Stroke-based extraction enhanced ancient text calligraphy style recommendation method and system

Abstract

The application provides a kind of based on stroke extraction enhanced ancient text calligraphy style recommendation method and system, the method comprises: obtaining target ancient text calligraphy image data;Extract the stroke image corresponding to each character in target ancient text calligraphy image data, obtain the stroke sequence parameters corresponding to stroke image according to preset brush touch model;Target ancient text calligraphy image data and the stroke sequence parameters corresponding to target ancient text calligraphy image data are input into style extraction model, and the ancient text calligraphy style features output by style extraction model are obtained;Determine ancient text calligraphy recommended style feature, and send the ancient text calligraphy image data corresponding to ancient text calligraphy recommended style feature to target user end, wherein, ancient text calligraphy recommended style feature is the ancient text calligraphy style feature determined based on the historical ancient text calligraphy browsing data of target user end.The application realizes the accurate recommendation of digital ancient text calligraphy.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for recommending the style of ancient Chinese inscriptions based on stroke extraction enhancement. Background Technology

[0002] Ancient inscriptions and rubbings are important materials for the study of traditional calligraphy. With the digitalization of public culture, the extraction and recommendation of digital features of ancient inscriptions and rubbings has become an important part of digital culture reading and dissemination. In particular, by calculating the style of ancient inscriptions and rubbings, we can achieve a more accurate digital description of them, which is conducive to the dissemination and research of public culture.

[0003] However, there is relatively little research on the style calculation of ancient inscriptions and rubbings. Most of them rely on static images of the text in the inscriptions and rubbings. Relying solely on static images of ancient inscriptions and rubbings can only obtain simple shape styles. It is difficult to obtain the writing process style, which is more crucial for ancient calligraphy. As a result, existing recommendation methods mainly rely on simple style features or simple attribute association strategies such as author and era when dealing with the recommendation problem of ancient inscriptions and rubbings, which cannot achieve accurate recommendation of ancient inscriptions and rubbings.

[0004] Therefore, there is an urgent need for a method and system for recommending ancient Chinese inscription styles based on stroke extraction enhancement to solve the above problems. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and system for recommending the style of ancient Chinese inscriptions based on stroke extraction enhancement.

[0006] This invention provides a method for recommending the style of ancient Chinese inscriptions based on stroke extraction enhancement, including:

[0007] Acquire the image data of the target ancient inscription;

[0008] Extract the stroke images corresponding to each character in the target ancient inscription image data, and obtain the stroke sequence parameters corresponding to the stroke images according to the preset stroke model;

[0009] The target ancient inscription image data and the corresponding brushstroke sequence parameters are input into the style extraction model to obtain the ancient inscription style features output by the style extraction model. The style extraction model is obtained by training a neural network model with sample ancient inscription image data labeled with ancient inscription style tags and the corresponding sample brushstroke sequence parameters.

[0010] Determine the ancient calligraphy recommended style features, and send the ancient calligraphy recommended style features corresponding to the ancient calligraphy image data to the target user end, wherein the ancient calligraphy recommended style features are the ancient calligraphy style features determined based on the historical ancient calligraphy browsing data of the target user end.

[0011] According to the ancient calligraphy style recommendation method based on stroke extraction enhancement provided by the application, the stroke image corresponding to each character in the target ancient calligraphy image data is extracted, which comprises:

[0012] Obtain the stroke template image data corresponding to the style of the target ancient calligraphy image data;

[0013] Based on the first encoder, the target ancient calligraphy image data and the stroke template image data are encoded to obtain the encoded target ancient calligraphy image data and the encoded stroke template image data, wherein the first encoder is constructed based on the first residual network;

[0014] The encoded target ancient calligraphy image data and the encoded stroke template image data are input into the character image stroke extraction model to obtain the stroke image corresponding to each character in the target ancient calligraphy image data, wherein the character image stroke extraction model is constructed by a convolutional long short-term memory neural network.

[0015] According to the ancient calligraphy style recommendation method based on stroke extraction enhancement provided by the application, the stroke image corresponding to each character in the target ancient calligraphy image data is extracted, which comprises:

[0016] The stroke image is subjected to image thinning processing to obtain the stroke thinning skeleton corresponding to the stroke image;

[0017] According to the writing form of the stroke thinning skeleton, a plurality of stroke positions are determined on the stroke thinning skeleton through the preset stroke model, and the stroke center position coordinate point information corresponding to each stroke position is obtained;

[0018] Obtain the stroke width data corresponding to each stroke position in the stroke thinning skeleton;

[0019] According to the stroke width data, the first distance data and the second distance data of each stroke position are obtained, wherein the first distance data represents the distance between the front end of the stroke and the stroke center in the stroke position, and the second distance data represents the distance between the rear end of the stroke and the stroke center in the stroke position;

[0020] According to the adjacent two stroke center position coordinate point information in the stroke thinning skeleton, the stroke direction corresponding to the stroke trajectory in the stroke image is obtained.

[0021] Based on the stroke direction, according to the stroke center position coordinate point information, the stroke width data, the first distance data and the second distance data, the stroke sequence parameters corresponding to the stroke image are constructed.

[0022] According to the ancient text and calligraphy style recommendation method based on stroke extraction enhancement provided by the application, the style extraction model is constructed by a multi-layer long short-term memory neural network and a second residual network, wherein the multi-layer long short-term memory neural network is used for feature extraction of the stroke sequence parameters to obtain stroke sequence style features; and the second residual network is used for feature extraction of the target ancient text and calligraphy image data to obtain image style features.

[0023] The target ancient text and calligraphy image data and the stroke sequence parameters corresponding to the target ancient text and calligraphy image data are input into the style extraction model to obtain the ancient text and calligraphy style features output by the style extraction model, which includes:

[0024] Based on the style extraction model, the stroke sequence style features and the image style features are weighted and summed according to the respective weights of the multi-layer long short-term memory neural network and the second residual network to obtain the ancient text and calligraphy style features.

