Poster design method based on artificial intelligence

Abstract

The invention relates to the technical field of image generation, and particularly discloses an artificial intelligence-based poster design method, which comprises the steps of data collection and preprocessing, model design, layout optimization, user feedback and poster output. According to the scheme, the generative adversarial network is adopted to automatically design the poster, a generator and a discriminator are trained through a large number of labeled real poster design data sets, and a more complex design mode is learned; in combination with the convolutional neural network, poster layout is automatically extracted, element positions are optimized, and the color matching degree is scientifically quantified, so that the overall design style is more coordinated and accords with visual aesthetics, and the optimization efficiency is effectively improved; a user real-time interaction and feedback mechanism is introduced, so that the generated design better meets personalized requirements, poster design images are updated in real time, an adjustment result is instant and visible, a user satisfaction function is designed to optimize a generative adversarial network, the system gradually learns aesthetic appreciation and requirements of users, and the continuous improvement capability of the system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of image generation, and in particular to an artificial intelligence-based poster design method. Background Art

[0002] With the development of artificial intelligence technology, using artificial intelligence for image design has become a trend. The general poster design method lacks diversity, tends to be single in style, lacks a scientific evaluation mechanism, and the rationality of layout and color matching is difficult to quantify, which is prone to incoordination problems; the traditional poster optimization process is cumbersome and lacks scientific basis, which easily leads to visual distraction and layout imbalance; the poster interaction method is inefficient, modification requirements need to be completed through repeated operations, and the lagging feedback mechanism leads to repeated communication and poor user experience. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a poster design method based on artificial intelligence. In view of the fact that the general poster design method lacks diversity, tends to be single in style, lacks a scientific evaluation mechanism, the rationality of layout and color matching is difficult to quantify, and is prone to incoordination problems, this scheme adopts a generative adversarial network to automatically design posters, trains the generator and discriminator through a large number of annotated real poster design data sets, learns more complex design patterns, dynamically responds to user design needs, improves design efficiency, designs a loss function to adjust the erroneous judgment in the generated layout, measures the rationality and authenticity of the generated layout, and provides an objective basis for design optimization; the traditional poster optimization process is cumbersome, lacks a scientific basis, and is prone to To address the problems of visual distraction and layout imbalance, this solution combines generative adversarial networks and convolutional neural networks to automatically extract poster layouts and optimize element positions, scientifically quantify color matching, make the overall design style more coordinated, conform to visual aesthetics, and effectively improve optimization efficiency; in response to the inefficiency of poster interaction methods, modification requirements need to be completed through repeated operations, and the lagging feedback mechanism leads to multiple repeated communications and poor user experience, this solution introduces a user real-time interaction and feedback mechanism to make the generated design more in line with personalized needs, update the poster design image in real time, make the adjustment results visible immediately, and design a user satisfaction function to optimize the generative adversarial network, so that the system can gradually learn the user's aesthetics and needs to ensure the system's continuous improvement capabilities.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a poster design method based on artificial intelligence, which comprises the following steps: Step S1: Data collection and preprocessing: Collect poster design samples, each poster design sample includes a poster image and design elements, preprocess the poster image, label the design elements, construct a data set, and divide the data set into a training set and a test set; Step S2: Model design, obtaining the user's design requirements, building a generative adversarial network, including a generator and a discriminator, using the generator to generate an initial poster layout image based on the user's design requirements, and using the discriminator to judge the authenticity of the initial poster layout image; Step S3: layout optimization, embedding a feature extraction module based on a convolutional neural network in the generator, and using the feature extraction module to optimize the design elements in the initial poster layout image to obtain a poster design image; Step S4: user feedback, providing a user-friendly interactive interface, through which the user adjusts the poster design image, and the parameters of the generative adversarial network are updated in real time according to the user's adjustment operation to obtain the adjusted poster design image; Step S5: Poster output, outputting the adjusted poster design image into an image file format available to the user.

