A method and device for generating word cloud images, and a computer-readable storage medium

Abstract

The embodiment of the present invention discloses a method and device for generating a word cloud image, and a storage medium, including: obtaining a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element; based on the first feature map and the second feature map, processing is performed through a target detection special layer to determine a fixed-dimensional feature vector; based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image; coloring is performed on the preliminary word cloud image to determine the generated word cloud image. In the above scheme, the preliminary word cloud image is generated through a preset convolutional neural network and a preset regression prediction network, thereby improving the efficiency of word cloud image generation.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a word cloud image generation method and device, and a computer-readable storage medium. Background Art

[0002] In the field of graphic design, there is a design of word cloud images. In addition to displaying a large amount of text data so that readers can quickly grasp the key points, word cloud images have good artistic effects. When placed in an article, they can quickly attract readers' attention. However, there are a large number of words in the word cloud. If each word is manually typeset and colored one by one, it will be time-consuming and labor-intensive. The existing technology establishes multi-level boxes for each word's graphic element, performs overlap detection between graphic elements, keeps a certain distance between graphic elements, and determines the size and position of the text according to the importance of the text. Layer-by-layer overlap detection between graphic elements is time-consuming and inefficient, and the filling effect is poor under some special background shape area conditions. Summary of the invention

[0003] The embodiments of the present invention provide a word cloud image generation method and device, and a computer-readable storage medium, which can solve the problems of low efficiency in word cloud image generation and poor filling effect.

[0004] The technical solution of the present invention is achieved in this way:

[0005] The embodiment of the present invention provides a method for generating a word cloud image, which is characterized by comprising:

[0006] Acquire a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element;

[0007] Based on the first foreground image element and the first background image element, feature extraction is performed by a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element;

[0008] Based on the first feature map and the second feature map, processing is performed through a target detection special layer to determine a fixed-dimensional feature vector;

[0009] Based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image;

[0010] The preliminary word cloud image is colorized to determine a generated word cloud image.

[0011] In the above scheme, before extracting features based on the foreground image element and the background image element by a preset convolutional neural network to obtain feature maps corresponding to the foreground image element and the background image element respectively, the method further includes:

[0012] Acquire multiple training images, wherein each training image includes multiple foreground image elements and training background image elements;

[0013] Based on the multiple foreground primitives, determining a training foreground primitive of a current primitive type through type selection processing;

[0014] Based on the training foreground primitive and the training background primitive, determining the parameters of the training foreground primitive by convolution scanning;

[0015] Based on the parameters of the training foreground primitives, initial word cloud images corresponding to the plurality of training images are obtained, and qualified images are screened out as training sample data;

[0016] Based on the training sample data, the initial convolutional neural network is trained to obtain the preset convolutional neural network.

[0017] In the above solution, the step of determining the parameters of the training foreground primitives by convolution scanning based on the training foreground primitives and the training background primitives includes:

[0018] Based on the training foreground image element, a first binary image of the training foreground image element is obtained through binarization processing;

[0019] Based on the training background image element, a second binary image of the training background image element is obtained through binarization processing.

[0020] Based on the first binarized image and the second binarized image, parameters of the training foreground primitive are determined.

[0021] In the above solution, the first binary image of the training foreground image element is obtained by binarization processing based on the training foreground image element, including:

[0022] Based on the training foreground primitive, obtaining multi-size training foreground sub-primitives through scaling processing;

[0023] Based on each foreground sub-pixel of the multi-size training foreground sub-pixel, obtain a multi-angle training foreground sub-pixel of each foreground sub-pixel through rotation processing;

[0024] Based on the multi-angle training foreground sub-pixel, the first binarized image of the training foreground sub-pixel is obtained through binarization processing.

[0025] In the above solution, the step of determining the parameters of the training foreground primitives by convolution scanning based on the training foreground primitives and the training background primitives includes:

[0026] Based on the training foreground image element and the training background image element, a third binarized image of the training foreground image element and a fourth binarized image of the training background image element are obtained through binarization processing;

[0027] Performing convolution scanning on the third binarized image on the fourth binarized image to obtain a product value at each step;

[0028] Based on the product value and the third binarized image, parameters of the training foreground primitive are determined.

[0029] In the above solution, determining the parameters of the training foreground primitive based on the first binarized image and the second binarized image includes:

[0030] Based on the second binary image, a preset blank detection algorithm is used to perform calculations to determine a blank center position of the second binary image;

[0031] Based on the blank center positions of the first binarized sub-image and the second binarized image, parameters of the first foreground primitive are determined through convolution scanning processing.

[0032] In the above solution, the method of determining the blank center position of the second binary image by performing calculations based on the second binary image through a preset blank detection algorithm includes:

[0033] Based on the second binarized image, a blank area is selected as an initial scanning position;

[0034] Based on the scanning initial position and the preset field strength calculation criterion, performing calculation to determine the field strength of each cell in the second binary image;

[0035] Based on the field intensity of each cell in the second binarized image, a calculation is performed using a preset gradient calculation criterion to obtain the field intensity gradient of each cell;

[0036] Based on the field intensity gradient of each cell, an iterative operation is performed until a maximum value of the field intensity is obtained, and a cell corresponding to the maximum value of the field intensity is determined;

[0037] The cell corresponding to the maximum value of the field intensity is taken as the blank center position.

[0038] In the above solution, the determining the first foreground image primitive by type selection processing based on the multiple foreground image primitives includes:

[0039] Based on the multiple foreground primitives, data management is performed by establishing a quadtree to obtain a data management structure;

[0040] Based on the data management structure, searching is performed to determine adjacent graphics elements of the current foreground graphics element;

[0041] Determining the type of the current foreground image element according to the type of the adjacent image element;

[0042] The first foreground primitive is determined based on the type of the current foreground primitive.

[0043] In the above solution, the first feature map and the second feature map are processed by a target detection special layer to determine a fixed-dimensional feature vector, including:

[0044] Based on the second feature map of the first background image element, determining a local feature map of the second feature map through feature extraction processing;

[0045] Based on the local feature maps of the first feature map and the second feature map, processing is performed through the target detection special layer to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map;

[0046] The fixed-dimensional feature vector is obtained by fusion processing based on the first fixed-dimensional vector and the second fixed-dimensional vector.

[0047] In the above solution, the second feature map based on the first background image element is determined by feature extraction to obtain a local feature map of the second feature map, including:

[0048] Based on the second feature map of the first background image element, a preset blank detection algorithm is used to detect and obtain a blank center position of the second feature map;

[0049] Based on the second feature map and the blank center position of the second feature map, a surround scanning process is performed to obtain feature points of the second feature map;

[0050] Based on the feature points of the second feature map, the local feature map of the second feature map is obtained by performing window clipping processing.