[0025] According to the ancient text and calligraphy style recommendation method based on stroke extraction enhancement provided by the application, the target ancient text and calligraphy image data is obtained, which includes:

[0026] Based on the ancient text and calligraphy resource block, the ancient text and calligraphy image is obtained.

[0027] Based on the calligraphy character detection model, the text extraction is performed on the ancient text and calligraphy image to obtain single character images corresponding to each character in the ancient text and calligraphy image.

[0028] The single character images are subjected to character shape correction processing to obtain single character images after character shape correction processing.

[0029] The single character images after character shape correction processing are subjected to binaryzation processing to obtain the target ancient text and calligraphy image data.

[0030] According to the ancient text and calligraphy style recommendation method based on stroke extraction enhancement provided by the application, the method further includes:

[0031] According to the ancient text and calligraphy style features and the ancient text and calligraphy image data corresponding to the ancient text and calligraphy style features, a style feature database is constructed.

[0032] The determination of the ancient text and calligraphy recommendation style features includes:

[0033] obtaining a plurality of historical style features according to historical ancient prose and calligraphy browsing data of the target user terminal in the style feature database, wherein the historical style features are ancient prose and calligraphy style features corresponding to the historical ancient prose and calligraphy browsing data;

[0034] clustering the plurality of historical style features to obtain historical style feature clustering results, and determining a weight of each historical style feature according to a browsing time stamp corresponding to the historical style feature;

[0035] obtaining a recommendation weight of each cluster center according to the weight of the historical style feature of each cluster center in the historical style feature clustering results;

[0036] determining a target cluster center according to the recommendation weight of each cluster center, and determining an ancient prose and calligraphy style feature of the target cluster center as the ancient prose and calligraphy recommendation style feature.

[0037] The application further provides a system for recommending ancient prose and calligraphy style based on stroke extraction enhancement, comprising:

[0038] a preprocessing module configured to obtain target ancient prose and calligraphy image data;

[0039] a writing trajectory serialization module configured to extract stroke image corresponding to each character in the target ancient prose and calligraphy image data, and obtain stroke sequence parameters corresponding to the stroke image according to a preset stroke model;

[0040] a style extraction module configured to input the target ancient prose and calligraphy image data and the stroke sequence parameters corresponding to the target ancient prose and calligraphy image data into a style extraction model to obtain ancient prose and calligraphy style features output by the style extraction model, wherein the style extraction model is obtained by training a neural network model using sample ancient prose and calligraphy image data labeled with ancient prose and calligraphy style labels and sample stroke sequence parameters corresponding to the sample ancient prose and calligraphy image data;

[0041] an ancient prose and calligraphy data recommendation module configured to determine an ancient prose and calligraphy recommendation style feature, and send ancient prose and calligraphy image data corresponding to the ancient prose and calligraphy recommendation style feature to a target user terminal, wherein the ancient prose and calligraphy recommendation style feature is an ancient prose and calligraphy style feature determined based on historical ancient prose and calligraphy browsing data of the target user terminal.

[0042] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for recommending ancient prose and calligraphy style based on stroke extraction enhancement when executing the program.

[0043] The application also provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the stroke extraction enhanced ancient text and calligraphy style recommendation method according to any one of the above.

[0044] The application also provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the stroke extraction enhanced ancient text and calligraphy style recommendation method according to any one of the above.

[0045] The stroke extraction enhanced ancient text and calligraphy style recommendation method and system provided by the application, by extracting the stroke image corresponding to each character in the ancient text and calligraphy image data, then obtaining the stroke sequence parameters corresponding to the stroke image according to the preset stroke model, and then inputting the ancient text and calligraphy image data and the stroke sequence parameters into the style extraction model, obtaining the ancient text and calligraphy style features extracted by the style extraction model, and based on the historical ancient text and calligraphy browsing data of the user end, matching the ancient text and calligraphy recommendation style features from the extracted ancient text and calligraphy style features, and sending the ancient text and calligraphy image data corresponding to the ancient text and calligraphy recommendation style features to the user end, the accurate recommendation of the digital ancient text and calligraphy is realized. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0047] Figure 1 The flowchart of the stroke extraction enhanced ancient text and calligraphy style recommendation method provided by the application;

[0048] Figure 2 The schematic diagram of the character stroke image extraction process provided by the application;

[0049] Figure 3 The schematic diagram of the preset stroke model provided by the application;

[0050] Figure 4 The schematic diagram of the stroke extraction enhanced ancient text and calligraphy style recommendation system provided by the application;

[0051] Figure 5 The structure schematic diagram of the stroke extraction enhanced ancient text and calligraphy style recommendation system provided by the application;

[0052] Figure 6 The overall schematic diagram of the stroke extraction enhanced ancient text and calligraphy style recommendation system provided by the application;

[0053] Figure 7 The structural schematic diagram of the electronic device provided by the present application is shown. DETAILED DESCRIPTION

[0054] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0055] At present, the style calculation research on ancient text stele and calligraphy has not been fully carried out, is mainly limited to static stele and calligraphy text images, only depends on these static images, and only simple shape styles can be obtained, and the more critical writing process style of ancient text calligraphy cannot be understood in depth. Therefore, the existing recommendation method often only depends on simple style features or adopts the association strategy of simple attributes such as author and age when processing the ancient text stele and calligraphy recommendation problem, and it is difficult to realize accurate recommendation of ancient text stele and calligraphy, and there is certain limitation.

