A graphic element library management method and system for packaging design

Abstract

The present invention relates to the field of data processing, and in particular to a method and system for managing a graphic element library for packaging design. A graphic element library management system for packaging design comprises: a graphic element acquisition module, a label recognition module, a feature extraction module, a graphic element annotation module, a candidate graphic element set construction module, and a candidate graphic element set update module. The present invention annotates the graphic elements required by the designer with the label dimension and the feature dimension of the graphic itself, and when the designer performs a graphic element retrieval, calculates the corresponding similarity based on the label dimension and the feature dimension of the graphic itself, and assists the designer in the retrieval work through the corresponding similarity. At the same time, the designer's retrieval operation is analyzed to determine the degree of inclination for the label dimension and the feature dimension of the graphic itself to adjust the retrieval direction, which can improve the retrieval speed.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method and system for managing a graphic element library for packaging design. Background Art

[0002] In the field of modern packaging design, the use and management of graphic elements are becoming increasingly important. With the diversification of market demands and the rapid changes in consumer preferences, designers need to frequently find and use various graphic elements to meet the packaging needs of different products. However, traditional graphic element libraries mostly use static classification methods. When designers look for specific elements, they often need to spend a lot of time manually searching in a large and complex image library, which results in low query efficiency and seriously restricts the work efficiency of designers. Summary of the invention

[0003] The present invention annotates the graphic elements required by the designer with label dimensions and the characteristic dimensions of the graphic itself, and when the designer performs graphic element retrieval, the corresponding similarity is calculated based on the label dimension and the characteristic dimension of the graphic itself, and the designer is assisted in the retrieval work through the corresponding similarity. At the same time, the designer's retrieval operation is analyzed to determine the degree of inclination towards the label dimension and the characteristic dimension of the graphic itself to adjust the retrieval direction, which can improve the retrieval speed.

[0004] A method for managing a graphic element library for packaging design, comprising:

[0005] Step S1: When generating a graphic element, perform the following steps:

[0006] Step S1.1: Obtaining graphic elements for packaging design;

[0007] Step S1.2: Send the graphic elements to the label recognition model for processing and output the graphic labels; convert all graphic labels into label vectors through word embedding operations, and then concatenate all label vectors to construct the first graphic feature vector;

[0008] Step S1.3: sending the graphic element to the feature extraction model for processing, and outputting the second graphic feature vector;

[0009] Step S1.4: labeling the graphic element by using the graphic label, the first graphic feature vector and the second graphic feature vector, and adding the labeled graphic element to the graphic element library;

[0010] Step S2: When searching for graphic elements in the graphic element library, the following steps are performed:

[0011] Step S2.1: obtaining an input tag, matching the input tag with the graphic tags corresponding to the graphic elements in the graphic element library, and randomly selecting N graphic elements from the successfully matched graphic elements to form a set of graphic elements to be selected;

[0012] Step S2.2: In response to the operation feedback, determine the first operation graphic element, the first operation graphic element being the graphic element in the set of graphic elements to be selected on which the user has performed an operation, traverse all the graphic elements in the graphic element library, and for each selected graphic element, calculate the first similarity δ1 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the first operation graphic element, calculate the second similarity δ2 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the first operation graphic element, and then calculate the first comprehensive similarity A1=α1δ1+α2δ2, wherein α1 and α2 are the first weight coefficient and the second weight coefficient, respectively, and satisfy α1+α2=1; sort the graphic elements in the graphic element library from large to small according to the first comprehensive similarity A1, and select the first N graphic elements to combine and replace the set of graphic elements to be selected;

[0013] Step S2.3: In response to the operation feedback, determine the second operation graphic element, the second operation graphic element is the graphic element on which the user has performed the operation in the new set of to-be-selected graphic elements, traverse all the graphic elements in the graphic element library, and for each selected graphic element, calculate the third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element, calculate the fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element, calculate the fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element, calculate the sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element, and then calculate the second comprehensive similarity , sorting the graphic elements in the graphic element library from large to small according to the second comprehensive similarity A2, and selecting the first N graphic elements to combine and replace the set of graphic elements to be selected; and then replacing the second operation graphic element with the first operation graphic element;

[0014] Step S2.4: Repeat step S2.3 until the search operation is completed to achieve the management of the graphic element library.