[0005] Furthermore, in step S2, the model design specifically includes the following steps: Step S21: demand input, obtaining the user's design requirements, and extracting the input data of the generator from the user's design requirements; Step S22: constructing a generator, including an embedding layer, a fully connected layer, a convolutional layer, and a transposed convolutional layer, and outputting an initial poster layout image based on the input data; Step S23: constructing a discriminator, including a convolutional layer and a fully connected layer, extracting a real layout image from the training set according to the label, inputting the initial poster layout image and the real layout image, the convolutional layer extracts features from the initial poster layout image and the real layout image respectively, and the fully connected layer maps the features extracted by the convolutional layer to probability values; Step S24: Define a loss function based on the probability value, train the generative adversarial network on the training set, maximize the correct judgment of the real layout image, and minimize the wrong judgment of the initial poster layout image. The formula used is as follows: ; In the formula, represents the loss function, Represents the real layout image, represents the data points in the real layout image, Represents the actual layout image Upsampled data points The expected value when represents the probability value output by the discriminator, represents the logarithmic function, represents a random noise vector, represents the random noise distribution, Indicates the noise distribution The expected value on Represents the initial poster layout image.

[0006] Furthermore, in step S22, the generator is constructed, specifically comprising the following steps: Step S221: construct an embedding layer to generate a feature vector based on the input data; Step S222: construct a fully connected layer to convert the feature vector into a low-resolution feature map; Step S223: construct a convolution layer, extract the spatial features in the low-resolution feature map through a filter, and each convolution layer outputs a convolution feature map of a different level; Step S224: construct a transposed convolution layer to gradually restore high-dimensional features of the convolution feature map to generate a high-resolution feature map; Step S225: Image output. The high-resolution feature map generated by the last transposed convolutional layer is the initial poster layout image, and the initial poster layout image is output.

[0007] Furthermore, in step S3, the layout optimization specifically includes the following steps: Step S31: constructing a feature extraction module based on a convolutional neural network, and using the feature extraction module to extract the position and color matching of the design elements of the initial poster layout image;

[0008] Step S32: Calculate the optimal position of each design element and optimize the position of the design element in the initial poster layout image. The formula used is as follows: ; In the formula, represents the optimal position of the design element, Represents the two-dimensional coordinates of the design element, represents the total number of design elements, represents the index of the design element, Indicates the attention value of the design element, represents the maximum value function; Step S33: Calculate the color matching between the design elements and optimize the color matching of the design elements. The formula used is as follows: ; In the formula, Indicates color matching. Represents the RGB color value of the design element, represents the color weight, represents the summation function, Indicates the distance between the RGB color values ​​of design elements.

[0009] Further, in step S4, the user feedback specifically includes the following steps: Step S41: interactive design, designing an interactive interface for users to adjust poster design images and design elements; Step S42: Real-time feedback optimization, updating the poster design image in real time according to the user's adjustment operation; Step S43: Test the generative adversarial network on the test set, define the user satisfaction function to optimize the generative adversarial network, and maximize the user satisfaction. The formula used is as follows: ; In the formula, Indicates the maximum value of user satisfaction. Indicates user satisfaction, Indicates readability, Indicates aesthetics, Indicates functionality, Indicates adjustable weight.

[0010] The beneficial effects achieved by the present invention using the above scheme are as follows: (1) In view of the fact that general poster design methods lack diversity, tend to have a single style, lack a scientific evaluation mechanism, and are difficult to quantify the rationality of layout and color matching, which is prone to incoordination problems, this scheme uses a generative adversarial network to automatically design posters. The generator and discriminator are trained through a large number of annotated real poster design data sets to learn more complex design patterns, dynamically respond to user design needs, and improve design efficiency. The loss function is designed to adjust the erroneous judgments in the generated layout, measure the rationality and authenticity of the generated layout, and provide an objective basis for design optimization.

[0011] (2) In view of the fact that the traditional poster optimization process is cumbersome, lacks scientific basis, and easily leads to visual distraction and layout imbalance, this solution combines generative adversarial networks and convolutional neural networks to automatically extract poster layouts and optimize element positions, scientifically quantify color matching, make the overall design style more coordinated, conform to visual aesthetics, and effectively improve optimization efficiency.