[0056] The present application proposes a kind of ancient text stele and calligraphy style calculation model and ancient text stele and calligraphy accurate recommendation method based on stroke extraction enhancement in view of existing problems. First, by Chinese stroke extraction method, stroke image is extracted according to stroke template order, and then the writing track of ancient text stele and calligraphy is restored based on preset stroke model;Then, using style extraction model, the writing style of ancient text stele and calligraphy is calculated using static image and serialized writing track of ancient text stele and calligraphy, and ancient text stele and calligraphy style database is constructed;Finally, the browsing history data of user terminal is used to realize the accurate recommendation of ancient text stele and calligraphy based on style calculation.

[0057] Figure 1 The flowchart of the ancient text stele and calligraphy style recommendation method based on stroke extraction enhancement provided by the present application is shown in Figure Figure 1 The present application provides an ancient text stele and calligraphy style recommendation method based on stroke extraction enhancement, comprising:

[0058] Step 101, obtain target ancient text stele and calligraphy image data.

[0059] In the present application, the required ancient text stele and calligraphy is first determined, including specific stele and calligraphy works, author, era and other information. Specifically, through various channels such as library, museum, online database, etc., the image data of ancient text stele and calligraphy is collected, and these images can be high-resolution digital images, photos or scanned images, so as to construct ancient text stele and calligraphy resource library through these image data of ancient text stele and calligraphy.

[0060] Further, based on the ancient text and epigraph resources, obtain the ancient text and epigraph image data required for style extraction, and in the present application, before the ancient text and epigraph image data is subjected to writing track serialization and style extraction, the ancient text and epigraph image data also needs to be preprocessed, so as to obtain the target ancient text and epigraph image data. On the basis of the above embodiment, the target ancient text and epigraph image data comprises:

[0061] Based on the ancient text and epigraph resources, obtain the ancient text and epigraph image;

[0062] Based on the calligraphy character detection model, extract the characters from the ancient text and epigraph image to obtain the single character image corresponding to each character in the ancient text and epigraph image;

[0063] Carry out the character shape correction processing on each single character image to obtain the single character image after the character shape correction processing;

[0064] Carry out the binary processing on the single character image after the character shape correction processing to obtain the target ancient text and epigraph image data.

[0065] In the present application, in the preprocessing process, mainly includes three processes of calligraphy character detection and recognition, calligraphy character shape correction and calligraphy character binary, that is, obtaining the single character image in the ancient text and epigraph, and carrying out the shape correction and binary processing, the specific process is as follows:

[0066] Firstly, based on the calligraphy detection model constructed by the SSD (Single Shot MultiBox Detector) target detection framework, the single character image extraction of the ancient text and epigraph characters is realized; at the same time, a calligraphy character shape correction and binary model is constructed to realize the recognition, correction and binary of single character, in the present application, the architecture of the calligraphy character shape correction and binary model adopts the symmetrical Encoder-Decoder structure, uses the intermediate feature output to identify and correct the results, and outputs the binary results through the Decoder, wherein the correction result outputs an affine matrix through the space translation network (Space Translation Net, STN).

[0067] In the preprocessing process provided by the present application, the ancient text and epigraph image is first input into the calligraphy detection model to obtain the single character image in the ancient text and epigraph image, and then the single character image is input into the calligraphy character shape correction and binary model to output the calligraphy single character image after the correction and binary. It should be noted that in the present application, since the subsequent writing track serialization process (i.e. calculating the stroke sequence parameters) depends on the template information, the input character image needs to have a very high recognition accuracy. Therefore, in the recognition result of the preprocessing of the present application, a classification score threshold (the classification result is normalized) is set, only the image data with a score higher than the threshold (the default is 0.95) will enter the subsequent processing.

[0068] In step 102, the stroke image corresponding to each character in the target ancient Chinese text and calligraphy image data is extracted, and the stroke sequence parameters corresponding to the stroke image are obtained according to a preset stroke model.

[0069] In the present application, the stroke sequence parameters corresponding to the stroke image are obtained by performing writing trajectory serialization processing on the stroke image corresponding to each character in the target ancient Chinese text and calligraphy image data. In the present application, the writing trajectory serialization processing includes two parts: character image stroke extraction and writing stroke trajectory construction. Specifically, the character image is disassembled into strokes in writing order through a stroke extraction algorithm, and then the serialized writing stroke trajectory is estimated through a writing stroke trajectory construction algorithm for the stroke sequence, so as to obtain the stroke sequence parameters corresponding to the stroke image.

[0070] In step 103, the target ancient Chinese text and calligraphy image data and the stroke sequence parameters corresponding to the target ancient Chinese text and calligraphy image data are input into a style extraction model to obtain the ancient Chinese text and calligraphy style features output by the style extraction model, wherein the style extraction model is obtained by training a neural network model using sample ancient Chinese text and calligraphy image data labeled with ancient Chinese text and calligraphy style labels and sample stroke sequence parameters corresponding to the sample ancient Chinese text and calligraphy image data.

[0071] In the present application, the style extraction model uses sample ancient Chinese text and calligraphy image data labeled with ancient Chinese text and calligraphy style labels in the training stage. These sample data not only contain the image itself, but also have sample stroke sequence parameters. The present application trains the neural network model with a large amount of sample data, and the model gradually learns how to identify and extract the style features of ancient Chinese text and calligraphy according to the input image data and corresponding stroke sequence parameters.

[0072] After the model is trained, the prepared target ancient Chinese text and calligraphy image data and corresponding stroke sequence parameters are input into the trained style extraction model. The style extraction model analyzes and processes the input data, and then outputs the style features of the ancient Chinese text and calligraphy. These features can be a set of numerical values, vectors or other forms of representation, and contain the unique style of the ancient Chinese text and calligraphy in terms of stroke, structure, layout, etc.

[0073] In step 104, the ancient Chinese text and calligraphy recommended style features are determined, and the ancient Chinese text and calligraphy image data corresponding to the ancient Chinese text and calligraphy recommended style features are sent to the target user end, wherein the ancient Chinese text and calligraphy recommended style features are the ancient Chinese text and calligraphy style features determined based on the historical ancient Chinese text and calligraphy browsing data of the target user end.