[0051] In the above scheme, the local feature map based on the first feature map and the second feature map is processed by the target detection special layer to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map, including:

[0052] Based on the first feature map and the local feature map of the second feature map, a first fixed feature point of the first feature map and a second fixed feature point of the local feature map of the second feature map are obtained by processing through a bilinear interpolation method;

[0053] Based on the first fixed feature points and the second fixed feature points, the first fixed dimensional vector of the first feature map and the second fixed dimensional vector of the second feature map are determined.

[0054] In the above scheme, based on the fixed-dimensional feature vector, the preset regression prediction network is used to perform predictive typesetting to obtain a preliminary word cloud image, including:

[0055] Based on the fixed-dimensional feature vector, a preset regression prediction network is used to process the fixed-dimensional feature vector to obtain regression parameters, wherein the regression parameters include position parameters, size parameters and angle parameters.

[0056] Based on the angle parameter, obtaining a predicted angle of the first foreground primitive;

[0057] Based on the regression parameters, a regression calculation is performed to obtain a predicted position and size of the first foreground primitive;

[0058] Based on the predicted angle of the first foreground image element, the predicted position and size of the first foreground image element, typesetting is performed to obtain the preliminary word cloud image.

[0059] In the above solution, the coloring process is performed on the preliminary word cloud image to determine the generated word cloud image, including:

[0060] Based on the word cloud image to be generated, spatial color clustering is performed to establish a feature vector;

[0061] Perform clustering based on the feature vector of the word cloud image to be generated to obtain a color center point;

[0062] Based on the color center point and the preliminary word cloud image, matching is performed by the nearest matching principle to obtain a matching result;

[0063] Based on the matching result, the preliminary word cloud image is colored to obtain the word cloud image.

[0064] In the above solution, the preliminary word cloud image includes a plurality of word cloud primitives;

[0065] The matching result is obtained by matching the color center point and the preliminary word cloud image according to the nearest matching principle, including:

[0066] Based on the color center point and the multiple word cloud primitives, matching is performed by the nearest matching principle, and if the word cloud primitive matches a unique color center, a matching result is obtained;

[0067] Based on the color center point and the multiple word cloud primitives, matching is performed using a nearest matching principle. If the word cloud primitive matches at least two color centers, a matching result is determined using a maximum weight matching algorithm.

[0068] The embodiment of the present invention provides a word cloud image generation device, which includes an acquisition unit and a determination unit; wherein:

[0069] The acquisition unit is used to acquire a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element;

[0070] The determining unit is used to determine a fixed-dimensional feature vector by processing the first feature map and the second feature map through a target detection special layer;

[0071] The acquisition unit is used to perform predictive typesetting based on the fixed-dimensional feature vector through a preset regression prediction network to obtain a preliminary word cloud image;

[0072] The determining unit is used to perform coloring processing on the preliminary word cloud image to determine the generated word cloud image.

[0073] The embodiment of the present invention provides a word cloud image generating device, the word cloud image generating device comprising:

[0074] A memory for storing executable instructions;

[0075] The processor is used to execute the executable instructions stored in the memory. When the executable instructions are executed, the processor executes the word cloud image generation method.

[0076] An embodiment of the present invention provides a computer-readable storage medium, characterized in that the storage medium stores executable instructions, which, when executed, are used to cause a processor to execute the word cloud image generation method as described in the embodiment of the present invention.

[0077] The embodiment of the present invention provides a method and device for generating a word cloud image, and a computer-readable storage medium, wherein the method includes: obtaining a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element; based on the first feature map and the second feature map, processing is performed through a target detection special layer to determine a fixed-dimensional feature vector; based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image; coloring is performed on the preliminary word cloud image to determine the generated word cloud image. In the above scheme, the preliminary word cloud image is generated through a preset convolutional neural network and a preset regression prediction network, which improves the efficiency of word cloud image generation, and the background image element and the foreground image element are used to simultaneously generate the word cloud image, which enhances the positioning capability of the foreground image element in the background and improves the filling effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 1 ;

[0079] Figure 2 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 2 ;

[0080] Figure 3 An optional window size diagram of a word cloud image generation method is provided for an embodiment of the invention;

[0081] Figure 4 An optional schematic diagram of generating fixed feature points of a word cloud image generation method is provided for an embodiment of the present invention;

[0082] Figure 5 An optional bilinear interpolation processing diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0083] Figure 6 An optional preset regression prediction network architecture diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0084] Figure 7 An optional prediction effect diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0085] Figure 8 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 3 ;

[0086] Fig. 9 An optional color center distribution diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0087] Fig.10 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 4 ;

[0088] Fig.11 An optional quadtree management schematic diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0089] Fig.12 An optional data management structure diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0090] FIG. 13( a ) is a schematic diagram of an optional primitive type of a word cloud image generation method provided by an embodiment of the present invention. Figure 1 ;

[0091] FIG. 13( b ) is a schematic diagram of an optional primitive type of a word cloud image generation method provided by an embodiment of the present invention. Figure 2 ;

[0092] Fig.14 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 5 ;

[0093] Fig.15 An optional training foreground primitive of a word cloud image generation method is provided for an embodiment of the present invention;

[0094] FIG. 16( a ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 1 ;

[0095] FIG. 16( b ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 2 ;

[0096] FIG. 16( c ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 3 ;

[0097] FIG. 16( d ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 4 ;

[0098] FIG. 17( a ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 5 ;

[0099] FIG. 17( b ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 6 ;

[0100] FIG. 17( c ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 7 ;

[0101] FIG. 17( d ) is an optional effect of binarization processing of a word cloud image generation method provided by an embodiment of the present invention Figure 8 ;

[0102] Fig.18 An optional training background primitive for a word cloud image generation method is provided for an embodiment of the present invention;

[0103] Fig.19 An optional effect of binarization processing of a word cloud image generation method provided in an embodiment of the present invention Figure 9 ;

[0104] Fig. 20 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 6 ;

[0105] Fig.21 An optional schematic diagram of finding a blank center position in a word cloud image generation method is provided for an embodiment of the present invention;

[0106] Fig. 22 An optional method for obtaining training foreground primitive parameters in a word cloud image generation method according to an embodiment of the present invention is provided. Figure 1 ;

[0107] Fig.23 An optional method for obtaining training foreground primitive parameters in a word cloud image generation method according to an embodiment of the present invention is provided. Figure 2 ;

[0108] Fig.24 An optional initial word cloud image of a word cloud image generation method is provided for an embodiment of the present invention;

[0109] Fig.25 An optional process diagram of a word cloud image generation method is provided for an embodiment of the present invention Figure 7 ;

[0110] Fig.26 A framework diagram of an optional training network for a word cloud image generation method is provided for an embodiment of the present invention;

[0111] Fig. 27An optional feature extraction schematic diagram of a word cloud image generation method is provided for an embodiment of the present invention;

[0112] Fig.28 A structural diagram of a word cloud image generation device is provided for an embodiment of the present invention Figure 1 ;

[0113] Fig.29 A structural diagram of a word cloud image generation device is provided for an embodiment of the present invention Figure 2 . DETAILED DESCRIPTION

[0114] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.

[0115] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 1 , will combine Figure 1 The steps shown are explained.