[0015] Preferably, the label recognition model includes a feature extraction layer and M parallel graphic label output layers, wherein the feature extraction layer is used to extract features of graphic elements to construct an intermediate feature vector; the graphic label output layer is used to process the intermediate feature vector to output the corresponding graphic label.

[0016] Preferably, training the label recognition model specifically includes the following steps:

[0017] Acquire several graphic element training samples, and annotate the graphic element training samples with graphic labels; combine all annotated graphic element training samples into a graphic element training set, and then send the graphic element training set to a parameter-initialized label recognition model for training to determine whether a first training condition is met. If so, output the trained label recognition model; otherwise, continue to train the label recognition model with the graphic element training set.

[0018] Preferably, the training of the feature extraction model specifically includes the following steps:

[0019] Acquire several sets of graphic element training samples, and sequentially send all the graphic element training samples to the pre-trained feature extraction model for processing, and output the second graphic feature vector corresponding to each graphic element training sample; perform unsupervised cluster analysis on all the second graphic feature vectors, construct several first clusters, and for each first cluster, select the cluster center of the first cluster to annotate all the graphic element training samples corresponding to the second graphic feature vectors in the first cluster; and form all the annotated graphic element training samples into a feature extraction training set;

[0020] The pre-trained feature extraction model is trained again through the feature extraction training set to determine whether the second training condition is met. If the second training condition is met, the trained feature extraction model is output; otherwise, the pre-trained feature extraction model is trained through the feature extraction training set.

[0021] Preferably, the process further includes strengthening the generation operation of the first set of candidate graphic elements for each designer's search history, and the specific steps are as follows:

[0022] For each designer, the corresponding search history data is obtained; the first graphic feature vectors corresponding to all graphic elements in the search history data are selected for cluster analysis, a number of second clusters are output, and the cluster center of the second cluster containing the most first graphic feature vectors is selected as the target graphic element;

[0023] Get an input label, match the input label with the graphic label corresponding to the graphic element in the graphic element library, and form the successfully matched graphic elements into an intermediate graphic element set, traverse the intermediate graphic element set, calculate the seventh similarity between the first graphic feature vector of the graphic element in the intermediate graphic element set and the first graphic feature vector of the target graphic element, sort the graphic elements in the intermediate graphic element set from large to small according to the seventh similarity, and select the first N graphic elements to combine and construct a selected graphic element set.

[0024] Preferably, the feature extraction layer and the feature extraction model in the tag recognition model are both set based on a convolutional neural network.

[0025] Preferably, the cluster analysis adopts the k-means algorithm.

[0026] The present invention also provides a graphic element library management system for packaging design, comprising:

[0027] A graphic element acquisition module, used for acquiring graphic elements for packaging design;

[0028] The label recognition module is used to send the graphic elements into the label recognition model for processing and output the graphic labels; convert all the graphic labels into label vectors through word embedding operations, and then concatenate all the label vectors to construct the first graphic feature vector;

[0029] A feature extraction module, used for sending the graphic elements into the feature extraction model for processing and outputting a second graphic feature vector;

[0030] A graphic element annotation module, used to annotate the graphic elements by using the graphic label, the first graphic feature vector and the second graphic feature vector, and add the annotated graphic elements to the graphic element library;

[0031] A module for constructing a set of graphic elements to be selected is used to obtain an input label, match the input label with a graphic label corresponding to a graphic element in a graphic element library, and randomly select N graphic elements from the successfully matched graphic elements to form a set of graphic elements to be selected;

[0032] A module for updating the set of graphic elements to be selected, used for updating the set of graphic elements to be selected in response to the operation feedback;

[0033] Update the set of selected graphic elements, which includes two parts:

[0034] When the designer performs the first operation, in response to the operation feedback, a first operation graphic element is determined, the first operation graphic element being the graphic element in the set of to-be-selected graphic elements on which the user performs the operation, all graphic elements in the graphic element library are traversed, and for each selected graphic element, a first similarity δ1 between a first graphic feature vector corresponding to the selected graphic element and a first graphic feature vector corresponding to the first operation graphic element is calculated, a second similarity δ2 between a second graphic feature vector corresponding to the selected graphic element and a second graphic feature vector corresponding to the first operation graphic element is calculated, and then a first comprehensive similarity A1=α1δ1+α2δ2 is calculated, wherein α1 and α2 are respectively the first weight coefficient and the second weight coefficient, and α1+α2=1 is satisfied; the graphic elements in the graphic element library are sorted from large to small according to the first comprehensive similarity A1, and the first N graphic elements are selected to be combined to replace the set of to-be-selected graphic elements;

[0035] When the designer performs the second and subsequent operations, in response to the operation feedback, a second operation graphic element is determined, the second operation graphic element is the graphic element on which the user performs the operation in the new set of to-be-selected graphic elements, all graphic elements in the graphic element library are traversed, and for each selected graphic element, a third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element is calculated, a fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element is calculated, a fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element is calculated, a sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element is calculated, and then a second comprehensive similarity is calculated. , sort the graphic elements in the graphic element library from large to small according to the second comprehensive similarity A2, and select the first N graphic elements to combine and replace the set of graphic elements to be selected; and then replace the second operation graphic element with the first operation graphic element.

[0036] The present invention has the following advantages:

[0037] The present invention annotates the graphic elements required by the designer with label dimensions and the characteristic dimensions of the graphic itself, and when the designer performs graphic element retrieval, the corresponding similarity is calculated based on the label dimension and the characteristic dimension of the graphic itself, and the designer is assisted in the retrieval work through the corresponding similarity. At the same time, the designer's retrieval operation is analyzed to determine the degree of inclination towards the label dimension and the characteristic dimension of the graphic itself to adjust the retrieval direction, which can improve the retrieval speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the structure of a graphic element library management system for packaging design adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0040] Embodiment 1, a method for managing a graphic element library for packaging design, comprising:

[0041] Step S1: When generating a graphic element, perform the following steps:

[0042] Step S1.1: Obtain graphic elements for packaging design. Graphic elements usually include icons, vector patterns, background patterns, buttons and illustrations. Designers will generate these graphic elements during the packaging design process. These graphic elements can constitute the packaging design and can be used as standardization in different projects to achieve the effect of maintaining consistent brand vision.

[0043] Step S1.2: Send the graphic elements to the label recognition model for processing, and output the graphic labels. The graphic labels include style labels and theme labels, such as minimalist style and retro style, and theme labels such as technology and nature. These graphic labels can facilitate designers to query graphic elements and improve work efficiency. All graphic labels are converted into label vectors through word embedding operations. It should be noted that word embedding can use the Word2vec model to convert graphic labels into encoding vectors with contextual connections, and then all label vectors are spliced ​​to construct the first graphic feature vector.

[0044] Step S1.3: Send the graphic element to the feature extraction model for processing, and output a second graphic feature vector. It should be noted that the second graphic feature vector here is different from the first graphic feature vector, and includes feature information that is more graphic-oriented, such as texture, contour and color;

[0045] Step S1.4: labeling the graphic element by using the graphic label, the first graphic feature vector and the second graphic feature vector, and adding the labeled graphic element to the graphic element library. It should be noted that, whenever the designer designs a new graphic element, intelligent labeling will be performed, and the graphic element library will be constructed accordingly;

[0046] Step S2: When searching for graphic elements in the graphic element library, the following steps are performed:

[0047] Step S2.1: Obtain input tags. The input tags are input by the designer who performs the search operation. The input tags have the same content as the graphic tags, including style tags and theme tags, etc. The input tags are matched with the graphic tags corresponding to the graphic elements in the graphic element library, and N graphic elements are randomly selected from the successfully matched graphic elements to form a set of selected graphic elements. The value of N is set by the developer based on the page size. It should be noted that in the process of searching for the required graphic elements, the designer will first input a rough tag, and then perform subsequent searches based on the tag;

