The role of deep neural network in the creation of traditional Chinese landscape painting

Abstract

With the wide application of deep learning technology in various fields, its potential in artistic creation has gradually attracted attention. This research focuses on the application of deep learning in the creation of traditional Chinese landscape painting and its cultural and aesthetic impact. First, the research comprehensively analyzes the existing deep learning algorithms and the basic elements of Chinese landscape painting to determine the most suitable model architecture. Then, through several rounds of experiments, various training parameters are adjusted and the optimal network configuration is determined. In terms of assessment, the study uses a variety of indicators, including visual quality and technical performance, as well as in-depth cultural and aesthetic analysis. The results show that deep learning not only effectively improves visual quality and technical performance, but also has a positive impact on culture and aesthetics. Although there are some limitations, such as high computational requirements and reliance on large amounts of training data, corresponding solutions are also proposed. This study provides a powerful experimental basis for the integration of Chinese traditional art and modern science and technology, and promotes the research in this field.

Keywords

Deep learning Chinese landscape painting visual quality technical performance cultural and aesthetic analysis

1. Introduction

At the intersection of contemporary technology and art, deep neural networks (DNNS) have become central to innovation, particularly in visual arts such as image generation and style transfer. At the same time, as an art form bearing rich historical and cultural significance, Chinese traditional landscape painting has been widely concerned and studied around the world. However, the intersection and interaction between two seemingly distinct fields, deep neural networks and traditional Chinese landscape painting, has not been fully explored.

Recent studies, such as Chung and Huang’s [1], have used edge-enhanced generative adversarial networks to transform Chinese ink paintings into realistic images, highlighting the potential of deep learning in traditional art. In the study of Jin [2], the application of genetic algorithm-based machine vision in image painting style methods and image processing optimization provides a new technical approach for the modern expression of traditional art. In addition, Chen explored the creation of Chinese ink-style painting art based on deep learning framework and convolutional neural network model [3], which demonstrated the great potential of deep learning in understanding and innovating traditional art forms.

As a highly abstract and symbolic art form, traditional Chinese landscape painting not only reproduces natural landscapes, but also deeply expresses emotions and philosophy. Facing the challenge of modern science and technology, its future development path and possibility are full of unknown. The purpose of this study is to explore the application of deep learning technology in the creation of traditional Chinese landscape painting, in order to fill the gap of existing research. Through case studies, for example, Guo et al. demonstrated how to enable robots to actively learn and paint through visual measurement and reproduction of the artistic creation process [4]. This study will explore how deep neural networks can help artists achieve style innovation and diversity of expression in landscape painting creation.

This research not only has high academic value, but also promotes the practical application and development of related fields. Overall, the goal of this study is to provide a solid foundation for understanding and utilizing the potential of deep learning in the field of traditional art, while providing new insights into the fusion of traditional art and modern technology.

Research exploring the role of deep neural networks in the creation of traditional Chinese landscape painting addresses the fusion of art and technology, and how emerging technologies can be used to interpret, preserve and even redefine traditional art forms. Here are some of the key studies and perspectives in related areas.

The application of deep neural networks in artistic creation is increasingly extensive, and its impact on visual art is particularly significant. The study of Zhou et al. [5] demonstrated the rapid generation method of 3D scenes in Chinese landscape painting, indicating that deep learning technology can add new dimensions to traditional art forms. Meanwhile, Qiu and Zhang [6] discussed the application of digital technology in painting, highlighting the importance of big data and new media technology in modern art creation. In the study of the combination of traditional landscape painting and modern technology, the preservation and innovation of traditional elements have become an important topic. Li [7] provided a new perspective to understand the transformation of traditional art in the digital age by studying 3D painting based on cloud computing. In addition, Liu et al. [8] revealed the possibility of using generative translation models for painting completion, demonstrating the innovative application of deep learning in artistic creation. Notably, the use of deep neural networks in artistic creation has also led to discussions about originality and cultural inheritance. The research of Yang et al. [9] showed that immersive virtual reality technology has a significant impact on art education, which emphasizes the important role of technology in inheriting and educating traditional art. Similarly, Xu and Cai [10] evaluate art design based on artificial intelligence and machine learning, highlighting the potential applications of technology in art evaluation and creation.

To sum up, the role of deep neural networks in the creation of traditional Chinese landscape painting is not only reflected in technical innovation, but also related to the reinterpretation of traditional art and the inheritance of cultural values. These studies provide a rich theoretical foundation and practical cases, and provide an important perspective for in-depth discussion of the potential and challenges of deep learning in the field of traditional arts.