[0116] S101. Obtain a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element.

[0117] In some embodiments of the present invention, the word cloud image to be generated is an original image taken by a camera or a picture downloaded from the Internet, but the embodiments of the present invention are not limited thereto.

[0118] In some embodiments of the present invention, the terminal may obtain pictures in local files and download pictures from the Internet and use them as the word cloud pictures to be generated.

[0119] S102: Based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element.

[0120] In some embodiments of the present invention, the preset convolutional neural network is a type of feedforward neural network that includes convolution calculations and has a deep structure, and is a representative algorithm of deep learning. The convolutional neural network has representation learning capabilities and can perform shift-invariant classification on input information according to its hierarchical structure.

[0121] In some embodiments of the present invention, the terminal may perform feature extraction on the first foreground image element through a preset convolutional neural network to obtain a first feature map of the first foreground image element; and perform feature extraction on the first background image element to obtain a second feature map of the first background image element.

[0122] S103: Based on the first feature map and the second feature map, a fixed-dimensional feature vector is determined by processing through a target detection special layer.

[0123] In some embodiments of the present invention, feature maps of different sizes are input to a special layer for object detection, which processes the input image and outputs a feature vector of fixed dimension.

[0124] In some embodiments of the present invention, the terminal can determine a local feature map of the second feature map through feature extraction processing through the second feature map of the first background image element; perform special layer processing for target detection through the local feature maps of the first feature map and the second feature map to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map; and obtain a fixed-dimensional feature vector through fusion processing of the first fixed-dimensional vector and the second fixed-dimensional vector.

[0125] In some embodiments of the present invention, Figure 2 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 2 , S103 can be implemented through S1031-S1033 as follows:

[0126] S1031. Determine a local feature map of the second feature map through feature extraction processing based on the second feature map of the first background image element.

[0127] In some embodiments of the present invention, the terminal can use a preset blank detection algorithm to detect the second feature map of the first background image element to obtain the blank center position of the second feature map; perform a surround scanning process through the second feature map and the blank center position of the second feature map to obtain the feature points of the second feature map; and obtain a local feature map of the second feature map by performing a window capture process through the feature points of the second feature map.

[0128] In some embodiments of the present invention, S1031 may be implemented through S10311-S10313 as follows:

[0129] S10311. Based on the second feature map of the second background image element, a preset blank detection algorithm is used to detect and obtain a blank center position of the second feature map.

[0130] In some embodiments of the present invention, the preset blank detection algorithm is an algorithm designed for filling text primitives of the present invention, and firstly selects a blank area in the second feature map as the initial scanning position, performs iterative calculation, and determines the blank center position. The blank center position is the blankest position in the second feature map.

[0131] In some embodiments of the present invention, the terminal may perform a detection operation on the second feature map of the first background image element by using a preset blank detection algorithm to obtain the blank center position of the second feature map.

[0132] S10312: Based on the second feature map and the blank center position of the second feature map, perform a surround scanning process to obtain feature points of the second feature map.

[0133] In some embodiments of the present invention, the terminal may scan the second feature map through a blank center position of the second feature map and around the blank center position to obtain feature points of the second feature map.

[0134] S10313. Based on the feature points of the second feature map, obtain a local feature map of the second feature map by performing window clipping processing.

[0135] In some embodiments of the present invention, the window clipping process is a process of clipping feature points of the second feature map through an anchor window.

[0136] In some embodiments of the present invention, the terminal may obtain a local feature map of the second feature map by performing window interception processing on feature points of the second feature map.

[0137] For example, Figure 3 is an optional window size diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 3 There are three sizes of windows in the figure, with ratios of 1:1, 1:2, and 2:1.

[0138] It can be understood that in some embodiments of the present invention, the terminal can determine the local feature map of the second feature map through the second feature map of the second background image element through feature extraction processing, thereby paving the way for subsequent determination of the second fixed dimensional vector of the second feature map.

[0139] S1032. Based on the local feature map of the first feature map and the second feature map, a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map are obtained by processing through a special layer for target detection.

[0140] In some embodiments of the present invention, the terminal may process the first feature map through a special target detection layer to obtain a first fixed-dimensional vector of the first feature map; and process the local feature map of the second feature map through a special target detection layer to obtain a second fixed-dimensional vector of the second feature map.

[0141] In some embodiments of the present invention, S1032 may be implemented by S10321-S10322 as follows:

[0142] S10321. Based on the first feature map and the local feature map of the second feature map, a bilinear interpolation method is used to obtain a first fixed feature point of the first feature map and a second fixed feature point of the local feature map of the second feature map.

[0143] In some embodiments of the present invention, bilinear interpolation (ROI ALGIN) is also called bilinear interpolation, and its core idea is to perform linear interpolation in two directions respectively. As an interpolation algorithm in numerical analysis, bilinear interpolation is widely used in signal processing, digital image and video processing, etc.

[0144] In some embodiments of the present invention, the terminal may process the local feature maps of the first feature map and the second feature map by bilinear interpolation to obtain the first fixed feature points of the first feature map and the second fixed feature points of the local feature map of the second feature map.

[0145] For example, Figure 4 FIG. 1 is an optional schematic diagram of generating fixed feature points of a word cloud image generation method provided by an embodiment of the present invention. Figure 4 As shown, for the 5*5 local feature map, bilinear interpolation is used to obtain 3*3 fixed feature points.

[0146] The process of bilinear interpolation is as follows Figure 5 As shown, according to Q 12 , Q 22 , Q 11 and Q 21 The four points get the P value of the x and y points. First, use linear interpolation to calculate the values ​​of the R1 and R2 coordinate points, and then interpolate R1 and R2 to calculate the value of point P.

[0147] Exemplarily, the interpolation calculation can be obtained by the following formula (1).

[0148]

[0149] Among them, Q 12 , Q 22 , Q 11 , Q 21 and P are five points respectively; x1, y2 are Q 12 The coordinates of Q; x2, y2 are 22 The coordinates of Q; x1, y1 are 11 The coordinates of Q; x2, y1 are 21 ’s coordinates; x, y are the coordinates of P.

[0150] S10322. Determine a first fixed dimensional vector of the first feature map and a second fixed dimensional vector of the second feature map based on the first fixed feature point and the second fixed feature point.

[0151] In some embodiments of the present invention, the terminal can obtain a first matrix composed of first fixed feature points through the first fixed feature points; obtain a second matrix composed of second fixed feature points through the second fixed feature points; obtain a first fixed-dimensional vector of the first feature map through the first matrix; and obtain a second fixed-dimensional vector of the second feature map through the second matrix.

[0152] It can be understood that in some embodiments of the present invention, the terminal can process the local feature maps of the first feature map and the second feature map through bilinear interpolation to obtain the first fixed feature point of the first feature map and the second fixed feature point of the local feature map of the second feature map; determine the first fixed dimensional vector of the first feature map and the second fixed dimensional vector of the second feature map through the first fixed feature point and the second fixed feature point, thereby paving the way for obtaining the fixed dimensional feature vector.