[0048] Step S2.2: In response to the operation feedback, determine the first operation graphic element, the first operation graphic element being the graphic element in the set of to-be-selected graphic elements on which the user has performed an operation. It should be noted that when the graphic elements in the set of to-be-selected graphic elements are displayed on the page, the designer will perform a corresponding operation, such as a click operation. Based on this operation feedback, the graphic element selected by the click will be taken as the first operation graphic element, and all graphic elements in the graphic element library will be traversed. For each selected graphic element, a first similarity δ1 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the first operation graphic element is calculated, and a similarity δ2 between the second graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the first operation graphic element is calculated. A second similarity δ2 between the second graphic feature vectors corresponding to the first operation graphic element, and then calculate the first comprehensive similarity A1=α1δ1+α2δ2, wherein α1 and α2 are the first weight coefficient and the second weight coefficient respectively, and satisfy α1+α2=1, and the first weight coefficient and the second weight coefficient are set by the developer according to the nature of the design project, and are generally 0.5; the graphic elements in the graphic element library are sorted from large to small according to the first comprehensive similarity A1, and the first N graphic elements are selected to be combined to replace the set of graphic elements to be selected; it should be noted that in the process of designer retrieval, based on the designer's operation, similar graphic elements are recommended, which can improve the retrieval speed;

[0049] Step S2.3: In response to the operation feedback, determine the second operation graphic element, the second operation graphic element is the graphic element on which the user performs the operation in the new set of graphic elements to be selected, where the second operation graphic element refers to the graphic element on which the click operation is performed in the new set of graphic elements to be selected, traverse all the graphic elements in the graphic element library, and for each selected graphic element, calculate the third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element, calculate the fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element, calculate the fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element, calculate the sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element, and then calculate the second comprehensive similarity It should be noted that, based on the first operation graphic element, the second operation graphic element is selected again, and the third similarity and the fourth similarity can be used to characterize the designer's tendency in selecting graphic elements, including the tendency to select labels or the tendency to select the characteristics of the image itself. The proportion of the fifth similarity and the sixth similarity in the second comprehensive similarity is adjusted by the size of the third similarity and the fourth similarity, so that the calculated second comprehensive similarity can be more in line with the designer's selection tendency, thereby making the subsequent retrieval operation more accurate and improving the retrieval speed. The graphic elements in the graphic element library are sorted from large to small according to the second comprehensive similarity A2, and the first N graphic elements are selected to be combined to replace the set of graphic elements to be selected; and then the second operation graphic element is replaced by the first operation graphic element;

[0050] Step S2.4: Repeat step S2.3 until the search operation is completed, that is, the designer has determined the selected graphic element and performed operations such as downloading, thereby achieving management of the graphic element library.

[0051] This application annotates the graphic elements required by the designer with label dimensions and the characteristic dimensions of the graphic itself, and when the designer performs graphic element retrieval, calculates the corresponding similarity based on the label dimension and the characteristic dimension of the graphic itself, and assists the designer in the retrieval work through the corresponding similarity. At the same time, the designer's retrieval operation is analyzed to determine the degree of inclination towards the label dimension and the characteristic dimension of the graphic itself to adjust the retrieval direction, which can improve the retrieval speed.

[0052] The label recognition model includes a feature extraction layer and M parallel graphic label output layers, wherein the feature extraction layer is set based on a convolutional neural network. In this embodiment, the feature extraction layer is constructed based on the ResNet50 model, and the feature extraction layer is used to extract features of graphic elements to construct an intermediate feature vector; the graphic label output layer is used to process the intermediate feature vector to output the corresponding graphic label;

[0053] Training the label recognition model includes the following steps:

[0054] A number of graphic element training samples are obtained, where the graphic element training samples are graphic elements actually obtained by developers. The graphic element training samples are labeled with graphic labels, which are obtained based on the design information of the graphic elements. When designers design graphic elements, they will attach the corresponding design information. All labeled graphic element training samples are combined into a graphic element training set, and then the graphic element training set is sent to a parameter-initialized label recognition model for training to determine whether a first training condition is met. The first training condition may be that the loss value obtained meets expectations. If the first training condition is met, the trained label recognition model is output; otherwise, the label recognition model is continuously trained with the graphic element training set.