The main purpose of this study is to deeply explore the practical application and potential impact of deep neural networks in the creation of traditional Chinese landscape painting. Specifically, the first hope is to understand how to use deep neural network technology to achieve efficiency improvement and artistic innovation in the creation process of landscape painting. At the same time, the research also focuses on the cultural and aesthetic challenges that this technology may face in its application. In order to achieve these goals, this study will focus on answering two core questions: First, what are the specific application scenarios of deep neural networks in landscape painting creation, and what are the advantages and limitations compared with traditional methods? Second, how to evaluate the effectiveness and cultural influence of deep neural network in landscape painting creation from both quantitative and qualitative aspects? By studying these key issues, we hope to not only provide strong theoretical support for the combination of traditional art and modern science and technology, but also open up new research paths for the application of deep learning technology in more specific cultural and artistic fields.

The purpose of this study is to comprehensively explore the application and influence of deep learning technology in the creation of traditional Chinese landscape painting. Specifically, the research first makes a comprehensive analysis of the existing deep learning algorithms and the basic elements of Chinese landscape painting in order to build a suitable model architecture. In the stage of model training, the research not only adjusts various training parameters, but also determines the optimal network configuration through multiple rounds of experiments. In terms of evaluation, this study uses a variety of indicators to quantify the visual quality and technical performance of the model. Visual quality is mainly assessed by factors such as the fineness of the picture, color vividness and artistic style. The technical performance includes the computational efficiency, accuracy and generalization ability of the model. At the same time, in-depth analysis at the cultural and aesthetic level is also carried out to understand the adaptability and potential value of deep learning in traditional arts. In the results obtained, the study found that deep learning technology can not only effectively improve the visual quality and technical performance of Chinese landscape painting, but also have a positive impact on the cultural and aesthetic aspects. However, these applications have also exposed some limitations, such as the model’s high computational requirements and reliance on large amounts of training data. In view of these problems, the corresponding solutions and improvement suggestions are put forward. In general, this study not only provides a powerful experimental basis for the application of deep learning in artistic creation, but also promotes the research on the integration and development of traditional Chinese art and modern science and technology. It is believed that with the continuous progress of technology and in-depth research, deep learning will play an important role in more art fields.

2. Cross application: Deep neural network and Chinese landscape painting

2.1 Current situation of contemporary landscape painting creation

In Chinese painting, ink painting as a unique form of expression, through the use of different shades of ink to form a unique “ink rhyme”. Its basic elements include simplicity, symbolism and naturalness. The beauty is mainly embodied in the beauty of image, the beauty of pen and ink, the beauty of artistic conception and the beauty of visual psychology. In the integration of art and science and technology, traditional Chinese landscape painting is facing the intersection of history and future, and encountering the dual influence of challenges and opportunities. The wave of globalization and digitalization is changing the face of culture and art, and landscape painting is no exception. This art form, deeply rooted in Chinese cultural traditions, is facing the challenge of a modern and multicultural environment.

Examples include collaborative projects between artists and scientists, such as the use of deep neural networks to reinterpret and create landscape paintings. For example, in one project, artists used AI algorithms to analyze thousands of historical landscape paintings, extract their style and technical characteristics, and then incorporate these elements into new creations [11]. This shows not only the traditional elements of landscape painting, but also how these elements can be combined with contemporary art forms to create entirely new visual experiences.

At the same time, the unique charm and cultural heritage of landscape painting also attract artists and researchers seeking innovation. They are not only committed to studying and inheriting this art form, but also to exploring new stages of its development through new creative methods and means. As a cutting-edge technology, deep neural network opens up new possibilities in landscape painting creation and research [12].

The high computational power and learning efficiency of deep neural networks allow artists and researchers to quickly extract and learn the basic elements and composition rules in a large number of historical and contemporary landscape painting samples. This not only improves the efficiency and quality of creation, but also provides technical support for the contemporary transformation of landscape painting [13]. Through machine learning algorithms, basic elements of traditional landscape painting can be accurately imitated and innovated.

It is worth noting that deep neural networks are not only a tool, but also a new perspective that can expand the boundaries of artistic creation and aesthetic experience. The application of the algorithm not only retains the traditional aesthetic and cultural elements, but also introduces a new social and cultural context, giving landscape painting more extensive and multi-dimensional significance. This is also in line with the views of many researchers and artists, who believe that the combination of art and technology contributes to the inheritance and innovation of traditional culture.

At the same time, the unique charm and cultural heritage of landscape painting also attract artists and researchers seeking innovation. Studies have shown the impact of immersive virtual reality technology in art education, providing insights on how modern technology can be used to enhance art experience and learning [14]. In addition, some studies have discussed the application of computer image technology in 3D painting based on cloud computing, providing technical support for the modernization of landscape painting [15].

To sum up, the status and value of landscape painting in contemporary society is a complex problem composed of many factors, and the intervention of deep neural network provides a new solution and thinking Angle for this problem. This is not only a successful fusion of art and technology, but also a new attempt to find a sustainable and creative balance between tradition and modernity, culture and technology. Therefore, for anyone who cares about the future direction of traditional culture, this is an important topic worthy of attention and research.