[0074] In the present application, the user terminal can be a mobile phone, a computer, etc., when the target user terminal browses ancient text calligraphy image data, the system will record the user's browsing behavior, including which calligraphy has been browsed, the duration and frequency of browsing, etc. These data are important basis for subsequent determination of recommended style features.

[0075] Further, according to the historical browsing data of the target user terminal, the ancient text calligraphy style features that the user may be interested in are analyzed, for example, which types of calligraphy does the user frequently browse, which calligraphy shows higher interest, etc. to infer the user's preferences and style preferences. Based on these analyses, the user's ancient text calligraphy recommended style features can be determined.

[0076] Further, according to the determined ancient text calligraphy recommended style features, matching is performed in the ancient text calligraphy style features extracted by the style extraction model in the early stage. Once the matching ancient text calligraphy image data is found, these data are sent to the target user terminal for the user to view and browse, improving the user's browsing experience.

[0077] The present application provides an ancient text calligraphy style recommendation method based on stroke extraction enhancement. By extracting the stroke image corresponding to each character in the ancient text calligraphy image data, and then obtaining the stroke sequence parameters corresponding to the stroke image according to the preset stroke model, the ancient text calligraphy image data and the stroke sequence parameters are input into the style extraction model to obtain the ancient text calligraphy style features extracted by the style extraction model. Based on the historical ancient text calligraphy browsing data of the user terminal, the ancient text calligraphy recommended style features are matched from the extracted ancient text calligraphy style features, and the ancient text calligraphy image data corresponding to the ancient text calligraphy recommended style features is sent to the user terminal, realizing the accurate recommendation of digital ancient text calligraphy.

[0078] On the basis of the above-mentioned embodiments, the extraction of the stroke image corresponding to each character in the target ancient text calligraphy image data comprises:

[0079] Obtaining the stroke template image data corresponding to the style of the target ancient text calligraphy image data;

[0080] Encoding the target ancient text calligraphy image data and the stroke template image data based on a first encoder to obtain encoded target ancient text calligraphy image data and encoded stroke template image data, wherein the first encoder is constructed based on a first residual network;

[0081] Inputting the encoded target ancient text calligraphy image data and the encoded stroke template image data into a character image stroke extraction model to obtain the stroke image corresponding to each character in the target ancient text calligraphy image data, wherein the character image stroke extraction model is constructed by a convolutional long short-term memory neural network.

[0082] Figure 2 The schematic diagram of the stroke image extraction process provided by the present application can refer to Figure 2 As shown in the figure, the stroke extraction algorithm structure is as shown in Figure 2 The stroke template image data serves as reference information in the stroke extraction process and contains binarized images of commonly used calligraphy characters and stroke information. Specifically, the to-be-extracted character image (i.e., the target ancient text calligraphy image data) and the stroke template image data (the binarized image therein) are encoded by an Encoder (i.e., a first encoder). In the present application, the first encoder is constructed based on a ResNet (first residual network) as a main framework structure.

[0083] Further, the stroke image of the to-be-extracted character image is extracted one by one based on the stroke information in the stroke template image data by using a convolution (Conv) long short-term memory neural network (LSTM).

[0084] On the basis of the above-mentioned embodiments, the stroke image corresponding to the stroke sequence parameter is obtained according to a preset stroke model, including:

[0085] The stroke image is subjected to image thinning processing to obtain a stroke thinning skeleton corresponding to the stroke image.

[0086] In the present application, the main structure of the stroke image is extracted through image thinning processing while maintaining its original shape. In this process, some non-key pixel points are removed, while important structural elements such as lines and edges are retained. For the stroke image, the main goal of thinning processing is to extract the skeleton of the stroke, i.e., the center line or the axis, which can represent the overall shape and structure of the stroke while removing redundant details and noise. Finally, the stroke thinning skeleton obtained after the stroke image is subjected to thinning processing can accurately reflect the main features and structure of the original stroke.

[0087] According to the writing form of the stroke thinning skeleton, a plurality of stroke positions are determined on the stroke thinning skeleton through the preset stroke model, and stroke center position coordinate point information corresponding to each stroke position is obtained;

[0088] Stroke width data corresponding to each stroke position at the stroke thinning skeleton is obtained.

[0089] According to the stroke width data, first distance data and second distance data of each stroke position are obtained, wherein the first distance data represents the distance between the front end of the stroke and the stroke center in the stroke position, and the second distance data represents the distance between the rear end of the stroke and the stroke center in the stroke position.

[0090] Based on the coordinate information of the center positions of two adjacent strokes in the stroke refinement skeleton, the stroke direction corresponding to the stroke trajectory in the stroke image is obtained;

[0091] Based on the stroke direction, and according to the coordinates of the center positions of each stroke, the stroke width data, the first distance data, and the second distance data, the stroke sequence parameters corresponding to the stroke image are constructed.

[0092] Figure 3 This is a schematic diagram of the preset brushstroke model provided by the present invention, which can be referred to. Figure 3 As shown, in this invention, Figure 3 (a) represents commonly used brushstrokes, serving as the preset brushstroke model in this invention. In constructing the brushstroke trajectory, to simplify the brushstroke model, [the model is used]. Figure 3 (b) uses a rhombus structure to simulate commonly used brushstrokes. The preset brushstroke model is described by five parameters, including the horizontal and vertical coordinates of the brushstroke center position, i.e., the coordinates of the brushstroke center position O(x, y); the single-sided width Ls (i.e., the brushstroke width data corresponding to the brushstroke position at the stroke thinning skeleton, which is half of the brushstroke width data); the distance Lt between the brushstroke front end and the brushstroke center O (i.e., the first distance data); and the distance Ld between the brushstroke rear end and the brushstroke center O (i.e., the second distance data).