[0055] The feature extraction model is set based on the convolutional neural network, and the ResNet50 model is also selected. The training of the feature extraction model includes the following steps:

[0056] Obtain several sets of graphic element training samples, and send all of them to the pre-trained feature extraction model for processing in turn, and output the second graphic feature vector corresponding to each graphic element training sample. The pre-training operation here is implemented based on the ImgNet data set. The ResNet50 model trained by the ImgNet data set can accurately extract graphic features; perform unsupervised cluster analysis on all second graphic feature vectors, and use the k-means algorithm to construct several first clusters. For each first cluster, select the cluster center of the first cluster to annotate all graphic element training samples corresponding to the second graphic feature vectors in the first cluster; form all annotated graphic element training samples into a feature extraction training set;

[0057] The pre-trained feature extraction model is trained again through the feature extraction training set to determine whether the second training condition is met. The second training condition here can be to meet a certain number of training times. If the second training condition is met, the trained feature extraction model is output. It should be noted that although the pre-trained feature extraction model has good performance for graphic feature extraction, there will be certain deviations for specific tasks. Therefore, the feature extraction training set is used to fine-tune the feature extraction model to improve the accuracy of graphic feature extraction; otherwise, the pre-trained feature extraction model is trained through the feature extraction training set.

[0058] When actually searching for graphic elements, since the set of graphic elements to be selected is obtained randomly according to the input tags, the search speed still cannot achieve the best effect. Therefore, the present application also strengthens the generation operation of the first set of graphic elements to be selected according to the search history of each designer, which specifically includes the following steps:

[0059] For each designer, the corresponding search history data is obtained, and the search history data includes the graphic elements finally selected by all designers in the search history; the first graphic feature vectors corresponding to all graphic elements in the search history data are selected for cluster analysis, and the k-means algorithm can be used to output a number of second clusters, and the cluster center of the second cluster containing the most first graphic feature vectors is selected as the target graphic element;

[0060] Get an input label, match the input label with the graphic label corresponding to the graphic element in the graphic element library, and form the successfully matched graphic elements into an intermediate graphic element set, traverse the intermediate graphic element set, calculate the seventh similarity between the first graphic feature vector of the graphic element in the intermediate graphic element set and the first graphic feature vector of the target graphic element, sort the graphic elements in the intermediate graphic element set from large to small according to the seventh similarity, and select the first N graphic elements to combine and construct a selected graphic element set.

[0061] Embodiment 2, a graphic element library management system for packaging design, such as Figure 1 As shown, including:

[0062] Graphic element acquisition module, used to acquire graphic elements for packaging design. Graphic elements usually include icons, vector patterns, background patterns, buttons and illustrations. Designers will generate these graphic elements during the packaging design process. These graphic elements can constitute the packaging design and can be used as standardization in different projects to achieve the effect of maintaining consistent brand vision.

[0063] The label recognition module is used to send the graphic elements to the label recognition model for processing and output graphic labels. The graphic labels include style labels and theme labels, such as minimalist style and retro style, and theme labels such as technology and nature. These graphic labels can facilitate designers to query graphic elements and improve work efficiency. All graphic labels are converted into label vectors through word embedding operations. It should be noted that word embedding can use the Word2vec model to convert graphic labels into encoding vectors with contextual connections, and then all label vectors are spliced ​​to construct the first graphic feature vector.

