoss function. The second generator model converts the first target font image set into a fake first source font image set. The second discriminator model compares the fake first source font image set with the first source font image set to obtain a second adversarial loss value. Specifically, the second generator model converts the target font image B into a fake first source font image set. A fake source font image is used. A second discriminator compares the fake source font image A' with the original font image A, and calculates a second adversarial loss value using a loss function, preferably the MSELoss function. The generator and discriminator models are trained through an adversarial training mechanism so that the generator model can learn the distribution of real data and generate diverse and creative samples. The overall adversarial loss value is obtained by weighting the first and second adversarial loss values ​​to improve data representativeness and facilitate subsequent adjustment of training parameters based on the overall adversarial loss value. The generator's overall loss is a weighted loss of the adversarial network loss and the cycle consistency loss. The adversarial network loss is preferably calculated using the MSELoos function to calculate the binary classification loss values ​​of the two discriminators, i.e., the first and second adversarial loss values. The cycle consistency loss is calculated using the L1Loss function to calculate the difference loss value between the output images of the two generators and the target image. The discriminator's overall loss is a weighted loss of the real image loss and the fake image loss, i.e., the weighted value of the first and second adversarial loss values. The discriminator is applied to both the real and generated images, and the loss value is calculated using the MSELoss function.

[0027] Furthermore, the generator model structure includes an upsampling region, a downsampling region, and a densely connected convolutional network. In this embodiment, the generator model is preferably as follows: the upsampling region includes two sets of deconvolutional layers, an instance normalization layer, a rectified linear unit (ReLU activation function), a reflection padding layer, a convolutional layer, and a hyperbolic tangent activation function (Tanh activation function); the downsampling region includes a reflection padding layer and three sets of convolutional layers, an instance normalization layer, and a rectified linear unit (ReLU activation function); the densely connected convolutional network includes four residual dense blocks, each containing a custom number of feature transfer layers, and each feature transfer layer includes a batch normalization layer and a rectified linear unit (ReLU activation function). The reason for choosing a generator containing a densely connected convolutional network in this embodiment is that each layer of the densely connected convolutional network can access the features of all previous layers, which can more effectively utilize feature information, improve feature propagation and reuse, enable the network to converge faster, and alleviate the gradient vanishing problem; at the same time, because the feature reuse rate is high, the function of each layer is more focused, reducing redundant feature learning.

[0028] The discriminator model includes convolutional layers, instance normalization layers, and leakage correction linear units. In this embodiment, the discriminator model preferably includes 6 sets of convolutional layers, instance normalization layers, and leakage ReLU activation functions. Finally, a single channel is output through a convolutional layer, and the feature map size is reduced to 1 after average pooling operation. A single evaluation value is then constructed for binary classification tasks.

[0029] Furthermore, in this embodiment, training the font generation model based on the first source font image set and the first target font image set further includes: adjusting the training parameters of the generator model and the discriminator model according to the overall adversarial loss value, repeatedly training the generator model and the discriminator model, and stopping the training of the generator model and the discriminator model when the number of repeated training rounds reaches a preset value. The training parameters of the generator model and the discriminator model are adjusted by the overall adversarial loss value, and the generator model and the discriminator model are repeatedly trained to optimize the model training effect and achieve a balanced state. The generator model generates a sufficiently realistic image set, and the discriminator model can hardly distinguish between real and fake images. When the number of model training rounds reaches the preset value, the training of the generator model and the discriminator model is stopped. Furthermore, the stability of the training process is improved by dynamically adjusting the training parameters. Preferably, the stopping condition for model training is set as follows: a preset value is set for the number of model training rounds, and training is stopped when the preset value is reached, rather than setting the stopping condition based on the overall adversarial loss value, because the overall adversarial loss value of the model fluctuates within a certain range and cannot be used as a good judgment condition.

[0030] Further, please refer to Figure 4, which is a result display diagram of a Chinese character font generation method according to an embodiment of the present invention. In this embodiment, the step of inputting the second source font image set into the trained font generation model to obtain the second target font image set includes: when the second source font image set is the same as the first source font image set, it is determined that the visual style of the second target font image set is similar to the visual style of the first target font image set. For example, if the original font C is the same as the first source font image set A, then the generated new font D is similar to the visual style of the first target font image set B; when the second source font image set is different from the first source font image set, it is determined that the second target font image set has a new visual style. For example, if the original font C is different from the first source font image set A, then the generated new font D has a unique visual style, and the character structure of the new font D is similar to that of the original font C. The font generation model can not only generate the visual style corresponding to the training data, but also convert the existing visual style into a new visual style, thereby generating Chinese character fonts with new styles, improving the diversity of font design, and using the font generation model to automatically generate, reducing the dependence on professional designers, reducing costs, and making font design more economical and efficient.