[0153] S1033. Obtain a fixed-dimensional feature vector through fusion processing based on the first fixed-dimensional vector and the second fixed-dimensional vector.

[0154] In some embodiments of the present invention, the terminal may perform fusion processing on the first fixed-dimensional vector and the second fixed-dimensional vector to fuse the two fixed-dimensional vectors into a fixed-dimensional feature vector.

[0155] It can be understood that in some embodiments of the present invention, the terminal can perform feature extraction processing through the second feature map of the first background image element to determine the local feature map of the second feature map; use the target detection special layer to process the local feature maps of the first feature map and the second feature map to obtain the first fixed dimension vector of the first feature map and the second fixed dimension vector of the second feature map; perform fusion processing on the first fixed dimension vector and the second fixed dimension vector to obtain a fixed dimension feature vector, which paves the way for obtaining regression parameters and performing predictive typesetting.

[0156] S104: Based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image.

[0157] In some embodiments of the present invention, the terminal may process a fixed-dimensional feature vector through a preset regression prediction network to obtain regression parameters; obtain the angle of prediction of the first foreground primitive through the angle parameters in the regression parameters; perform regression calculation through the regression parameters to obtain the predicted position and size of the first foreground primitive; perform typeset calculation through the predicted angle of the first foreground primitive, the predicted position and size of the first foreground primitive, and obtain a preliminary word cloud image.

[0158] In some embodiments of the present invention, S104 may be implemented through S1041-S1044 as follows:

[0159] S1041. Based on the fixed-dimensional feature vector, a preset regression prediction network is used for processing to obtain regression parameters.

[0160] In some embodiments of the present invention, the preset regression prediction network includes an input layer, two hidden layers and an output layer; the regression parameters include position parameters, size parameters and angle parameters.

[0161] In some embodiments of the present invention, the terminal can input a fixed-dimensional feature vector through a preset regression prediction network, and obtain the regression parameters of the output layer through two hidden layers.

[0162] For example, Figure 6 As shown in the figure, the preset regression prediction network includes an input layer, two hidden layers and an output layer. The fixed-dimensional feature vector is input and the output layer d can be obtained through two hidden layers. x (A), d y (A), d w (A) and d θθ (A), are the size parameter, position parameter and angle parameter, respectively.

[0163] S1042: Based on the angle parameter, obtain the predicted angle of the first foreground primitive.

[0164] In some embodiments of the present invention, the terminal may predict the predicted angle of the first foreground image element by using an angle parameter to obtain the predicted angle of the first foreground image element.

[0165] S1043: Perform regression calculation based on the regression parameters to obtain the predicted position and size of the first foreground primitive.

[0166] In some embodiments of the present invention, the terminal may perform regression calculation using regression parameters to obtain the predicted position and size of the first foreground primitive.

[0167] For example, Figure 7 As shown in the figure, the actual position and size are G, the anchor box is A, and the purpose of regression is to translate and scale A to get the position of G', which is the prediction of the position and size of the foreground primitive.

[0168] Exemplarily, the position and size prediction can be obtained by the following formula (2).

[0169]

[0170] Among them, d x (A), d y (A) is the length parameter and width parameter in the size parameter, d w (A) is the position parameter, G' x and G' y To predict the size, G' w is the predicted position, A w and A h are the actual length and height respectively, A x and A y They are the length and width of the anchor box respectively.

[0171] S1044 , based on the predicted angle of the first foreground image element, the predicted position and size of the first foreground image element, perform typesetting to obtain a preliminary word cloud image.

[0172] In some embodiments of the present invention, the terminal may perform typeset operation according to the angle, position and size predicted by the first foreground image element to obtain a preliminary word cloud image.

[0173] It can be understood that in some embodiments of the present invention, the terminal can use a preset regression prediction network to process a fixed-dimensional feature vector to obtain regression parameters; based on the regression parameters, regression calculations are performed to obtain the predicted angle, position and size of the first foreground element, and then typeset to obtain a preliminary word cloud image, thereby improving the efficiency of generating word cloud images.

[0174] S105 , coloring the preliminary word cloud image to determine the generated word cloud image.

[0175] In some embodiments of the present invention, the terminal can perform spatial color clustering through the word cloud image to be generated to establish a feature vector; cluster the feature vector of the word cloud image to be generated to obtain a color center point; match the color center point and the preliminary word cloud image using the nearest match principle to obtain a matching result; and color the preliminary word cloud image based on the matching result to obtain a word cloud image.

[0176] It can be understood that in some embodiments of the present invention, the terminal can obtain a word cloud image to be generated; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element; based on the first feature map and the second feature map, processing is performed through a special layer of target detection to determine a fixed-dimensional feature vector; based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image; coloring is performed on the preliminary word cloud image to determine the generated word cloud image, thereby improving the efficiency of word cloud image generation, and simultaneously using background images and foreground images to simultaneously generate word cloud images, thereby enhancing the positioning capability of foreground images in the background and improving the filling effect.

[0177] In some embodiments of the present invention, Figure 8 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 3 , S105 can be implemented through S1051-S1054 as follows:

[0178] S1051. Based on the word cloud image to be generated, perform spatial color clustering and establish a feature vector.

[0179] In some embodiments of the present invention, spatial color clustering uses mean shift clustering, which is a sliding window-based algorithm that attempts to find dense areas of data points. This is a centroid-based algorithm that aims to locate the center point of each group / class by updating the candidate points of the center point to the mean of the points in the sliding window. These candidate windows are then filtered in the post-processing stage to eliminate near-duplicates to form the final set of center points and their corresponding groups.

[0180] In some embodiments of the present invention, the terminal may perform spatial color clustering on the word cloud image to be generated, obtain a color center template, and then establish a feature vector.

[0181] S1052, clustering is performed based on the feature vector of the word cloud image to be generated to obtain a color center point.

[0182] In some embodiments of the present invention, the terminal may perform clustering based on the feature vectors of the word cloud image to be generated to obtain the color center point.

[0183] For example, the feature vector is V, and each pixel in the word cloud image to be generated can be converted into a vector as shown in formula (3):

[0184] V i =R i ,,G i ,B i ,X i ,Yi (3)

[0185] Among them, i represents the pixel point, R i ,,G i ,B i ,X i ,Y i They represent different color centers respectively.

[0186] For example, Fig. 9 As shown, 1, 2, 3, 4, 5 and 6 are color center points, and A, B, C and D are word cloud primitives of the preliminary word cloud image.

[0187] S1053. Based on the color center point and the preliminary word cloud image, matching is performed using a nearest matching principle to obtain a matching result.

[0188] In some embodiments of the present invention, the preliminary word cloud image includes a plurality of word cloud primitives.

[0189] In some embodiments of the present invention, the terminal can match the color center point and multiple word clouds according to the nearest matching principle. If the word cloud matches the only color center, a matching result is obtained; if the word cloud matches at least two color centers, the matching result is determined by the maximum weight matching algorithm.