[0012] (3) In order to address the problems of low efficiency of poster interaction methods, modification requirements need to be completed through repeated operations, lagging feedback mechanism leading to multiple repeated communications, and poor user experience, this solution introduces a user real-time interaction and feedback mechanism to make the generated design more in line with personalized needs, update the poster design image in real time, make the adjustment results visible immediately, design a user satisfaction function to optimize the generative adversarial network, so that the system can gradually learn the user's aesthetics and needs, and ensure the system's continuous improvement capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a poster design method based on artificial intelligence proposed by the present invention; Figure 2is a schematic flow chart of step S2; Figure 3 It is a schematic diagram of the process of step S4.

[0014] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] Example 1, see Figure 1 The present invention provides a poster design method based on artificial intelligence, which comprises the following steps: Step S1: Data collection and preprocessing: Collect poster design samples. Each poster design sample contains a poster image and design elements. The design elements include theme colors, keywords, and design styles. Preprocess the poster images and label the design elements. Construct a data set. Randomly select 70% of the data from the data set as a training set and 30% of the data as a test set. Step S2: Model design, obtaining the user's design requirements, building a generative adversarial network, including a generator and a discriminator, using the generator to generate an initial poster layout image based on the user's design requirements, and using the discriminator to judge the authenticity of the initial poster layout image; Step S3: layout optimization, embedding a feature extraction module based on a convolutional neural network in the generator, and using the feature extraction module to optimize the design elements in the initial poster layout image to obtain a poster design image; Step S4: user feedback, providing a user-friendly interactive interface, through which the user adjusts the poster design image, and the parameters of the generative adversarial network are updated in real time according to the user's adjustment operation to obtain the adjusted poster design image; Step S5: Poster output, outputting the adjusted poster design image into an image file format available to the user.

[0017] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the poster image is preprocessed, specifically including the following steps: Step S11: image standardization, adjusting the size of the poster image to 224×224 and converting it into RGB format; Step S12: labeling, labeling the design elements in the poster image, wherein the label content includes the position, size and category of the design elements, and obtaining the labeled poster image; Step S13: Image enhancement: Process the annotated poster image by rotating, cropping, scaling, and color conversion to obtain a processed poster image, construct a data set, and add the annotated poster image and the processed poster image into the data set.

[0018] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the model design specifically includes the following steps: Step S21: demand input, obtaining the user's design requirements, and extracting the input data of the generator from the user's design requirements, including theme color, keywords, and design style; Step S22: Generator construction, including embedding layer, fully connected layer, convolution layer and transposed convolution layer. The specific steps are as follows: Step S221: construct an embedding layer, convert the theme color into RGB value, process keywords through the pre-trained natural language model, introduce the style classification network to process the design style, and output the feature vector; Step S222: construct a fully connected layer to convert the feature vector into a low-resolution feature map; Step S223: Construct a convolution layer to extract spatial features in the low-resolution feature map through filters. The spatial features include edges, corners, and color distribution. Each convolution layer outputs convolution feature maps at different levels. The formula used is as follows: ; In the formula, represents the hierarchical index of the convolutional layer, Respectively represent Layer, The convolution feature map output by the convolution layer, represents the convolution kernel weight, represents the convolution operation, represents the bias term; Step S224: construct a transposed convolution layer to gradually restore the high-dimensional features of the convolution feature map to generate a high-resolution feature map. The formula used is as follows: ; In the formula, Respectively represent Layer, The convolutional feature map of the convolutional layer input is transposed. represents the transposed convolution kernel, represents the bias term, represents the activation function; Step S225: Image output: the high-resolution feature map generated by the last transposed convolution layer is the initial poster layout image, and the initial poster layout image is output; Step S23: construct a discriminator, including a convolutional layer and a fully connected layer, extract the real layout image from the training set according to the label, input the initial poster layout image and the real layout image, and judge whether the initial poster layout image is close to the real layout image. The specific steps are as follows: Step S231: construct a convolution layer to extract features from the initial poster layout image and the real layout image respectively, wherein the features include edges, textures, color blocks, shapes, layout structures, and element relationships; Step S232: construct a fully connected layer, and map the features extracted by the convolutional layer into probability values: if the real layout image is input, the probability value output by the discriminator is close to 1; if the initial poster layout image is input, the probability value output by the discriminator is close to 0; Step S24: Define a loss function based on the probability value, train the generative adversarial network on the training set, maximize the correct judgment of the real layout image, and minimize the wrong judgment of the initial poster layout image. The formula used is as follows: ; In the formula, represents the loss function, Represents the real layout image, represents the data points in the real layout image, Represents the actual layout image Upsampled data points The expected value when represents the probability value output by the discriminator, represents the logarithmic function, represents a random noise vector, represents the random noise distribution, Indicates the noise distribution The expected value on Represents the initial poster layout image.