[0031] S3. Convert the second target font image set into a vector graphic set, and create a new font file based on the vector graphic set.

[0032] In this embodiment, by converting the second target font image set into a vector graphic set and creating a new font file based on the vector graphic set, the tedious manual operations in traditional font design methods are greatly simplified, improving the efficiency and diversity of font design.

[0033] Furthermore, in this embodiment, step S3 specifically involves: using a vector graphics conversion tool to convert the second target font image set into a vector graphic set, and modifying the vector graphic set, including standardizing the view frame size and offset matrix of the vector graphic set to meet different application design requirements, achieve better visual effects, and improve the flexibility of font design; using a script file to batch import the modified vector graphic set into the font editing software to create new fonts, which can process multiple vector graphic sets simultaneously, reduce repetitive work, and improve the font generation speed.

[0034] Referring to Figure 2, Embodiment 2 of the present invention is: a Chinese character font generation terminal 1, including a memory 2, a processor 3, and a computer program stored on the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, it implements the various steps of the Chinese character font generation method of Embodiment 1.

[0035] In summary, the present invention provides a Chinese character font generation method and terminal. By acquiring a font file and extracting a first source font image set and a first target font image set from the font file, the reliability of the data source is ensured, providing training data for the subsequent font generation model. The font generation model is trained based on the first source font image set and the first target font image set, including training a generator model and a discriminator model. The generator model includes a first generator model and a second generator model, and the discriminator model includes a first discriminator model and a second discriminator model. The first generator model converts the first source font image set into a fake first target font image set. The first discriminator model compares the fake first target font image set with the first target font image set to obtain a first adversarial loss value. The second generator model converts the first target font image set into a fake first source font image set. The second discriminator model compares the fake first source font image set with the first source font image set to obtain a second adversarial loss value. The generator model and discriminator model are trained through an adversarial training mechanism, enabling the generator model to learn the distribution of real data and generate diverse and creative samples. The first adversarial loss value and the second adversarial loss value are weighted to obtain the overall adversarial loss value, which improves the representativeness of the data and facilitates the subsequent adjustment of training parameters based on the overall adversarial loss value. The training parameters of the generator model and the discriminator model are adjusted according to the overall adversarial loss value, and the generator model and the discriminator model are repeatedly trained to optimize the model training effect and reach a balance state. The generator model generates a set of images that are realistic enough to be mistaken for real images, and the discriminator model can hardly distinguish between real and fake images. When the overall adversarial loss value meets the preset value, the training of the generator model and the discriminator model is stopped. The second source font image set is input into the trained font generation model to obtain the second target font image set. When the second source font image set is the same as the first source font image set, it is determined that the visual style of the second target font image set is similar to that of the first target font image set. When the second source font image set is different from the first source font image set, it is determined that the second target font image set has a new visual style. The font generation model can not only generate the visual style corresponding to the training data, but also convert the existing visual style into a new visual style, thereby generating Chinese fonts with new styles and improving the diversity of font design.The second target font image set is converted into a vector graphic set using a vector graphics conversion tool. The vector graphic set is then modified, including standardizing the viewport size and offset matrix to meet different application design requirements, achieve better visual effects, and improve the flexibility of font design. A script file is used to batch import the modified vector graphic sets into font editing software to create new fonts. This allows for the simultaneous processing of multiple vector graphic sets, reducing repetitive work and increasing font generation speed.

[0036] Therefore, the embodiments of the present invention can achieve the following beneficial effects: Automating the generation of target images using a font generation model and converting the generated target images into vector images to create new fonts significantly shortens the font design cycle and improves font design efficiency; the font generation model in the present invention can not only convert existing font styles into various novel styles, but also generate entirely new Chinese character fonts, increasing the diversity of font design; the automated generation technology using the font generation model reduces reliance on professional design, lowers labor costs, and makes font design more economical and efficient; using a trained font generation model to generate new fonts does not require a specific number of source and target font images, nor does it require a one-to-one correspondence of content. Users can design a small number of character images as the target font image set during the training phase. After training a model, the model can be used to generate all other characters of the same style, improving the flexibility of font generation and thus adapting to diverse design needs and application scenarios.