[0190] In some embodiments of the present invention, S1053 may be implemented by S10531 and S10532 as follows:

[0191] S10531. Based on the color center point and multiple word cloud primitives, matching is performed using the nearest matching principle. If the word cloud primitive matches a unique color center, a matching result is obtained.

[0192] In some embodiments of the present invention, the terminal may perform matching through the color center point and multiple word cloud primitives using the nearest matching principle. If a word cloud primitive matches a unique color center, a matching result of the color center and multiple word cloud primitives is obtained.

[0193] S10532. Based on the color center point and multiple word cloud primitives, matching is performed using the nearest matching principle. If the word cloud primitive matches at least two color centers, the matching result is determined using a maximum weight matching algorithm.

[0194] In some embodiments of the present invention, a maximum weight matching (Kuhn-Munkras, KM) algorithm is used to find the best matching algorithm of a weighted bipartite graph.

[0195] In some embodiments of the present invention, the terminal can use the nearest matching principle to match the color center point and multiple word cloud primitives. If a word cloud primitive matches at least two color centers, it is processed through the maximum weight matching algorithm to obtain the matching results of the color center and multiple word cloud primitives.

[0196] It is understandable that in some embodiments of the present invention, the terminal can match the word cloud primitives in the preliminary word cloud image using the color center point and the nearest match principle to obtain a matching result, paving the way for subsequent coloring of the word cloud primitives.

[0197] S1054. Based on the matching results, the preliminary word cloud image is colored to obtain a word cloud image.

[0198] In some embodiments of the present invention, the terminal may color the word cloud primitives in the preliminary word cloud image according to the matching results to obtain the word cloud image.

[0199] It can be understood that in some embodiments of the present invention, the terminal can perform spatial color clustering through the word cloud image to be generated to establish a feature vector; cluster the feature vector of the word cloud image to be generated to obtain the color center point; match the color center point and the preliminary word cloud image using the nearest match principle to obtain a matching result; color the preliminary word cloud image based on the matching result to obtain a word cloud image, so that the word cloud image can be obtained quickly, thereby improving the efficiency of word cloud image generation.

[0200] In some embodiments of the present invention, Fig.10 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 4 , before S102, S106-S1010 are also executed, combining Fig.10 The steps shown are explained.

[0201] S106: Obtain multiple training images.

[0202] In some embodiments of the present invention, each training picture includes a plurality of foreground primitives and training background primitives.

[0203] In some embodiments of the present invention, the terminal can obtain multiple training images through two ways: local files and online downloads.

[0204] S107 . Based on multiple foreground primitives, determine a training foreground primitive of the current primitive type through type selection processing.

[0205] In some embodiments of the present invention, the terminal can establish a quadtree through multiple foreground elements for data management to obtain a data management structure; search through the data management structure to determine the adjacent elements of the current foreground element; determine the type of the current foreground element based on the type of the adjacent elements; and determine the training foreground element based on the type of the current foreground element.

[0206] In some embodiments of the present invention, S107 may be implemented by S1071-S1074 as follows:

[0207] S1071. Based on multiple foreground graphics elements, a quadtree is established to perform data management to obtain a data management structure.

[0208] In some embodiments of the present invention, the basic idea of ​​the quadtree is to divide a raster map or image into four equal parts and check the grid attribute value (or grayscale) block by block.

[0209] In some embodiments of the present invention, the terminal may perform data management by establishing a quadtree for multiple foreground primitives, dividing multiple primitives in each picture, and obtaining a foreground primitive data management structure for each image.

[0210] For example, the foreground primitive in each image is taken as a center point, and the entire image area is recursively divided into four regions, and a minimum size threshold is specified. If there are multiple primitive center points in an area, the division continues until the division ends at the minimum threshold, and each primitive center point is attached to the partition. If there are multiple primitives within the minimum threshold area, multiple primitives are attached. The division process is as follows: Fig.11 As shown, Fig.11 Each black dot in represents a different foreground primitive. After the division, we get Fig.12 The data management structure is as follows: root represents the current foreground primitive, L and R represent left and right; U and D represent up and down, LU represents that there is a foreground primitive to the upper left of the current primitive, and RU represents that there is a foreground primitive to the upper right of the current primitive.

[0211] S1072. Search based on the data management structure to determine adjacent graphics elements of the current foreground graphics element.

[0212] In some embodiments of the present invention, the terminal may search through the foreground image element data management structure of each picture to obtain adjacent image elements of the current foreground image element.

[0213] S1073. Determine the type of the current foreground primitive according to the types of the adjacent primitives.

[0214] In some embodiments of the present invention, the terminal may exclude foreground image elements of the same type according to the types of adjacent image elements, and determine the type of the current foreground image element.

[0215] For example, as shown in FIG13( a), ● represents one type of foreground primitive, ○ represents another type of foreground primitive; A, B, C, D, E, F and G are adjacent primitives of the current foreground primitive; as can be seen from FIG13( a), A, D, B and C are of the same primitive type, and E, F and G are of the same primitive type.

[0216] Exemplarily, as shown in FIG13( b ), the gray squares represent one type of primitive, and the white squares represent another type of primitive; after obtaining the adjacent primitives of the current foreground primitive, the types of E, F, and G are excluded from the list based on the adjacent types of A, D, B, and C, and any one of E, F, and G is randomly selected as the type of the current primitive.

[0217] S1074: Determine a training foreground primitive based on the type of the current foreground primitive.

[0218] In some embodiments of the present invention, the terminal may use the type of the current foreground image element as a training foreground image element according to the type of the current foreground image element.

[0219] It can be understood that in some embodiments of the present invention, the terminal can perform quadtree data management and type selection processing on multiple foreground graphics elements, determine the training foreground graphics element of the current graphics element type, pave the way for subsequent determination of training sample data, and improve the effectiveness of the training sample data.

[0220] S108 . Determine parameters of the training foreground primitives through convolution scanning based on the training foreground primitives and the training background primitives.

[0221] In some embodiments of the present invention, the terminal can obtain a first binarized image of the training foreground image element through a binarization process by training the foreground image element; obtain a second binarized image of the training background image element through a binarization process by training the background image element; and determine the parameters of the training foreground image element through the first binarization image and the second binarization image.

[0222] In some embodiments of the present invention, Fig.14 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 5 , S108 can be implemented through S1081-S1083 as follows:

[0223] S1081. Based on the training foreground image element, a first binarized image of the training foreground image element is obtained through binarization processing.

[0224] In some embodiments of the present invention, the terminal can obtain multiple training foreground sub-picture elements of different sizes through a training foreground picture element and a scaling process; obtain a binary image of the training foreground picture element through a binarization process through the training foreground sub-picture element; and obtain a second binary image of the training background picture element through a binarization process through the training background picture element.

[0225] In some embodiments of the present invention, S1081 may be implemented by S10811, S10812, and S10813 as follows:

[0226] S10811. Based on the training foreground primitive, obtain multi-size training foreground sub-primitives through scaling processing.