[0093] Figure 4 This is a schematic diagram of the pen stroke trajectory construction process provided by the present invention, which can be referred to. Figure 4 As shown, in this invention, with Figure 4 To explain the writing form of strokes, first, the stroke image is refined. Then, 10 to 15 points are evenly extracted from the refined stroke skeleton as the center positions of the strokes (e.g., ...). Figure 4 (8 center points in the model). Further, based on the preset stroke model, after determining the stroke position, three more morphological parameters need to be determined to obtain complete stroke information. Specifically, in this invention, for each stroke position, the coordinate point information of the stroke center position is first obtained, and the stroke direction can be determined based on the previous and subsequent stroke positions; further, in order to improve accuracy, the stroke information of each stroke is calculated separately, and the distance Ls of the stroke character edge is obtained based on the vertical distance from one side of each stroke position to the stroke center point, that is, the stroke width data corresponding to the stroke thinning skeleton at the stroke position.

[0094] Further, according to the stroke width data corresponding to the stroke position at the stroke refinement skeleton, the first distance data and the second distance data can be estimated, preferably, in the present application, through the analysis of the stroke refinement skeleton and the actual stroke parameters in advance, it is known that the stroke width data corresponding to the stroke position at the stroke refinement skeleton and the first distance and the second distance satisfy a certain conversion relationship, so that the first distance and the second distance can be quickly estimated according to the conversion relationship, that is, the single-side width Ls is approximately equal to half of the first distance data Lt, and the single-side width Ls is almost equal to the second distance data Ld, so that the stroke parameters of each stroke position are obtained in sequence based on the writing trajectory of the stroke, and finally the stroke sequence parameters corresponding to the stroke image are constructed according to the stroke parameters. Referring to Figure 4 As shown in the stroke in Figure 4 , the writing direction is from left to right, and the stroke sequence parameters are composed of the stroke parameters of the leftmost stroke center position to the stroke parameters of the rightmost stroke center position. It should be noted that in the present application, the first distance data and the second distance data calculated by the above conversion relationship can be adjusted according to the actual situation to make the calculated distance data more accurate.

[0095] On the basis of the above embodiment, the style extraction model is constructed by a multi-layer long short-term memory neural network and a second residual network, wherein the multi-layer long short-term memory neural network is used for feature extraction of the stroke sequence parameters to obtain stroke sequence style features; the second residual network is used for feature extraction of the target ancient text calligraphy image data to obtain image style features;

[0096] The target ancient text calligraphy image data and the stroke sequence parameters corresponding to the target ancient text calligraphy image data are input into the style extraction model to obtain the ancient text calligraphy style features output by the style extraction model, which comprises:

[0097] Based on the style extraction model, the stroke sequence style features and the image style features are weighted and summed according to the respective weights of the multi-layer long short-term memory neural network and the second residual network to obtain the ancient text calligraphy style features.

[0098] In the present application, the style extraction module is used to extract the style feature data of ancient text calligraphy, in order to obtain more complete feature data, the target ancient text calligraphy image data and the stroke sequence parameters corresponding to the target ancient text calligraphy image data are input into the style extraction model.

[0099] Specifically, the style extraction model is composed of a multi-layer LSTM network and a ResNet network (a second residual network), wherein the multi-layer LSTM network is used to extract style features of brush stroke sequence parameters, the ResNet network is used to extract style features of target ancient text calligraphy image data, and in style extraction, weights of 0.7 and 0.3 are respectively applied for summation to obtain final style features. In the application, based on an ancient text calligraphy resource library, style features of all calligraphy image resources are extracted by the style extraction model and stored in a style feature database for subsequent ancient text calligraphy data recommendation.

[0100] On the basis of the above-mentioned embodiments, the method further comprises:

[0101] According to the ancient text calligraphy style features and the ancient text calligraphy image data corresponding to the ancient text calligraphy style features, a style feature database is constructed;

[0102] The determination of the ancient text calligraphy recommendation style feature comprises:

[0103] According to historical ancient text calligraphy browsing data of the target user terminal in the style feature database, a plurality of historical style features are obtained, wherein the historical style features are the ancient text calligraphy style features corresponding to the historical ancient text calligraphy browsing data;

[0104] The plurality of historical style features are clustered to obtain a historical style feature clustering result, and according to browsing time stamps corresponding to each of the historical style features, weights of each of the historical style features are determined;

[0105] According to the weights of the historical style features of each of the clustering centers in the historical style feature clustering result, recommendation weights of each of the clustering centers are obtained;

[0106] According to the recommendation weights of each of the clustering centers, a target clustering center is determined, and an ancient text calligraphy style feature of the target clustering center is determined as the ancient text calligraphy recommendation style feature.

[0107] In the present application, based on the historical ancient text and epigraph data of the target user terminal, new ancient text and epigraph data is accurately recommended to the user terminal. Specifically, according to the user browsing history data, the style feature data of the historical ancient text and epigraph is obtained from the style feature library as the historical style feature. The present application is aimed at the user's browsing around multiple interest points, clustering based on the historical style feature to obtain multiple cluster centers; at the same time, the browsing data is weighted according to the browsing timestamp, that is, the longer the browsing time is, the smaller the weight is; further, according to the clustering result, the weight sum of each cluster center is calculated and normalized to obtain the recommendation weight of each cluster center; finally, the style feature similar to each cluster center is searched in the style feature database to determine one or more target cluster centers, and the recommendation quantity of the corresponding style feature of each target cluster center is determined according to the size of the recommendation weight, and the corresponding ancient text and epigraph data is recommended to the target user terminal.

[0108] The ancient text and epigraph style recommendation system based on stroke extraction enhancement provided by the present application is described below, and the ancient text and epigraph style recommendation system based on stroke extraction enhancement described below can be correspondingly referred to the ancient text and epigraph style recommendation method based on stroke extraction enhancement described above.