2.2 The role of deep neural networks in art

In recent years, the fusion trend of art and technology has become increasingly obvious, and the potential of deep neural network (DNN) in art creation is particularly eye-catching [16]. This advanced computational model not only efficiently processes and analyzes large data sets, but also generates artworks with specific styles or attributes through training and optimization. The multifaceted nature of deep neural networks makes them a transformative tool in multiple art domains, where the extraction of image features is a key step in the entire process.

In the visual arts, algorithms such as generative Adversarial Networks (Gans) have been successfully applied to a variety of art styles, from realistic portraiture to abstract art to complex multimedia works. These algorithms are not only able to mimic existing artistic styles, but also create entirely new forms of visual expression through techniques such as “style transfer.”

Especially in the creation of traditional Chinese landscape painting, deep neural networks have begun to show their unique role and innovative potential [17]. For example, by analyzing a large number of historical landscape paintings, deep learning models can learn and simulate the styles and techniques of traditional landscape paintings to generate new works with traditional characteristics and a modern sense. This technique can not only imitate the colors, strokes and textures of traditional landscape painting, but also innovate and change on the basis of retaining traditional elements.

Applications of deep neural networks also extend to other art forms such as music, literature, and the performing arts. In music, recurrent neural networks (RNNS) are used to generate melodies and harmonies, simulating complex musical structures and styles. In literary creation, natural language processing (NLP) models are used to create poetry, prose and even plays.

Deep neural network-based painter style learning breaks down images through convolutional neural networks (CNN) and learns through deep learning algorithms to find the result of the target image [18]. This method is particularly suitable for analyzing and learning the painting styles of artists who excel in color, brush strokes, and texture.

However, the application of deep neural networks also raises questions about creative originality, cultural inheritance, and the relationship between art and technology. One of the most interesting questions is how the “originality” of a work can be determined when the creative process is increasingly dependent on algorithms and data. How can we ensure that technological applications do not erode or dilute the uniqueness and value of particular cultural and art forms?

Especially in the creation of traditional Chinese landscape painting, deep neural networks began to show their unique role and innovative potential. For example, by analyzing a large number of historical landscape painting works, some researchers use deep learning models to learn and simulate the styles and techniques of traditional landscape painting, thus generating new works with traditional characteristics and modern sense. This technique can not only imitate the colors, strokes and textures of traditional landscape painting, but also innovate and change on the basis of retaining traditional elements [19].

These questions challenge traditional categories of artistic creation and evaluation, as well as provoke broader academic and ethical discussions. For example, deep neural networks have the potential to redefine the roles and relationships of “artist” and “viewer” in the future, and may even affect the functioning of the art market.

To sum up, deep neural networks have become an innovative tool and method in the field of modern art. It not only expands the boundaries of artistic creation, but also challenges and reshapes the understanding and perception of the relationship between art, culture and technology. This provides a rich theoretical support and practical basis for in-depth research on the application of deep neural networks in specific cultural and artistic backgrounds such as traditional Chinese landscape painting. Therefore, this field deserves close attention not only from artists and researchers of science and technology, but also from all those interested in contemporary cultural and social developments.

2.3 Intersection and challenge of landscape painting and deep neural network

In the intersection of traditional Chinese landscape painting and modern deep neural network technology, a diversified and complicated phenomenon has been witnessed. With rapid advances in the fields of artificial intelligence and art, this interdisciplinary discipline is attracting more and more artists and researchers. They attempt to interpret, simulate and even innovate the basic elements and structures of landscape painting through deep learning techniques. These efforts are not only expected to improve the efficiency of the creation of artistic works, but may also lead to new aesthetic experiences and cultural connotations.

However, this innovation path also faces multiple technical, cultural and ethical challenges:

Technical Challenge: Accurately capturing and simulating the unique artistic elements and aesthetic laws of landscape painting is an extremely challenging task. The subtle artistic details such as brush strokes, composition and ink color of landscape painting are difficult to reproduce completely by algorithms. Solutions include continuing to optimize deep learning models, particularly in terms of image processing and generation algorithms, as well as increasing training data on specific properties of art works.

Cultural challenge: Ensure that landscape paintings generated by deep neural networks truly convey their rich cultural and historical connotations. Solutions to this challenge include increasing the model’s understanding of the Chinese cultural context and history of landscape painting, as well as collaborating with art history experts to ensure the cultural accuracy and depth of the generated works.

Ethical and legal challenges: Controversies over copyright, originality, and cultural identity of model-generated landscape paintings [20]. Addressing these challenges requires the development of relevant legal and ethical guidelines that clarify copyright attribution and balance the role of AI and the contribution of artists in the creation of art.

Future research directions should include:

Develop more advanced deep learning techniques, especially in image recognition and generation, to better capture and represent the unique style of landscape painting.

Deepen the understanding of traditional Chinese art and culture, and integrate these elements more deeply into technological developments to ensure the cultural authenticity and depth of the generated works.

Explore models of interdisciplinary collaboration, particularly between art history, cultural studies and technology development, to promote more comprehensive and in-depth understanding and innovation.