[0227] In some embodiments of the present invention, multi-size training foreground sub-primitives refer to training foreground sub-primitives of different sizes.

[0228] In some embodiments of the present invention, the terminal may obtain a plurality of training foreground sub-picture elements of different sizes through a training foreground picture element and a scaling process.

[0229] S10812, based on each foreground sub-pixel of the multi-size training foreground sub-pixel, obtain a multi-angle training foreground sub-pixel for each foreground sub-pixel through rotation processing.

[0230] In some embodiments of the present invention, the terminal may rotate each foreground sub-picture element of the multi-size training foreground sub-picture element to obtain a multi-angle training foreground sub-picture element for each foreground sub-picture element.

[0231] S10813. Based on the multi-angle training foreground sub-pixel, a first binarized image of the training foreground pixel is obtained through binarization processing.

[0232] In some embodiments of the present invention, the multi-angle training foreground sub-pixels refer to training foreground sub-pixels at different angles.

[0233] In some embodiments of the present invention, the terminal may perform binarization processing on the training foreground sub-pixels at different angles to obtain binarized images of the training foreground sub-pixels, thereby determining a first binarized image of the training foreground sub-pixels.

[0234] In some embodiments of the present invention, the terminal may perform grid processing on the training foreground primitive and the training background primitive to obtain the grid-processed training foreground primitive and training background primitive. When performing binarization processing on the training foreground primitive and the training background primitive, it is based on whether the primitive occupies the grid, and the assigned values for occupation and non-occupation can be different, or binary assignment can be used, which is not limited in the embodiments of the present invention. In addition, the grids corresponding to occupation and non-occupation can be visually displayed in different colors according to different assigned values, which is not limited in the embodiments of the present invention.

[0235] Exemplarily, if the assigned value for occupation is 1 and the assigned value for non-occupation is 0, a binary matrix of 0 and 1 can be obtained, which is the binarization processing result and is also reflected in the grid graph. 1 indicates that the grid is black, and 0 indicates that the grid is white.

[0236] Exemplarily, Fig.15 is a training foreground primitive, which is obtained after grid processing. It can be seen that Fig.15 the Chinese character "word" is in it. Fig. 16(a) is Fig.15 the effect diagram of the image shown in it after binarization processing. It can be seen that Fig.15 in the third row, 8 grids are occupied, which are grids 4, 5, 6, 7, 8, 9, 10, and 11 respectively. There are 8 black grids at the corresponding positions after binarization in Fig. 16(a). Similarly, the effect diagrams of all grids after binarization can be obtained. Fig. 16(b), Fig. 16(c), and Fig. 16(d) are Fig.15 the effect diagrams of the image shown in it after different degrees of scaling processing and then binarization. It can be seen that the basic outline of the word "word" still exists in Fig. 16(b), but the details have become a black area due to scaling; the outlines cannot be seen in Fig. 16(c) and Fig. 16(d), only two black squares of different sizes. The black square in Fig. 16(c) is composed of 9 black grids, and the black square in Fig. 16(d) is composed of 4 black grids.

[0237] Exemplarily, Fig. 17(a), Fig. 17(b), Fig. 17(c), and Fig. 17(d) are Fig.15 the effect diagrams of the image shown in it after binarization processing after being rotated by different angles. Fig. 17(a) is Fig.15 the image shown in it after being rotated by 45 degrees and then binarized; Fig. 17(b) is Fig.15 the image shown in it after being rotated by 120 degrees and then binarized; Fig. 17(c) is the image shown in Fig. 15 after being rotated by 180 degrees and then binarized; Fig. 17(d) is Fig.15 the image shown in it after being rotated by 270 degrees and then binarized.

[0238] It can be understood that in some embodiments of the present invention, the terminal can obtain multiple training foreground sub-elements by scaling and rotating the training foreground element, and then perform binarization processing to obtain a first binary image of the training foreground element, thereby improving the diversity of the first binary image and improving the effectiveness of the training sample data.

[0239] S1082. Based on the training background image element, obtain a second binarized image of the training background image element through binarization processing.

[0240] In some embodiments of the present invention, the terminal may perform binarization processing on the training background image element to obtain a second binarized image of the training background image element.

[0241] For example, Fig.18 This is a training background primitive, which is obtained after gridding. You can see Fig.18 The middle one is an ellipse. The second binary image of the training background primitive after binary processing is as follows Fig.19 shown. Fig.19 is based on Fig.18 Whether the ellipse occupies the grid for binarization can be seen Fig.18 The second row occupies 3 grids, such as Fig.18 The grids 7, 8, and 9 shown (here, as long as part of the grid is occupied, it is considered occupied), Fig.19 The corresponding 7, 8, and 9 grids in the image are binarized to black. Fig.18 The unoccupied grid cells in the second row are 1, 2, 3, 4, 5, 6, 10, 11, 12, and 13. Fig.19 The corresponding 1, 2, 3, 4, 5, 6, 10, 11, 12 and 13 grids are binarized to white. Similarly, the binarized effect of all grids can be obtained.

[0242] S1083. Determine parameters of a training foreground primitive based on the first binarized image and the second binarized image.

[0243] In some embodiments of the present invention, the terminal can use a second binarized image to perform operations using a preset blank detection algorithm to determine the blank center position of the second binarized image; and use convolution scanning processing to determine the parameters of the training foreground primitives using the blank center positions of the first binarized sub-image and the second binarized image.

[0244] It can be understood that in some embodiments of the present invention, the terminal can obtain a first binarized image of the training foreground element and a second binarized image of the training background element through the training foreground element and the training background image by binarization processing; and determine the parameters of the training foreground element through the first binarized image and the second binarized image, so as to pave the way for the subsequent generation of the initial word cloud image.

[0245] In some embodiments of the present invention, Fig. 20 This is an optional process diagram of a word cloud image generation method provided by an embodiment of the present invention. Figure 6 , S1083 can be implemented through S10831-S10832 as follows:

[0246] S10831. Based on the second binary image, a preset blank detection algorithm is used to perform calculations to determine a blank center position of the second binary image.

[0247] In some embodiments of the present invention, the preset blank detection algorithm is an algorithm designed to enhance the intelligent positioning capability of foreground primitives.

[0248] In some embodiments of the present invention, the terminal can select a blank area as the initial scanning position through the second binary image; perform operations by scanning the initial position and a preset field strength calculation criterion to determine the field strength of each cell in the second binary image; perform operations by using the field strength of each cell in the second binary image and a preset gradient calculation criterion to obtain the field strength gradient of each cell; perform iterative operations by using the field strength gradient of each cell until the maximum value of the field strength is obtained, determine the cell corresponding to the maximum value of the field strength, and use it as the blank center position.

[0249] In some embodiments of the present invention, S10831 may be implemented by S108311 and S108315 as follows:

[0250] S108311. Based on the second binarized image, select a blank area as the initial scanning position.

[0251] In some embodiments of the present invention, the terminal may select a blank area in the second binarized image as an initial scanning position.