[0109] Figure 5 The structure diagram of the ancient text and epigraph style recommendation system based on stroke extraction enhancement provided by the present application is shown in Figure 5 The present application provides an ancient text and epigraph style recommendation system based on stroke extraction enhancement, which comprises a preprocessing module 501, a writing trajectory serialization module 502, a style extraction module 503 and an ancient text and epigraph data recommendation module 504. The preprocessing module 501 is used to obtain target ancient text and epigraph image data; the writing trajectory serialization module 502 is used to extract the stroke image corresponding to each character in the target ancient text and epigraph image data, and obtain the stroke sequence parameters corresponding to the stroke image according to a preset stroke model; the style extraction module 503 is used to input the target ancient text and epigraph image data and the stroke sequence parameters corresponding to the target ancient text and epigraph image data into a style extraction model to obtain the ancient text and epigraph style feature output by the style extraction model, wherein the style extraction model is obtained by training a neural network model with sample ancient text and epigraph image data labeled with ancient text and epigraph style labels and sample stroke sequence parameters corresponding to the sample ancient text and epigraph image data; the ancient text and epigraph data recommendation module 504 is used to determine ancient text and epigraph recommendation style features and send the ancient text and epigraph image data corresponding to the ancient text and epigraph recommendation style features to a target user terminal, wherein the ancient text and epigraph recommendation style features are ancient text and epigraph style features determined based on the historical ancient text and epigraph browsing data of the target user terminal.

[0110] In the present application, first, the required ancient calligraphy is determined, including specific calligraphy works, author, era and other information. Specifically, through various channels such as libraries, museums, online databases, etc., image data of ancient calligraphy is collected, which can be high-resolution digital images, photos or scanned images, so as to construct an ancient calligraphy resource library through the image data of the ancient calligraphy.

[0111] Further, based on the ancient calligraphy resource library, the preprocessing module 501 obtains the ancient calligraphy image data required for style extraction this time, and in the present application, the preprocessing module 501 also needs to preprocess the ancient calligraphy image data, so as to obtain target ancient calligraphy image data. On the basis of the above embodiment, the target ancient calligraphy image data includes:

[0112] Figure 6 The overall schematic diagram of the ancient calligraphy style recommendation system based on stroke extraction enhancement provided by the present application is shown in Figure 6 As shown, first, the preprocessing module 501 realizes single character image extraction of ancient calligraphy text based on the calligraphy detection model constructed by the SSD target detection framework; at the same time, a calligraphy character shape correction and binarization model is constructed to realize single character recognition, correction and binarization. In the present application, the architecture of the calligraphy character shape correction and binarization model adopts a symmetrical Encoder-Decoder structure, uses intermediate feature output to identify and correct the results, and outputs the binarization results through the Decoder, wherein the correction result outputs an affine matrix through a spatial transformation network.

[0113] Further, the preprocessing module 501 obtains single character images in the ancient calligraphy image through the calligraphy detection model, and then inputs these single character images into the calligraphy character shape correction and binarization model to output corrected and binarized calligraphy single character images. It should be noted that in the present application, since the subsequent writing trajectory serialization process (i.e. calculating stroke sequence parameters) depends on template information, the input text image needs to have very high recognition accuracy. Therefore, in the recognition result of the preprocessing, a classification score threshold (the classification result is normalized) is set, only the image data with a score exceeding the threshold (the default is 0.95) will enter the subsequent processing.

[0114] In the present application, the writing trajectory serialization module 502 obtains the stroke sequence parameters corresponding to the stroke image by performing writing trajectory serialization processing on the strokes image corresponding to each character in the target ancient text calligraphy image data. The writing trajectory serialization processing includes two parts: stroke image extraction and stroke trajectory construction. Specifically, the writing trajectory serialization module 502 extracts strokes from the character image in writing order through a stroke extraction algorithm, and then estimates the serialized stroke trajectory through a stroke trajectory construction algorithm for the stroke sequence, to obtain the stroke sequence parameters corresponding to the stroke image.

[0115] Further, the style extraction module 503 extracts the ancient text style through the pre-trained style extraction model. The style extraction model uses sample ancient text calligraphy image data labeled with ancient text calligraphy style tags during the training stage. These sample data not only include the image itself, but also have corresponding sample stroke sequence parameters. The present application trains the neural network model with a large amount of sample data, and the model gradually learns how to identify and extract the style features of ancient text calligraphy according to the input image data and corresponding stroke sequence parameters. After the model is trained, the style extraction module 503 inputs the prepared target ancient text calligraphy image data and corresponding stroke sequence parameters into the trained style extraction model. The style extraction model analyzes and processes the input data, and then outputs the style features of the ancient text calligraphy. These features can be a set of numerical values, vectors or other forms of representation, and contain the unique style of ancient text calligraphy in terms of stroke, structure, layout, etc.

[0116] In the present application, the user terminal can be a mobile phone, a computer, etc. When the target user terminal browses the ancient text calligraphy image data, the system will record the user's browsing behavior, including which calligraphy has been browsed, the duration and frequency of browsing, etc. These data are important basis for subsequent determination of recommended style features. Further, the ancient text calligraphy data recommendation module 504 analyzes the ancient text calligraphy style features that the user may be interested in based on the historical browsing data of the target user terminal, such as which types of calligraphy the user frequently browses, which calligraphy the user shows high interest in, etc., to infer the user's preferences and style preferences. Based on these analyses, the user's ancient text calligraphy recommended style features can be determined. Finally, the ancient text calligraphy data recommendation module 504 matches the determined ancient text calligraphy recommended style features with the ancient text calligraphy style features extracted by the style extraction model in the early stage. Once the matching ancient text calligraphy image data is found, these data are sent to the target user terminal for the user to view and browse, improving the user's browsing experience.