In conclusion, the intersection of traditional Chinese landscape painting and deep neural networks not only demonstrates the great potential of cooperation between art and technology, but also faces multiple practical challenges. Overcoming these challenges requires interdisciplinary collaboration, in-depth research, and a careful examination of multi-dimensional interrelationships and implications. The study of the future is not only a combination of technology and art, but also a comprehensive social, cultural and ethical issue that deserves in-depth study and sustained attention. Only by comprehensively and accurately understanding these multi-dimensional challenges and opportunities can we better grasp the current situation and future development trend of this cross-cutting field, and open up new possibilities for future artistic creation and cultural inheritance. Therefore, the exploration of this field is not only a technical or artistic issue, but also a comprehensive social, cultural and ethical issue that deserves in-depth study and continuous attention.

In future research, more attention should be paid to how to balance the relationship between modern technology and traditional art, so as to ensure that the progress of technology will not destroy or distort the core values and cultural spirit of traditional art. At the same time, explore how to use modern technologies such as deep neural networks to inject new creativity and expression on the basis of maintaining the original appearance of traditional art, so as to bring new vitality and perspective to the inheritance and development of traditional art. This process requires not only technological innovation, but also the deep involvement and cooperation of experts in the fields of culture and art to ensure that the development of science and technology can promote rather than destroy the rich and profound heritage of traditional culture.

3. Experimental design and method

3.1 Data set preparation

3.1.1 Source and type of data

In this study, data set preparation is a crucial step, as it will directly affect the subsequent training and evaluation of deep neural network models. The goal of this study is to analyze, interpret and generate works conforming to the aesthetic and technical skills of traditional Chinese landscape painting through deep neural networks. Therefore, the construction of data sets requires special attention. The details of the data set are shown in Table 1.

Table 1
Data sources and types

Data type	Data source	Description	Format	Quantity
Landscape painting images	Museums, galleries, artists available	High quality hand-painted or printed landscape paintings	JPG, PNG	2000 sheets
Landscape painting metadata	Literature and database	Contains author, year, materials used, etc	CSV, JSON	2000 pieces
Modern works of art	Open databases, social media	Works of modern art for comparative analysis	JPG, PNG	500 sheets
Deep neural networks generate works	Early experiment	Landscape paintings produced by different models	JPG, PNG	300 sheets

(1)

Landscape painting pictures: Obtain high-quality landscape painting pictures from famous museums and galleries, and cooperate with artists to ensure the quality and diversity of pictures.

(2)

Landscape painting metadata: Collect metadata related to each landscape painting work, such as author, creation year, materials used, etc., to provide necessary background information for subsequent in-depth analysis.

(3)

Modern art works: Through comparative analysis of traditional landscape painting and modern art works, explore the application effect of deep neural network in different artistic styles.

(4)

Works generated by deep neural networks: landscape paintings generated by different types of deep neural network models in early experiments are collected for evaluation and analysis.

With such a detailed and comprehensive data set, the research will be able to more accurately assess the application potential and limitations of deep neural networks in the creation of traditional Chinese landscape painting. This not only contributes to the progress of technology, but also contributes to the modernization and inheritance of traditional culture and art.

3.1.2 Data preprocessing

After obtaining the raw data set, the next step is data preprocessing. The pre-processing steps include image correction, standardization, data enhancement, and metadata cleaning. The specific pretreatment process is shown in Table 2.

Table 2
Image preprocessing

Operation procedure	Method	Instructions
Image correction	Color balance, contrast adjustment	Fix color and contrast distortion caused by scanning or shooting
Image standardization	Size adjustment, uniform format	Adjust all images to the same resolution and format
Data enhancement	Rotate, flip, crop, etc	Increase the diversity of data set and improve the generalization ability of model
Data quality control	Manual screening, automatic detection	Ensure clarity and representativeness of images
Data screening criteria	Artistic, historical value, technical	Select data that meets research needs according to
	characteristics	specific criteria

Evaluation indicators Metadata preprocessing

Data cleaning: For missing or inaccurate metadata, the corresponding null value filling or outlier processing is studied.

Data standardization: In order to facilitate subsequent data analysis and model training, the study standardized all metadata (e.g., when the work was created, author, etc.) into a uniform format and coding.

Through this series of meticulous data preprocessing steps, this study ensures the quality and consistency of the data set, providing a solid foundation for subsequent model training and evaluation.

3.2 Model design and training

3.2.1 Selected model type

After data preprocessing, the next step is model selection and training. Based on previous experiments and literature research, this study selected the following types of deep neural network models for comparison and analysis, as shown in Table 3.

Table 3
Comparison of model types

Model type	Symbolic representation	Main application
Convolutional Neural Networks (CNN)	CNN	Image classification, feature extraction
Recurrent neural network (RNN)	RNN	Sequential data processing
Generative adversarial network (GAN)	GAN	Image generation

(1) Convolutional Neural Network (CNN):

The mathematical representation is shown in the following Eq. (1):

$\displaystyle f(x)=\max(0,Wx+b)$ (1)

Convolutional neural network is one of the earliest deep learning models used in image processing, and it has excellent feature extraction ability.