[0064] A feature extraction module is used to send the graphic elements into the feature extraction model for processing and output a second graphic feature vector. It should be noted that the second graphic feature vector here is different from the first graphic feature vector and includes feature information that is more graphic-oriented, such as texture, contour and color;

[0065] A graphic element annotation module is used to annotate the graphic elements by using the graphic label, the first graphic feature vector and the second graphic feature vector, and add the annotated graphic elements to the graphic element library. It should be noted that whenever a designer designs a new graphic element, intelligent annotation work will be performed, and the graphic element library will be constructed accordingly;

[0066] The module for constructing the set of graphic elements to be selected is used to obtain input tags. The input tags are input by the designer who performs the search operation. The input tags have the same content as the graphic tags, including style tags and theme tags, etc. The input tags are matched with the graphic tags corresponding to the graphic elements in the graphic element library, and N graphic elements are randomly selected from the successfully matched graphic elements to form the set of graphic elements to be selected. The value of N is set by the developer based on the page size. It should be noted that in the process of searching for the required graphic elements, the designer will first input a rough tag, and then perform subsequent searches based on the tag;

[0067] A module for updating the set of graphic elements to be selected, used for updating the set of graphic elements to be selected in response to the operation feedback;

[0068] Update the set of selected graphic elements, which includes two parts:

[0069] When the designer performs the first operation, in response to the operation feedback, a first operation graphic element is determined, and the first operation graphic element is a graphic element in the set of to-be-selected graphic elements on which the user performs the operation. It should be noted that when the graphic elements in the set of to-be-selected graphic elements are displayed on the page, the designer will perform a corresponding operation, such as a click operation. Based on this operation feedback, the graphic element selected by the click will be taken as the first operation graphic element, and all graphic elements in the graphic element library will be traversed. For each selected graphic element, a first similarity δ1 between a first graphic feature vector corresponding to the selected graphic element and a first graphic feature vector corresponding to the first operation graphic element is calculated, and a second graphic feature vector corresponding to the selected graphic element is calculated. The second similarity δ2 between the second graphic feature vector corresponding to the first operation graphic element is measured, and then the first comprehensive similarity A1=α1δ1+α2δ2 is calculated, wherein α1 and α2 are respectively the first weight coefficient and the second weight coefficient, and α1+α2=1 is satisfied, and the first weight coefficient and the second weight coefficient are set by the developer according to the nature of the design project, and generally, both are 0.5; the graphic elements in the graphic element library are sorted from large to small according to the first comprehensive similarity A1, and the first N graphic elements are selected to be combined to replace the set of graphic elements to be selected; it should be noted that in the process of designer retrieval, based on the designer's operation, similar graphic elements are recommended, which can improve the retrieval speed;

[0070] When the designer performs the second and subsequent operations, in response to the operation feedback, a second operation graphic element is determined, the second operation graphic element is the graphic element on which the user performs the operation in the new set of graphic elements to be selected, and the second operation graphic element here refers to the graphic element on which the click operation is performed in the new set of graphic elements to be selected, and all graphic elements in the graphic element library are traversed, and for each selected graphic element, a third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element is calculated, a fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element is calculated, a fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element is calculated, a sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element is calculated, and then a second comprehensive similarity is calculated. It should be noted that, based on the first operation graphic element, the second operation graphic element is selected again, and the third similarity and the fourth similarity can be used to characterize the designer's tendency in selecting graphic elements, including the tendency to select labels or the tendency to select the characteristics of the image itself. The proportion of the fifth similarity and the sixth similarity in the second comprehensive similarity is adjusted by the size of the third similarity and the fourth similarity, so that the calculated second comprehensive similarity can be more in line with the designer's selection tendency, thereby making subsequent retrieval operations more accurate and improving the retrieval speed. The graphic elements in the graphic element library are sorted from large to small according to the second comprehensive similarity A2, and the first N graphic elements are selected to combine and replace the set of graphic elements to be selected; and then the second operation graphic element is replaced by the first operation graphic element.