[0019] By performing the above operations, the general poster design method has insufficient generation diversity, tends to be single in style, lacks a scientific evaluation mechanism, the rationality of layout and color matching is difficult to quantify, and is prone to incoordination problems. This scheme adopts a generative adversarial network to automatically design posters. The generator and discriminator are trained through a large number of annotated real poster design data sets to learn more complex design patterns, dynamically respond to user design needs, improve design efficiency, design loss function to adjust the erroneous judgment in the generated layout, measure the rationality and authenticity of the generated layout, and provide an objective basis for design optimization.

[0020] Example 4, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S223, a total of 5 convolution layers are set: the first convolution layer extracts low-level spatial features, namely edges and textures; the second to third convolution layers extract intermediate spatial features, namely color blocks and shapes; the fourth to fifth convolution layers extract high-level spatial features, namely layout structures and element relationships.

[0021] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S3, the layout optimization specifically includes the following steps: Step S31: constructing a feature extraction module based on a convolutional neural network, and using the feature extraction module to extract the position and color matching of the design elements of the initial poster layout image; Step S32: Calculate the optimal position of each design element and optimize the position of the design element in the initial poster layout image. The formula used is as follows: ; In the formula, represents the optimal position of the design element, Represents the two-dimensional coordinates of the design element, represents the total number of design elements, represents the index of the design element, Indicates the attention value of the design element, represents the maximum value function;

[0022] Step S33: Calculate the color matching between the design elements and optimize the color matching of the design elements. The formula used is as follows: ; In the formula, Indicates color matching. Represents the RGB color value of the design element, represents the color weight, represents the summation function, Indicates the distance between the RGB color values ​​of design elements.

[0023] By performing the above operations, the traditional poster optimization process is complicated, lacks scientific basis, and easily leads to visual distraction and layout imbalance. This solution combines generative adversarial networks and convolutional neural networks to automatically extract poster layouts and optimize element positions, scientifically quantify color matching, make the overall design style more coordinated, conform to visual aesthetics, and effectively improve optimization efficiency.

[0024] Example 6, see Figure 1 and Figure 3This embodiment is based on the above embodiment. In step S4, the user feedback specifically includes the following steps: Step S41: interactive design, designing an interactive interface for users to adjust the poster design image and design elements, the interactive content includes position modification, content modification, and color modification of the design elements; Step S42: Real-time feedback optimization, updating the poster design image in real time according to the user's adjustment operation; Step S43: Test the generative adversarial network on the test set, define the user satisfaction function to optimize the generative adversarial network, and maximize the user satisfaction. The formula used is as follows: ; In the formula, Indicates the maximum value of user satisfaction. Indicates user satisfaction, Indicates readability, Indicates aesthetics, Indicates functionality, Indicates adjustable weight.

[0025] By executing the operations, this solution introduces a real-time user interaction and feedback mechanism to address the problems of low efficiency of poster interaction methods, modification requirements needing to be completed through repeated operations, lagging feedback mechanism leading to multiple repeated communications, and poor user experience. This allows the generated design to be more in line with personalized needs, updates the poster design image in real time, makes the adjustment results visible immediately, and designs a user satisfaction function to optimize the generative adversarial network, so that the system can gradually learn the user's aesthetics and needs, ensuring the system's ability to continuously improve.

[0026] Embodiment 7, see Figure 1 This embodiment is based on the above embodiment. In step S5, the image file formats available to the user include BMP, TIF, PNG, JPG, PDF, and RAW formats.