[0117] The present invention provides a style recommendation system for ancient Chinese inscriptions based on stroke extraction enhancement. This system extracts stroke images corresponding to each character from the image data of ancient Chinese inscriptions, obtains stroke sequence parameters corresponding to each stroke image according to a preset stroke model, and then inputs the ancient Chinese inscription image data and stroke sequence parameters into a style extraction model to obtain style features of the ancient Chinese inscriptions extracted by the style extraction model. Based on the user's historical browsing data of ancient Chinese inscriptions, the system matches the extracted style features to a recommended style feature of the ancient Chinese inscriptions and sends the image data of the ancient Chinese inscription corresponding to the recommended style feature to the user, thus achieving accurate recommendation of digital ancient Chinese inscriptions.

[0118] The system provided by this invention is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0119] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 7 As shown, the electronic device may include: a processor 701, a communication interface 702, a memory 703, and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704. The processor 701 can call logical instructions in the memory 703 to execute a method for recommending ancient Chinese inscription styles based on stroke extraction enhancement. This method includes: acquiring target ancient Chinese inscription image data; extracting stroke images corresponding to each character in the target ancient Chinese inscription image data, and obtaining stroke sequence parameters corresponding to the stroke images according to a preset stroke model; inputting the target ancient Chinese inscription image data and the corresponding stroke sequence parameters into a style extraction model to obtain ancient Chinese inscription style features output by the style extraction model, wherein the style extraction model is obtained by training a neural network model with sample ancient Chinese inscription image data labeled with ancient Chinese inscription style tags and the corresponding sample stroke sequence parameters; determining recommended ancient Chinese inscription style features, and sending the ancient Chinese inscription image data corresponding to the recommended ancient Chinese inscription style features to a target user terminal, wherein the recommended ancient Chinese inscription style features are determined based on historical ancient Chinese inscription browsing data of the target user terminal.

[0120] Further, the logic instructions in the memory 703 described above can be implemented in the form of software functional units and sold or used as standalone products, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or partially contribute to the prior art, or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0121] In another aspect, the present application also provides a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, when the program instructions are executed by a computer, the computer can execute the stroke extraction enhanced ancient text and calligraphy style recommendation method provided by the above-mentioned method, which comprises: obtaining target ancient text and calligraphy image data; extracting the stroke image corresponding to each character in the target ancient text and calligraphy image data, and obtaining the stroke sequence parameters corresponding to the stroke image according to a preset stroke model; inputting the target ancient text and calligraphy image data and the stroke sequence parameters corresponding to the target ancient text and calligraphy image data into a style extraction model to obtain the ancient text and calligraphy style features output by the style extraction model, wherein the style extraction model is obtained by training a neural network model based on sample ancient text and calligraphy image data marked with ancient text and calligraphy style labels and sample stroke sequence parameters corresponding to the sample ancient text and calligraphy image data; determining ancient text and calligraphy recommendation style features, and sending the ancient text and calligraphy image data corresponding to the ancient text and calligraphy recommendation style features to a target user terminal, wherein the ancient text and calligraphy recommendation style features are ancient text and calligraphy style features determined based on historical ancient text and calligraphy browsing data of the target user terminal.

[0122] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a method for recommending a style of ancient Chinese calligraphy based on stroke extraction enhancement as provided by any of the above embodiments, the method comprising: obtaining target ancient Chinese calligraphy image data; extracting stroke images corresponding to each character in the target ancient Chinese calligraphy image data, and obtaining stroke sequence parameters corresponding to the stroke images according to a preset stroke model; inputting the target ancient Chinese calligraphy image data and the stroke sequence parameters corresponding to the target ancient Chinese calligraphy image data into a style extraction model to obtain ancient Chinese calligraphy style features output by the style extraction model, wherein the style extraction model is obtained by training a neural network model using sample ancient Chinese calligraphy image data labeled with ancient Chinese calligraphy style labels and sample stroke sequence parameters corresponding to the sample ancient Chinese calligraphy image data; determining an ancient Chinese calligraphy recommended style feature, and sending ancient Chinese calligraphy image data corresponding to the ancient Chinese calligraphy recommended style feature to a target user terminal, wherein the ancient Chinese calligraphy recommended style feature is an ancient Chinese calligraphy style feature determined based on historical ancient Chinese calligraphy browsing data of the target user terminal.

[0123] The apparatus embodiments described above are merely illustrative, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0124] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary universal hardware platforms, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0125] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for recommending the style of ancient Chinese inscriptions based on stroke extraction enhancement, characterized in that, include: Acquire the image data of the target ancient inscription; Extract the stroke images corresponding to each character in the target ancient inscription image data, and obtain the stroke sequence parameters corresponding to the stroke images according to the preset stroke model; The target ancient inscription image data and the corresponding brushstroke sequence parameters are input into the style extraction model to obtain the ancient inscription style features output by the style extraction model. The style extraction model is obtained by training a neural network model with sample ancient inscription image data labeled with ancient inscription style tags and the corresponding sample brushstroke sequence parameters. Determine the recommended style features of ancient inscriptions and send the image data of ancient inscriptions corresponding to the recommended style features to the target user terminal. The recommended style features of ancient inscriptions are determined based on the historical browsing data of ancient inscriptions on the target user terminal. The step of obtaining the stroke sequence parameters corresponding to the stroke image according to the preset stroke model includes: The stroke image is thinned to obtain the thinned skeleton of the stroke image. Based on the writing form of the stroke refinement skeleton, multiple stroke positions are determined on the stroke refinement skeleton using the preset stroke model, and the coordinate point information of the stroke center position corresponding to each stroke position is obtained. Obtain the stroke width data corresponding to each stroke position at the stroke refinement skeleton; Based on the stroke width data, first distance data and second distance data are obtained for each stroke position, wherein the first distance data represents the distance between the tip of the stroke and the center of the stroke at the stroke position, and the second distance data represents the distance between the back end of the stroke and the center of the stroke at the stroke position; Based on the coordinate information of the center positions of two adjacent strokes in the stroke refinement skeleton, the stroke direction corresponding to the stroke trajectory in the stroke image is obtained; Based on the stroke direction, and according to the coordinates of the center positions of each stroke, the stroke width data, the first distance data, and the second distance data, the stroke sequence parameters corresponding to the stroke image are constructed.