[0071] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for managing a graphic element library for packaging design, characterized in that: include: Step S1: When generating a graphic element, perform the following steps: Step S1.1: Obtaining graphic elements for packaging design; Step S1.2: Send the graphic elements to the label recognition model for processing and output the graphic labels; convert all graphic labels into label vectors through word embedding operations, and then concatenate all label vectors to construct the first graphic feature vector; Step S1.3: sending the graphic element to the feature extraction model for processing, and outputting the second graphic feature vector; Step S1.4: labeling the graphic element by using the graphic label, the first graphic feature vector and the second graphic feature vector, and adding the labeled graphic element to the graphic element library; Step S2: When searching for graphic elements in the graphic element library, the following steps are performed: Step S2.1: obtaining an input tag, matching the input tag with the graphic tags corresponding to the graphic elements in the graphic element library, and randomly selecting N graphic elements from the successfully matched graphic elements to form a set of graphic elements to be selected; Step S2.2: In response to the operation feedback, determine the first operation graphic element, the first operation graphic element being the graphic element in the set of graphic elements to be selected on which the user has performed an operation, traverse all the graphic elements in the graphic element library, and for each selected graphic element, calculate the first similarity δ1 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the first operation graphic element, calculate the second similarity δ2 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the first operation graphic element, and then calculate the first comprehensive similarity A1=α1δ1+α2δ2, wherein α1 and α2 are the first weight coefficient and the second weight coefficient, respectively, and satisfy α1+α2=1; sort the graphic elements in the graphic element library from large to small according to the first comprehensive similarity A1, and select the first N graphic elements to combine and replace the set of graphic elements to be selected; Step S2.3: In response to the operation feedback, determine the second operation graphic element, the second operation graphic element is the graphic element on which the user has performed the operation in the new set of to-be-selected graphic elements, traverse all the graphic elements in the graphic element library, and for each selected graphic element, calculate the third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element, calculate the fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element, calculate the fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element, calculate the sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element, and then calculate the second comprehensive similarity , sorting the graphic elements in the graphic element library from large to small according to the second comprehensive similarity A2, and selecting the first N graphic elements to combine and replace the set of graphic elements to be selected; and then replacing the second operation graphic element with the first operation graphic element; Step S2.4: Repeat step S2.3 until the search operation is completed to achieve the management of the graphic element library.

2. A method for managing a graphic element library for packaging design according to claim 1, characterized in that: The label recognition model includes a feature extraction layer and M parallel graphic label output layers, wherein the feature extraction layer is used to extract features of graphic elements to construct an intermediate feature vector; the graphic label output layer is used to process the intermediate feature vector to output the corresponding graphic label.

3. A method for managing a graphic element library for packaging design according to claim 2, characterized in that: Training the label recognition model includes the following steps: Acquire several graphic element training samples, and annotate the graphic element training samples with graphic labels; combine all annotated graphic element training samples into a graphic element training set, and then send the graphic element training set to a parameter-initialized label recognition model for training to determine whether a first training condition is met. If so, output the trained label recognition model; otherwise, continue to train the label recognition model with the graphic element training set.

4. A method for managing a graphic element library for packaging design according to claim 3, characterized in that: The training of the feature extraction model includes the following steps: Acquire several sets of graphic element training samples, and sequentially send all the graphic element training samples to the pre-trained feature extraction model for processing, and output the second graphic feature vector corresponding to each graphic element training sample; perform unsupervised cluster analysis on all the second graphic feature vectors, construct several first clusters, and for each first cluster, select the cluster center of the first cluster to annotate all the graphic element training samples corresponding to the second graphic feature vectors in the first cluster; and form all the annotated graphic element training samples into a feature extraction training set; Retrain the pre-trained feature extraction model using the feature extraction training set to determine whether the second training condition is met, and if so, output the trained feature extraction model; Otherwise, the pre-trained feature extraction model is trained using the feature extraction training set.