(2) Recurrent neural network (RNN):

The mathematical representation is shown in the following Eq. (2):

$\displaystyle h_{t}=\sigma(W_{hh}h_{t-1}+W_{xh}x_{t})$ (2)

RNN is suitable for processing data with sequential relationships, and although it is not as widely used as CNN in image processing, it has its advantages in some specific application scenarios.

(3) Generate adversarial network (GAN):

The mathematical representation is shown in the following Eq. (3):

$\displaystyle G(z)=\text{argmin}_{G}\text{max}_{D}V(D,G)$ (3)

GAN is a model that can generate new images, which is especially suitable for creative art production.

After a series of comparison and experiments, generative adversarial network (GAN) is chosen as the main model type. This is because GAN can not only generate high-quality images, but also integrate more creative factors into the generation process, which is particularly important for the study of the role of deep neural networks in the creation of traditional Chinese landscape painting.

3.2.2 Training parameters and configuration

Based on the selected generative adversarial network (GAN) model, the next research needs to determine the training parameters and configuration of the network. These parameters are very critical in the model training process, affecting the performance of the model and the quality of the final product. As shown in Table 4, the main training parameters and corresponding Settings are listed.

Table 4
Parameter settings

Parameter name	Symbol	Set value	Description
Learning rate	$\alpha$	0.0002	Control the speed of model weight update
Lot size	$N$	64	The number of samples used per training
Number of iterations	$T$	50000	The number of iterations required to complete the training
Optimizer	–	Adam	An algorithm for updating network weights
Loss function	${\cal L}$	BCE	A function used to measure the performance of the model
Dropout rate	$p$	0.5	Probability Settings to prevent overfitting
Momentum	$\beta$	0.5	Momentum factor in the optimizer

In the training process, the research adopts Binary Cross-Entropy (BCE) as a loss function, which is expressed mathematically, as shown in Eq. (4) below:

$\displaystyle{\cal L}=-\frac{1}{N}\sum\limits_{i=1}^{N}{[{y_{i}\log({p({y_{i}}% )})+({1-y_{i}})\log({1-p({y_{i}})})}]}$ (4)

Where $y_{i}$ is the actual label and $p(y_{i})$ is the probability predicted by the model.

The optimizer adopts Adam, whose update rules are shown in the following Eq. (5):

$\displaystyle\theta_{t+1}=\theta_{t}-\alpha\cdot\textit{Adam}(\nabla{\cal L})$ (5)

Here, $\theta$ represents the model parameters, and $\nabla{\cal L}$ is the gradient of the loss function with respect to the model parameters.

Through these training parameters and configuration, the research aims to ensure that the model can achieve high-quality landscape painting production.

3.3 Evaluation indicators

3.3.1 Defining indicators

In order to comprehensively and accurately evaluate the performance of the model in generating traditional Chinese landscape painting, the following key evaluation indicators are defined in this study.

(1) Generation Quality (GQ): Measures the visual quality of the generated work. The following Eq. (6) is shown:

$\displaystyle GQ=\frac{1}{N}\sum\limits_{i=1}^{N}{S_{i}}$ (6)

Where, $S_{i}$ indicates the score of the generated work for the first $i$ . The value ranges from 0 to 10.

(2) Technical Accuracy (TA): Evaluating the technical details and accuracy of the generated work. The following Eq. (7) is shown:

$\displaystyle TA=\frac{\text{Correct number of technical details}}{\text{Total% number of technical details}}$ (7)

(3) Cultural Relevance (CR): Evaluate the cultural relevance of the generated works with traditional Chinese landscape paintings. The following Eq. (8) is shown:

$\displaystyle CR=\frac{\text{The number of culturally relevant elements}}{% \text{Total number of elements}}$ (8)

(4) Model Complexity (MC, Model Complexity): The total number of parameters calculated for the model. The following Eq. (9) is shown:

$\displaystyle MC=\text{Total parameters of the model}$ (9)

(5) Running Time (RT): Measures the time it takes for the model to go from input to production. The following Eq. (10) is shown:

$\displaystyle RT=t_{\text{end }}-t_{\text{start }}$ (10)

Table 5 shows sample data for calculating these evaluation metrics on a test set.

Table 5

Examples of evaluation indicators

Index name	Symbol	Numerical value	Unit/range
Generation quality	$G Q$	7.6	0–10
Technical accuracy	$T A$	0.85	0–1
Cultural relevance	$C R$	0.9	0–1
Model complexity	$M C$	1.2M	Parameter number
Running time	$R T$	200 ms	ms

Through these evaluation indicators, this study can comprehensively understand the performance of the model in different aspects, so as to guide the subsequent optimization and adjustment of the model more accurately.