[0252] S108312. Perform calculations based on the scan initial position and a preset field strength calculation criterion to determine the field strength of each cell in the second binary image.

[0253] In some embodiments of the present invention, the terminal may perform operations on the cells of the second binary image by scanning the initial position and presetting the field strength calculation criterion to obtain the field strength of each cell in the second binary image.

[0254] Exemplarily, the field strength of each cell in the second binary image can be obtained by the following formula (4).

[0255]

[0256] Where q is the total field strength at x, y, which is the sum of the field strengths of the empty cells (set E) minus the sum of the field strengths of the filled cells (set F), x k represents a blank cell, x i Indicates the filled cell; f indicates the field strength calculation criterion; x, y are the coordinates of the cell.

[0257] f represents the field strength calculation criterion, which can be obtained by the following formula (5).

[0258]

[0259] Among them, x, y are the coordinates of the cell; x k Represents a blank cell; and They represent the coefficients corresponding to the horizontal and vertical coordinates of the cell respectively.

[0260] S108313. Based on the field intensity of each cell in the second binary image, a calculation is performed using a preset gradient calculation criterion to obtain a field intensity gradient of each cell.

[0261] In some embodiments of the present invention, the terminal may calculate the field intensity of each cell in the second binarized image using a preset gradient calculation criterion to obtain the field intensity gradient of each cell.

[0262] Exemplarily, the field intensity gradient can be obtained by the following formula (6).

[0263]

[0264] Among them, dx represents the lateral field intensity gradient of each cell; dy represents the longitudinal field intensity gradient of each cell; q(x,y) is the total field intensity at x,y, q(x+1,y) is the total field intensity at x+1,y; q(x,y+1) is the total field intensity at x,y+1.

[0265] S108314. Based on the field intensity gradient of each cell, perform iterative calculations until the maximum value of the field intensity is obtained, and determine the cell corresponding to the maximum value of the field intensity.

[0266] In some embodiments of the present invention, the terminal may perform iterative calculations on the field intensity gradient of each cell until the maximum value of the field intensity is obtained, and determine the cell corresponding to the maximum value of the field intensity.

[0267] Exemplarily, the iterative operation can be obtained by the following formula (7).

[0268]

[0269] Among them, dx represents the lateral field intensity gradient of each cell; dy represents the longitudinal field intensity gradient of each cell; x, y are the field intensity of the cell; μ is the field intensity coefficient; x' and y' are the updated field intensity.

[0270] S108315. The cell corresponding to the maximum value of the field strength is used as the blank center position.

[0271] In some embodiments of the present invention, the terminal may use the cell corresponding to the maximum value of the field intensity as the blank center position.

[0272] For example, Fig.21 is a schematic diagram for finding the blank center position, such as Fig.21 As shown, Fig.21 The area indicated by the shaded area is the second binary image, which is Fig.21 The starting point 1 shown in the figure is iterated to obtain the starting point 2; the iterative operation is continued based on the starting point 2 to obtain the starting point 3, and the iteration is continued to obtain the blank center position.

[0273] It is understandable that in some embodiments of the present invention, the terminal can operate on the second binary image through a preset blank detection algorithm to determine the blank center position of the second binary image, laying the foundation for subsequent determination of parameters of the training foreground image element.

[0274] S10832. Based on the blank center positions of the first binarized sub-image and the second binarized image, determine the parameters of the training foreground primitive through convolution scanning processing.

[0275] In some embodiments of the present invention, the terminal may scan the first binarized sub-image from the blank center position of the second binarized image in the direction of the spiral line to determine the parameters of the training foreground primitive.

[0276] For example, Fig. 22 This is a parameter diagram for obtaining the training foreground primitive. Figure 1 . Start scanning in a clockwise spiral from 1 to obtain parameters represented by 0 to 35. 0 means starting scanning from the center of the blank, and the first overlapping area of ​​the overlapping part of the first binary sub-image of the training foreground element and the second binary sub-image of the training background element is recorded as 0, and the parameters it represents are recorded. Similarly, 1 represents the second overlapping area of ​​the overlapping part of the two, and the parameters it represents are recorded. The areas where all overlapping parts are located and the parameters they represent are obtained in turn, thereby obtaining the parameters of the training foreground element.

[0277] It can be understood that in some embodiments of the present invention, the terminal can use the second binarized image to perform operations using a preset blank detection algorithm to determine the blank center position of the second binarized image; perform convolution scanning processing through the blank center positions of the first binarized sub-image and the second binarized image to determine the parameters of the training foreground primitives, thereby paving the way for subsequently obtaining the initial word cloud image.

[0278] In some embodiments of the present invention, S108 may also be implemented through S1084-S1086 as follows:

[0279] S1084. Based on the training foreground image element and the training background image element, a third binarized image of the training foreground image element and a fourth binarized image of the training background image element are obtained through binarization processing.

[0280] In some embodiments of the present invention, the terminal may directly perform binarization processing on the training foreground image elements and the training background image elements to obtain a third binarized image of the training foreground image elements and a fourth binarized image of the training background image elements.

[0281] S1085. Perform convolution scanning on the third binarized image on the fourth binarized image to obtain a product value of each step.

[0282] In some embodiments of the present invention, the terminal may perform convolution scanning on the third binarized image on the fourth binarized image to obtain a product value at each step.

[0283] S1086. Determine parameters of the training foreground primitive based on the product value and the third binarized image.

[0284] In some embodiments of the present invention, the terminal can compare the product value with the number of the third binary image. If they are equal, it means that the training background image element can cover the training foreground image element, indicating that the parameter of the training foreground image element is acceptable. If they are not equal, the third binary image should be moved continuously until the product value is equal to the number of the third binary image to obtain the parameters of the training foreground image element.

[0285] For example, Fig.23 This is a parameter diagram for training foreground primitives. Figure 2 ,like Fig.23 As shown, A is the fourth binarized image of the training background primitive; B is the third binarized image of the training foreground primitive; B gradually moves toward A to determine the parameters. When B moves toward A, the parameters of the overlapping parts are recorded from left to right and from top to bottom. The overlapping area at the beginning is 1, and then 2, 3, 4..., until 31, and the parameters represented by 1 to 31 are recorded respectively. However, it can be seen that A does not completely cover B. Therefore, this parameter is not acceptable. Continue to move until A completely covers B to obtain the parameters.

[0286] It can be understood that in some embodiments of the present invention, the terminal can perform binarization processing based on the training foreground image element and the training background image element to obtain a third binarized image of the training foreground image element and a fourth binarized image of the training background image element; perform convolution scanning on the third binarized image on the fourth binarized image to obtain the product value of each step; determine the parameters of the training foreground image element through the product value and the third binarized image, and pave the way for subsequently obtaining the initial word cloud image.

[0287] S109, based on the parameters of the training foreground primitives, initial word cloud images corresponding to the plurality of training images are obtained, and qualified images are screened out as training sample data.