5. A method for managing a graphic element library for packaging design according to claim 4, characterized in that: It also includes strengthening the generation operation of the first candidate graphic element set according to the search history of each designer. The specific steps are as follows: For each designer, the corresponding search history data is obtained; the first graphic feature vectors corresponding to all graphic elements in the search history data are selected for cluster analysis, a number of second clusters are output, and the cluster center of the second cluster containing the most first graphic feature vectors is selected as the target graphic element; Get an input label, match the input label with the graphic label corresponding to the graphic element in the graphic element library, and form the successfully matched graphic elements into an intermediate graphic element set, traverse the intermediate graphic element set, calculate the seventh similarity between the first graphic feature vector of the graphic element in the intermediate graphic element set and the first graphic feature vector of the target graphic element, sort the graphic elements in the intermediate graphic element set from large to small according to the seventh similarity, and select the first N graphic elements to combine and construct a selected graphic element set.

6. A method for managing a graphic element library for packaging design according to claim 5, characterized in that: The feature extraction layer and feature extraction model in the label recognition model are both set up based on convolutional neural networks.

7. A method for managing a graphic element library for packaging design according to claim 6, characterized in that: Cluster analysis used the k-means algorithm.

8. A graphic element library management system for packaging design, characterized in that: The system applies a graphic element library management method for packaging design as described in any one of claims 1 to 7, including: A graphic element acquisition module, used for acquiring graphic elements for packaging design; The label recognition module is used to send the graphic elements into the label recognition model for processing and output the graphic labels; convert all the graphic labels into label vectors through word embedding operations, and then concatenate all the label vectors to construct the first graphic feature vector; A feature extraction module, used for sending the graphic elements into the feature extraction model for processing and outputting a second graphic feature vector; A graphic element annotation module, used to annotate the graphic elements by using the graphic label, the first graphic feature vector and the second graphic feature vector, and add the annotated graphic elements to the graphic element library; A module for constructing a set of graphic elements to be selected is used to obtain an input label, match the input label with a graphic label corresponding to a graphic element in a graphic element library, and randomly select N graphic elements from the successfully matched graphic elements to form a set of graphic elements to be selected; A module for updating the set of graphic elements to be selected, used for updating the set of graphic elements to be selected in response to the operation feedback; Update the set of selected graphic elements, which includes two parts: When the designer performs the first operation, in response to the operation feedback, a first operation graphic element is determined, the first operation graphic element being the graphic element in the set of to-be-selected graphic elements on which the user performs the operation, all graphic elements in the graphic element library are traversed, and for each selected graphic element, a first similarity δ1 between a first graphic feature vector corresponding to the selected graphic element and a first graphic feature vector corresponding to the first operation graphic element is calculated, a second similarity δ2 between a second graphic feature vector corresponding to the selected graphic element and a second graphic feature vector corresponding to the first operation graphic element is calculated, and then a first comprehensive similarity A1=α1δ1+α2δ2 is calculated, wherein α1 and α2 are respectively the first weight coefficient and the second weight coefficient, and α1+α2=1 is satisfied; the graphic elements in the graphic element library are sorted from large to small according to the first comprehensive similarity A1, and the first N graphic elements are selected to be combined to replace the set of to-be-selected graphic elements; When the designer performs the second and subsequent operations, in response to the operation feedback, a second operation graphic element is determined, the second operation graphic element is the graphic element on which the user performs the operation in the new set of to-be-selected graphic elements, all graphic elements in the graphic element library are traversed, and for each selected graphic element, a third similarity δ3 between the first graphic feature vector corresponding to the second operation graphic element and the first graphic feature vector corresponding to the first operation graphic element is calculated, a fourth similarity δ4 between the second graphic feature vector corresponding to the second operation graphic element and the second graphic feature vector corresponding to the first operation graphic element is calculated, a fifth similarity δ5 between the first graphic feature vector corresponding to the selected graphic element and the first graphic feature vector corresponding to the second operation graphic element is calculated, a sixth similarity δ6 between the second graphic feature vector corresponding to the selected graphic element and the second graphic feature vector corresponding to the second operation graphic element is calculated, and then a second comprehensive similarity is calculated. , sort the graphic elements in the graphic element library from large to small according to the second comprehensive similarity A2, and select the first N graphic elements to combine and replace the set of graphic elements to be selected; and then replace the second operation graphic element with the first operation graphic element.

Images