3.3.2 Result evaluation

In order to comprehensively evaluate the model performance, this study will evaluate the results in several aspects, and the specific evaluation process is as follows:

(1) Build Quality Assessment: Build quality ( $G Q$ ) is scored by a panel of experts through manual review. The following Eq. (11) is shown:

$\displaystyle\text{Average generating mass}=\frac{\sum\limits_{i=1}^{N}GQ_{i}}% {N}$ (11)

Where $GQ_{i}$ is the generation quality of the first $i$ sample.

(2) Technical accuracy evaluation: investigates the use of technical accuracy ( $T A$ ) to quantify the technical accuracy of the image generated by the model. The following Eq. (12) is shown:

$\displaystyle TA_{\text{avg}}=\frac{\sum\limits_{i=1}^{N}TA_{i}}{N}$ (12)

Where $TA_{i}$ is the technical accuracy of the $i$ sample.

(3) Cultural Relevance assessment: Cultural relevance is calculated by calculating the frequency of culturally relevant elements in the sample ( $C R$ ). As shown in the following Eq. (13):

$\displaystyle CR_{\text{avg}}=\frac{\sum\limits_{i=1}^{N}CR_{i}}{N}$ (13)

Where $CR_{i}$ is the cultural correlation of the $i$ sample.

(4) Cultural Relevance assessment: Cultural relevance is calculated by calculating the frequency of culturally relevant elements in the sample ( $C R$ ). As shown in the following Eq. (14):

$\displaystyle CI=\alpha\times GQ_{\text{avg}}+\beta\times TA_{\text{avg}}+% \gamma\times CR_{\text{avg}}$ (14)

Where $\alpha,\beta,\gamma$ are the weight coefficients.

As shown in Fig. 1, the evaluation results of the sample data are presented.

Figure 1.

Data evaluation.

According to the calculation of the above data, the average value of the generated quality index in the sample data is 7.77, the average value of the technical accuracy index is 0.837, and the average value of the cultural relevance index is 0.907. Meanwhile, the comprehensive evaluation index of sample 1 was 7.84; The comprehensive evaluation index of sample 2 was 7.83; The composite evaluation index of sample 3 was 7.89. The average value of the comprehensive evaluation index was 7.853.

This evaluation method is both comprehensive and specific, and can effectively evaluate the model’s performance in generating traditional Chinese landscape painting from multiple perspectives.

4. Results and analysis

4.1 Visual quality assessment

In order to evaluate the visual quality of deep neural networks in generating traditional Chinese landscape paintings, this study invited five experts in art and technology to rate the works generated by the model. The rating scale is 1 to 10, with 10 representing high quality that is almost indistinguishable from hand-painted traditional landscape paintings.

In terms of average user ratings, we paid special attention to differences in visual quality perception among different groups. Average users include students, teachers, art lovers, IT professionals, etc., ranging in age from 20 to 60. With this diverse assessment approach, we aim to get broader and more objective feedback on visual quality.

Visual quality assessment data are shown in Fig. 2.

Figure 2.

Visual quality assessment form.

The analysis results show that there are some differences in the visual quality evaluation of the generated works among different groups, but the overall results still indicate that the generated works are highly evaluated in terms of visual quality. For example, the expert and general user ratings for Work 1 are 8.08 and 7.90, respectively, and the difference between the two is small, indicating that the work has a high acceptance among different audience groups.

Based on the above data, the overall average rating of the four works is between 7.73 and 8.08, indicating that the generated works are highly evaluated by experts and ordinary users in terms of visual quality. This diverse evaluation approach not only provides a comprehensive understanding of the visual quality of deep neural network-generated works, but also provides valuable feedback for further optimization and tuning of the model.

4.2 Technical performance evaluation

On the basis of visual quality assessment, the research also carried out a comprehensive evaluation of technical performance. The running time, accuracy and other relevant technical indexes of the model are mainly investigated here. As shown in Fig. 3.

Figure 3.

Technical performance evaluation.

(1)

Cultural similarity: Evaluate the similarity between the generated landscape painting and the traditional landscape painting in terms of cultural elements, styles and symbols. The average cultural similarity was 87.8%, reflecting a high degree of cultural similarity.

(2)

Thematic communication: Evaluate the effect of the generated works in conveying the theme and spirit of traditional Chinese landscape painting. The average thematic communication rate was 92.2%, indicating that the works successfully communicated traditional themes.

(3)

Aesthetic acceptance: Measuring the aesthetic satisfaction of ordinary audiences to the generated works. The average aesthetic acceptability was 85.4%, which showed good aesthetic acceptability.

In addition, this study also carries out cultural and aesthetic analysis from more perspectives to explore the positioning of landscape paintings generated by deep neural networks in the context of traditional art. For example, the expressive power of symbols and the innovation of works. The expressive power of symbols is to analyze the use and expression effect of traditional symbols and images in the generated works. The average symbolic expression score is 90.1%, indicating that the work does well in the expression of traditional symbols and images. Innovation is to evaluate the degree of innovation exhibited by a work while maintaining traditional characteristics. The average innovation score was 82.3%, indicating innovation while retaining tradition.