2. The method for recommending ancient Chinese inscription styles based on stroke extraction enhancement according to claim 1, characterized in that, The step of extracting the stroke images corresponding to each character in the target ancient inscription image data includes: Obtain the stroke template image data of the style corresponding to the target ancient inscription image data; Based on the first encoder, the target ancient inscription image data and the stroke template image data are encoded to obtain the encoded target ancient inscription image data and the encoded stroke template image data, wherein the first encoder is constructed based on the first residual network; The encoded target ancient inscription image data and the encoded stroke template image data are input into the text image stroke extraction model to obtain the stroke images corresponding to each character in the target ancient inscription image data. The text image stroke extraction model is constructed by a convolutional long short-term memory neural network.

3. The method for recommending ancient Chinese inscription styles based on stroke extraction enhancement according to claim 1, characterized in that, The style extraction model is constructed from a multilayer long short-term memory neural network and a second residual network. The multilayer long short-term memory neural network is used to extract features from the brushstroke sequence parameters to obtain brushstroke sequence style features; the second residual network is used to extract features from the target ancient inscription image data to obtain image style features. The step of inputting the target ancient inscription image data and the corresponding brushstroke sequence parameters into the style extraction model to obtain the ancient inscription style features output by the style extraction model includes: Based on the style extraction model, the brushstroke sequence style features and the image style features are weighted and summed according to the weights of the multilayer long short-term memory neural network and the second residual network to obtain the style features of the ancient inscription.

4. The method for recommending ancient Chinese inscription styles based on stroke extraction enhancement according to claim 1, characterized in that, The acquisition of the target ancient inscription image data includes: Based on ancient text rubbings resource blocks, obtain images of ancient text rubbings; Based on the calligraphy character detection model, the ancient inscription image is used to extract characters, and the single character image corresponding to each character in the ancient inscription image is obtained. Each of the individual character images is subjected to character shape correction processing to obtain the character image after character shape correction processing; The single-character image after the character shape correction process is binarized to obtain the target ancient inscription image data.

5. The method for recommending ancient Chinese inscription styles based on stroke extraction enhancement according to claim 1, characterized in that, The method further includes: A style feature database is constructed based on the stylistic features of the ancient inscriptions and the corresponding image data of the ancient inscriptions. The determination of recommended stylistic features for ancient inscriptions and rubbings includes: Based on the historical ancient text inscription browsing data of the target user terminal in the style feature database, multiple historical style features are obtained, wherein the historical style features are the ancient text inscription style features corresponding to the historical ancient text inscription browsing data; Clustering is performed on multiple historical style features to obtain historical style feature clustering results, and the weight of each historical style feature is determined according to the browsing timestamp corresponding to each historical style feature. Based on the weights of the historical style features of each cluster center in the historical style feature clustering results, the recommendation weights of each cluster center are obtained. Based on the recommendation weights of each cluster center, a target cluster center is determined, and the ancient inscription style features of the target cluster center are determined as the recommended style features of the ancient inscriptions.

6. A style recommendation system for ancient Chinese inscriptions based on stroke extraction enhancement, characterized in that, include: The preprocessing module is used to acquire the target ancient inscription image data; The writing trajectory serialization module is used to extract the stroke images corresponding to each character in the target ancient inscription image data, and obtain the stroke sequence parameters corresponding to the stroke images according to the preset stroke model. The style extraction module is used to input the target ancient inscription image data and the brushstroke sequence parameters corresponding to the target ancient inscription image data into the style extraction model to obtain the ancient inscription style features output by the style extraction model. The style extraction model is obtained by training a neural network model with sample ancient inscription image data labeled with ancient inscription style tags and sample brushstroke sequence parameters corresponding to the sample ancient inscription image data. The ancient text inscription data recommendation module is used to determine the recommended style features of ancient text inscriptions and send the ancient text inscription image data corresponding to the recommended style features to the target user terminal. The recommended style features of ancient text inscriptions are determined based on the historical ancient text inscription browsing data of the target user terminal. The writing trajectory serialization module is specifically used for: The stroke image is thinned to obtain the thinned skeleton of the stroke image. Based on the writing form of the stroke refinement skeleton, multiple stroke positions are determined on the stroke refinement skeleton using the preset stroke model, and the coordinate point information of the stroke center position corresponding to each stroke position is obtained. Obtain the stroke width data corresponding to each stroke position at the stroke refinement skeleton; Based on the stroke width data, first distance data and second distance data are obtained for each stroke position, wherein the first distance data represents the distance between the tip of the stroke and the center of the stroke at the stroke position, and the second distance data represents the distance between the back end of the stroke and the center of the stroke at the stroke position; Based on the coordinate information of the center positions of two adjacent strokes in the stroke refinement skeleton, the stroke direction corresponding to the stroke trajectory in the stroke image is obtained; Based on the stroke direction, and according to the coordinates of the center positions of each stroke, the stroke width data, the first distance data, and the second distance data, the stroke sequence parameters corresponding to the stroke image are constructed.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the ancient inscription style recommendation method based on stroke extraction enhancement as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the ancient inscription style recommendation method based on stroke extraction enhancement as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the ancient inscription style recommendation method based on stroke extraction enhancement as described in any one of claims 1 to 5.

Images