[0027] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0028] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0029] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A poster design method based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1: Data collection and preprocessing: Collect poster design samples, each poster design sample includes a poster image and design elements, preprocess the poster image, label the design elements, construct a data set, and divide the data set into a training set and a test set; Step S2: Model design, obtaining the user's design requirements, building a generative adversarial network, including a generator and a discriminator, using the generator to generate an initial poster layout image based on the user's design requirements, and using the discriminator to judge the authenticity of the initial poster layout image; Step S3: layout optimization, embedding a feature extraction module based on a convolutional neural network in the generator, and using the feature extraction module to optimize the design elements in the initial poster layout image to obtain a poster design image; Step S4: user feedback, providing a user-friendly interactive interface, through which the user adjusts the poster design image, and the parameters of the generative adversarial network are updated in real time according to the user's adjustment operation to obtain the adjusted poster design image; Step S5: poster output, outputting the adjusted poster design image into an image file format available to the user; In step S2, the model design includes the following steps: Step S21: demand input, obtaining the user's design requirements, and extracting the input data of the generator from the user's design requirements; Step S22: constructing a generator, including an embedding layer, a fully connected layer, a convolutional layer, and a transposed convolutional layer, and outputting an initial poster layout image based on the input data; Step S23: constructing a discriminator, including a convolutional layer and a fully connected layer, extracting a real layout image from the training set according to the label, inputting the initial poster layout image and the real layout image, the convolutional layer extracts features from the initial poster layout image and the real layout image respectively, and the fully connected layer maps the features extracted by the convolutional layer to probability values; Step S24: Define a loss function based on the probability value, train the generative adversarial network on the training set, maximize the correct judgment of the real layout image, and minimize the wrong judgment of the initial poster layout image. The formula used is as follows: ； In the formula, represents the loss function, Represents the real layout image, represents the data points in the real layout image, Represents the actual layout image Upsampled data points The expected value when represents the probability value output by the discriminator, represents the logarithmic function, represents a random noise vector, represents the random noise distribution, Indicates the noise distribution The expected value on Represents the initial poster layout image.

2. The poster design method based on artificial intelligence according to claim 1, characterized in that: In step S22, the generator is constructed, including the following steps: Step S221: construct an embedding layer to generate a feature vector based on the input data; Step S222: construct a fully connected layer to convert the feature vector into a low-resolution feature map; Step S223: construct a convolution layer, extract the spatial features in the low-resolution feature map through a filter, and each convolution layer outputs a convolution feature map of a different level; Step S224: construct a transposed convolution layer to gradually restore high-dimensional features of the convolution feature map to generate a high-resolution feature map; Step S225: Image output. The high-resolution feature map generated by the last transposed convolutional layer is the initial poster layout image, and the initial poster layout image is output.

3. The poster design method based on artificial intelligence according to claim 1, characterized in that: In step S3, the layout optimization includes the following steps: Step S31: constructing a feature extraction module based on a convolutional neural network, and using the feature extraction module to extract the position and color matching of the design elements of the initial poster layout image; Step S32: Calculate the optimal position of each design element and optimize the position of the design element in the initial poster layout image. The formula used is as follows: ； In the formula, represents the optimal position of the design element, Represents the two-dimensional coordinates of the design element, represents the total number of design elements, represents the index of the design element, Indicates the attention value of the design element, represents the maximum value function; Step S33: Calculate the color matching between the design elements and optimize the color matching of the design elements. The formula used is as follows: ； In the formula, Indicates color matching. Represents the RGB color value of the design element, represents the color weight, represents the summation function, Indicates the distance between the RGB color values ​​of design elements.

4. The poster design method based on artificial intelligence according to claim 1, characterized in that: In step S4, the user feedback includes the following steps: Step S41: interactive design, designing an interactive interface for users to adjust poster design images and design elements; Step S42: Real-time feedback optimization, updating the poster design image in real time according to the user's adjustment operation; Step S43: Test the generative adversarial network on the test set, define the user satisfaction function to optimize the generative adversarial network, and maximize the user satisfaction. The formula used is as follows: ； In the formula, Indicates the maximum value of user satisfaction. Indicates user satisfaction, Indicates readability, Indicates aesthetics, Indicates functionality, Indicates adjustable weight.

Images