These technical performance indicators further confirm that the model not only has good visual performance in generating traditional Chinese landscape painting, but also has efficient and accurate technical performance.

4.3 Cultural and aesthetic analysis

In addition to the assessment of technical performance and visual quality, the study also considers the cultural and aesthetic value of the generated work. This part mainly evaluates from three aspects: cultural similarity, thematic communication and aesthetic acceptance. As shown in Fig. 4.

Figure 4.

Cultural and aesthetic evaluation.

(1)

Cultural similarity: the average cultural similarity is 87.8%, which indicates that the generated landscape paintings have a high degree of similarity with traditional landscape paintings on the cultural level.

(2)

Theme communication: The average theme communication reached 92.2%, indicating that the generated works successfully conveyed the theme and spirit of traditional Chinese landscape painting.

(3)

Aesthetic acceptance: The average aesthetic acceptance was 85.4%, indicating that most people would find these works aesthetically acceptable or satisfactory.

These cultural and aesthetic indicators, together with the previous technical performance and visual quality, prove the validity and feasibility of the deep neural network model in the creation of traditional Chinese landscape painting.

4.4 Limitations and solutions

In the experimental process and results analysis, some obvious limitations were also observed. The following is a detailed description of these limitations and possible solutions, as shown in Table 6.

Table 6
Limitations and solutions

Limitation	Description	Solution
Data set bias	The dataset may not adequately cover all styles and themes of traditional Chinese landscape painting	Expand the dataset to include more diverse landscape paintings
Computing resource	Deep neural network training and evaluation require a lot of computational resources	Use cloud or distributed computing to address resource constraints
Aesthetic standard	The criteria for assessing aesthetic quality may be influenced by individual differences and cultural background	Diverse aesthetic evaluation criteria for design, including multicultural factors
Technical limitation	Models may not perform well in some complex scenes, such as fine lines and colors	Continue optimizing the model, or try different types of deep learning architectures
Cultural adaptation	The works generated by the model may not fully meet the requirements of traditional Chinese culture	Add more culturally relevant factors in the model training phase
The balance between	Explore how to incorporate innovative elements	Explore ways to incorporate innovative
innovation and tradition	to promote the modernization of traditional art	elements while maintaining traditional aesthetics
User acceptance	Increase public awareness and acceptance of new technologies	Increase public participation and increase awareness and acceptance of new technologies
Long-term cultural impact	Ensure the sustainable development of culture	The long-term trend of the influence of deep neural networks on traditional landscape painting culture is studied

(1)

Data set bias: The limitations of the data set may cause the model to fail to generate diverse or complex landscape paintings. This not only limits the scope of application of the model, but also may produce works that do not conform to some traditional aesthetic concepts.

(2)

Computational resources: Deep learning models require a lot of computational resources, which may limit the participation of some researchers and artists.

(3)

Aesthetic criteria: Because aesthetic is subjective and pluralistic, using a single criterion to evaluate a generated work may lead to bias in the evaluation results.

(4)

Technical limitations: The current deep learning technology has not reached the level of fully simulating the creative process of complex landscape painting, especially in the processing of fine textures and complex scenes.

(5)

Cultural adaptability: Even if the model performs well in terms of technical performance and visual quality, its application value will be greatly reduced if the works generated by it fail to meet the requirements of traditional culture.

(6)

Balance between innovation and tradition: Explore how to incorporate innovative elements while maintaining traditional landscape painting aesthetics to promote the modernization of traditional art.

(7)

User acceptance: Increase public awareness of new technologies and improve their acceptance through public education and exhibitions.

(8)

Long-term cultural impact: Study the impact of deep neural network technology on traditional landscape painting culture in the long run to ensure the sustainable development of culture.

The above discussion of limitations and solutions is intended to provide a comprehensive and in-depth analysis to guide future research and applications within this intersectional area.

5. Conclusions

This study is devoted to exploring the role and influence of deep neural networks in the creation of traditional Chinese landscape painting, covering the whole process from theoretical background, experimental design, practical application and multi-dimensional evaluation. First of all, we start with the history of Chinese landscape painting and the status quo of deep neural networks in art creation, and confirm the urgency and innovation of the research. The experimental design process ensures the scientificity and rigor of the study through detailed data set preparation, model training and evaluation index setting.

The main contribution of this study is to demonstrate the potential of deep neural networks in generating high-quality traditional Chinese landscape paintings, thus providing new possibilities for the integration of modern technology and traditional art. Through multi-dimensional evaluation methods, including visual quality, cultural aesthetics and technical performance, it provides a comprehensive evaluation framework for the application of deep neural networks in the field of artistic creation. The limitations of deep learning in dealing with complex cultural and artistic works are identified, and possible solutions are proposed, providing directions for future research.

Future research directions include:

Algorithm Optimization and Innovation: Explore more efficient deep learning models to better handle complex artistic elements and textures, enhancing the creativity and flexibility of the models.