[0037] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for generating Chinese character fonts, characterized in that, The steps include: S1, obtaining a font file and extracting a set of Chinese character images from the font file, wherein the set of Chinese character images includes a first source font image set and a first target font image set; S2. Train a font generation model based on the first source font image set and the first target font image set, and input the second source font image set into the trained font generation model to obtain the second target font image set; S3. Convert the second target font image set into a vector graphic set, and create a new font file based on the vector graphic set.

2. The method for generating Chinese character fonts according to claim 1, characterized in that, The step of training a font generation model based on the first source font image set and the first target font image set includes: training a generator model and a discriminator model. The generator model includes a first generator model and a second generator model, and the discriminator model includes a first discriminator model and a second discriminator model. The generator model structure includes an upsampling region, a downsampling region, and a densely connected convolutional network. The discriminator model structure includes a convolutional layer, an instance normalization layer, and a leakage correction linear unit. The first generator model converts the first source font image set into a fake first target font image set. The first discriminator model compares the fake first target font image set with the first target font image set to obtain a first adversarial loss value. The second generator model converts the first target font image set into a fake first source font image set. The second discriminator model compares the fake first source font image set with the first source font image set to obtain a second adversarial loss value. The overall adversarial loss value is obtained by weighting the first adversarial loss value and the second adversarial loss value.

3. The method for generating Chinese character fonts according to claim 2, characterized in that, The step of training the font generation model based on the first source font image set and the first target font image set further includes: adjusting the training parameters of the generator model and the discriminator model according to the overall adversarial loss value, repeatedly training the generator model and the discriminator model, and stopping the training of the generator model and the discriminator model when the number of repeated training rounds reaches a preset value.

4. The method for generating Chinese character fonts according to claim 1, characterized in that, The step of inputting the second source font image set into the trained font generation model to obtain the second target font image set includes: when the second source font image set is the same as the first source font image set, it is determined that the visual style of the second target font image set is similar to the visual style of the first target font image set; when the second source font image set is different from the first source font image set, it is determined that the second target font image set has a new visual style.

5. The method for generating Chinese character fonts according to claim 1, characterized in that, Step S3 specifically involves: using a vector graphics conversion tool to convert the second target font image set into a vector graphic set, and modifying the vector graphic set, including standardizing the view frame size and offset matrix of the vector graphic set; Use a script file to batch import modified vector graphics sets into font editing software and create new font files.

6. A Chinese character font generation terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it performs the following steps: S1, obtains a font file and extracts a set of Chinese character images from the font file, wherein the set of Chinese character images includes a first source font image set and a first target font image set; S2. Train a font generation model based on the first source font image set and the first target font image set, and input the second source font image set into the trained font generation model to obtain the second target font image set; S3. Convert the second target font image set into a vector image set, and create a new font file based on the vector image.

7. A Chinese character font generation terminal according to claim 6, characterized in that, The step of training a font generation model based on the first source font image set and the first target font image set includes: training a generator model and a discriminator model. The generator model includes a first generator model and a second generator model, and the discriminator model includes a first discriminator model and a second discriminator model. The generator model structure includes an upsampling region, a downsampling region, and a densely connected convolutional network. The discriminator model structure includes a convolutional layer, an instance normalization layer, and a leakage correction linear unit. The first generator model converts the first source font image set into a fake first target font image set. The first discriminator model compares the fake first target font image set with the first target font image set to obtain a first adversarial loss value. The second generator model converts the first target font image set into a fake first source font image set. The second discriminator model compares the fake first source font image set with the first source font image set to obtain a second adversarial loss value. The overall adversarial loss value is obtained by weighting the first adversarial loss value and the second adversarial loss value.

8. A Chinese character font generation terminal according to claim 7, characterized in that, The step of training the font generation model based on the first source font image set and the first target font image set further includes: adjusting the training parameters of the generator model and the discriminator model according to the overall adversarial loss value, repeatedly training the generator model and the discriminator model, and stopping the training of the generator model and the discriminator model when the number of repeated training rounds reaches a preset value.

9. A Chinese character font generation terminal according to claim 6, characterized in that, The step of inputting the second source font image set into the trained font generation model to obtain the second target font image set includes: when the second source font image set is the same as the first source font image set, it is determined that the visual style of the second target font image set is similar to the visual style of the first target font image set; when the second source font image set is different from the first source font image set, it is determined that the second target font image set has a new visual style.

10. A Chinese character font generation terminal according to claim 6, characterized in that, Step S3 specifically involves: using a vector graphics conversion tool to convert the second target font image set into a vector graphic set, and modifying the vector graphic set, including standardizing the view frame size and offset matrix of the vector graphic set; Use a script file to batch import modified vector graphics sets into font editing software and create new font files.

Images