Deep integration of cultural elements: In-depth study of the cultural background and aesthetic characteristics of traditional Chinese landscape painting, and more deeply integrate these elements into the model training.

Expand application scenarios: Explore the application of deep neural networks in other types of traditional art creation, such as calligraphy, Beijing Opera, etc., to broaden the scope of technology application.

Interdisciplinary collaboration and experimentation: Collaboration between artists, technologists, and academics is encouraged to explore new possibilities for deep neural networks in the creation of art through experimental projects.

Overall, this study not only has high academic value, but also provides a series of useful insights at the practical level. Through comprehensive and rigorous research design and execution, as well as multi-angle analysis and interpretation of the results, this research has laid a solid foundation for the cross-research between deep neural networks and art, especially traditional Chinese art. Facing the future, there are still many directions worth exploring in this research field, including algorithm optimization, deep integration of cultural elements, and expansion of practical application scenarios, which are all topics worthy of further study and discussion.

References

Chung

Huang

. Interactively transforming Chinese ink paintings into realistic images using a border enhance generative adversarial network. Multimed Tools Appl. 2023; 82: 11663-11696.

Jin

. Application of machine vision based on genetic algorithm in image painting style method and image processing optimization. Soft Comput. 2023.

Chen

. Exploration of artistic creation of Chinese ink style painting based on deep learning framework and convolutional neural network model. Soft Comput. 2020; 24: 7873-7884.

Guo

Bai

Wang

Zhang

Dai

Wang

. ShadowPainter: Active learning enabled robotic painting through visual measurement and reproduction of the artistic creation process. J Intell Robot Syst. 2022; 105: 61.

Zhou

Wei

Wang

Zhou

. Fast generation method of 3D scene in Chinese landscape painting. Multimed Tools Appl. 2020; 79: 16441-16457.

Qiu

Zhang

. Application of digital technology in painting using new media and big data. Soft Comput. 2023; 27: 12691-12709.

. Application of computer image technology in 3D painting based on cloud computing. Soft Comput. 2023.

Liu

Yang

Shi

Jin

. Painting completion with generative translation models. Multimed Tools Appl. 2020; 79: 14375-14388.

Yang

Cheng

Liu

Shih

. The impact of immersive virtual reality on art education: A study of flow state, cognitive load, brain state, and motivation. Educ Inf Technol. 2023.

10.

Cai

. Assessing the best art design based on artificial intelligence and machine learning using GTMA. Soft Comput. 2023; 27: 149-156.

11.

Zhang

Ding

. Art image inpainting via embedding multiple attention dilated convolutions. Multimed Tools Appl. 2023.

12.

Scalera

Seriani

Gasparetto

Gallina

. Watercolour robotic painting: A novel automatic system for artistic rendering. J Intell Robot Syst. 2019; 95: 871-886.

13.

Lan

Khan

. Evaluating the scope of digital media art education using the decision support system. Soft Comput. 2023.

14.

Zhang

Zhou

Xia

Jiang

Liu

. A novel automatic image segmentation method for Chinese literati paintings using multi-view fuzzy clustering technology. Multimedia Syst. 2020; 26: 37-51.

15.

Otsuki

Sugihara

Toda

Shibata

Kimura

. A brush device with visual and haptic feedback for virtual painting of 3D virtual objects. Virtual Real-London. 2018; 22: 167-181.

16.

Fiorillo

Matteucci

Cataldo

Apicella

Vandini

. A multi-analytical approach for the characterisation of 20th century paintings. Eur Phys J Plus. 2019; 134: 373.

17.

Yang

Zhang

Shi

Liu

Guo

Sheng

. Framework of personalized layout for a museum exhibition hall. Multimed Tools Appl. 2023.

18.

Deng

Tang

Dong

Deussen

. Selective clustering for representative paintings selection. Multimed Tools Appl. 2019; 78: 19305-19323.

19.

Jin

Chen

Xiao

Tian

Lan

. Deep semantic space guided multi-scale neural style transfer. Multimed Tools Appl. 2022; 81: 3915-3938.

20.

Wang

Sang

. An abstract painting generation method based on deep generative model. Neural Process Lett. 2020; 52: 949-960.

The role of deep neural network in the creation of traditional Chinese landscape painting

Abstract

Keywords

1. Introduction

2. Cross application: Deep neural network and Chinese landscape painting

2.1 Current situation of contemporary landscape painting creation

2.2 The role of deep neural networks in art

2.3 Intersection and challenge of landscape painting and deep neural network

3. Experimental design and method

3.1 Data set preparation

3.1.1 Source and type of data

Table 1 Data sources and types

Table 2 Image preprocessing

3.2.1 Selected model type

Table 3 Comparison of model types

Table 4 Parameter settings

3.3.1 Defining indicators

4.1 Visual quality assessment

Table 6 Limitations and solutions

References

Table 1
Data sources and types

Table 2
Image preprocessing

Table 3
Comparison of model types

Table 4
Parameter settings

Table 6
Limitations and solutions