[0288] In some embodiments of the present invention, the terminal can layout the word cloud by training the parameters of the foreground graphic element, determine the initial word cloud images corresponding to each of the multiple training images, and select qualified images as training sample data, and record the parameters of the training sample data.

[0289] For example, Fig.24 This is the initial word cloud image obtained. It can be seen that there are 7 initial word cloud images, of which the sizes, rotation angles and locations are different. Among them, the images numbered 1-5 are selected as training sample data.

[0290] S1010: Based on the training sample data, the initial convolutional neural network is trained to obtain a preset convolutional neural network.

[0291] In some embodiments of the present invention, the terminal can train the initial convolutional neural network through training sample data until the parameters of the training sample data are obtained, and save the trained convolutional neural network model to obtain a preset convolutional neural network.

[0292] It can be understood that in some embodiments of the present invention, the terminal can process multiple training images to obtain sample data, and then train the initial convolutional neural network through the sample data to obtain a preset convolutional neural network, thereby paving the way for the subsequent generation of preliminary word cloud images and improving the efficiency of generating preliminary word cloud images.

[0293] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0294] The embodiment of the present invention provides an optional flowchart of a method for generating a word cloud image. Fig.25 shown.

[0295] In some embodiments of the present invention, the terminal realizes word cloud image generation through four parts: creating a training set, selecting a training set, training a network, and matching colors of graphic elements.

[0296] S1. Create a training set.

[0297] In some embodiments of the present invention, the terminal can obtain multiple pictures, thereby obtaining the primitive background (equivalent to the training background primitive) and the primitive foreground (equivalent to the training foreground primitive) of each picture, and select the primitive foreground and the primitive background; by binarizing the primitive foreground and the primitive background, a binarized primitive foreground (equivalent to the third binarized image) and a binarized primitive background (equivalent to the fourth binarized image) are obtained; blank area detection and convolution scanning are performed on the binarized primitive foreground on the binarized primitive background to obtain foreground primitive parameters (equivalent to the parameters of the training foreground primitive); according to the obtained foreground primitive parameters, the word cloud in the primitive foreground is typeset to obtain an initial word cloud image.

[0298] S2. Select a training set.

[0299] In some embodiments of the present invention, the terminal may select an initial word cloud image based on a plurality of initial word cloud images, and save the selected training sample data (equivalent to parameter data).

[0300] S3. Train the network.

[0301] In some embodiments of the present invention, the terminal can create a feature extraction network (a preset convolutional neural network) to extract the feature maps corresponding to the first foreground image element and the first background image element in the word cloud image to be generated, and use blank area detection to obtain the blank center position of the feature map of the first background image element, thereby obtaining the corresponding local map, and then extracting the background and foreground feature vectors to establish a size, position and angle regression network (equivalent to the preset regression network).

[0302] For example, the framework structure of the training network is as follows Fig.26 As shown, Fig.26 In the process, the terminal can obtain the background (equivalent to the first background image element) and foreground (equivalent to the first foreground image element) of the word cloud image to be generated; obtain the background feature map (equivalent to the second feature map) and the foreground feature map (equivalent to the first feature map) through the convolution layer; obtain the blank center position of the background feature map through blank center detection; perform a surround scan on the blank center position of the background feature map to obtain feature points; perform anchor interception on the feature points to obtain a local feature map of the background feature map; use the ROI ALGIN method to obtain the second fixed dimensional vector of the local feature map and the first fixed dimensional vector of the foreground feature map; merge the first fixed dimensional vector and the second fixed dimensional vector to obtain a fixed dimensional vector (equivalent to the fixed dimensional feature vector).

[0303] For example, Fig. 27 is a schematic diagram of feature extraction. Fig.26 The convolutional layers in can be Fig. 27 The left half of the convolutional autoencoder shown in the figure is obtained. For any input image, a feature map is obtained by convolution and then the original image is obtained by deconvolution. The feature map contains all the information of the original image. The encoded part on the left is used to extract the features of the background and foreground.

[0304] S4. Color matching of graphic elements.

[0305] In some embodiments of the present invention, the terminal may color the primitives (preliminary word cloud images) through color learning.

[0306] It is understandable that in some embodiments of the present invention, the terminal can generate a word cloud image by creating a training set, selecting a training set, training a network, and matching the colors of graphics elements, and use a convolutional network to predict the size and angle of foreground graphics elements to accelerate the generation efficiency of the word cloud typesetting.

[0307] Based on the word cloud image generation method of the above embodiment, the embodiment of the present invention also provides a word cloud image generation device, such as Fig.28 As stated, Fig.28 A schematic diagram of a word cloud image generation device provided by an embodiment of the present invention Figure 1 , the device includes: an acquisition unit 2801 and a determination unit 2802; wherein,

[0308] The acquisition unit 2801 is used to acquire a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element;

[0309] The determining unit 2802 is used to determine a fixed-dimensional feature vector by processing the first feature map and the second feature map through a target detection special layer;

[0310] The acquisition unit 2801 is used to perform predictive typesetting based on the fixed-dimensional feature vector through a preset regression prediction network to obtain a preliminary word cloud image;

[0311] The determining unit 2802 is used to perform coloring processing on the preliminary word cloud image to determine the generated word cloud image.

[0312] In some embodiments of the present invention, the acquisition unit 2801 is used to acquire a plurality of training pictures; wherein each training picture includes a plurality of foreground picture elements and training background picture elements;

[0313] The determining unit 2802 is configured to determine a training foreground primitive of a current primitive type through type selection processing based on the multiple foreground primitives; and determine a parameter of the training foreground primitive through convolution scanning based on the training foreground primitive and the training background primitive;

[0314] The acquisition unit 2801 is used to obtain the initial word cloud images corresponding to each of the multiple training images based on the parameters of the training foreground image element, and screen out qualified images as training sample data; based on the training sample data, train the initial convolutional neural network to obtain the preset convolutional neural network.

[0315] In some embodiments of the present invention, the acquisition unit 2801 is used to obtain a first binarized image of the training foreground image element through binarization processing based on the training foreground image element; and obtain a second binarized image of the training background image element through binarization processing based on the training background image element;

[0316] The determining unit 2802 is used to determine the parameters of the training foreground primitive based on the first binarized image and the second binarized image.

[0317] In some embodiments of the present invention, the acquisition unit 2801 is used to obtain a multi-size training foreground sub-pixel through scaling processing based on the training foreground sub-pixel; obtain a multi-angle training foreground sub-pixel for each foreground sub-pixel through rotation processing based on the multi-size training foreground sub-pixel; and obtain the first binarized image of the training foreground sub-pixel through binarization processing based on the multi-angle training foreground sub-pixel.

[0318] In some embodiments of the present invention, the acquisition unit 2801 is used to obtain a third binarized image of the training foreground image element and a fourth binarized image of the training background image element through binarization processing based on the training foreground image element and the training background image element; perform convolution scanning on the third binarized image on the fourth binarized image to obtain a product value at each step;

[0319] The determining unit 2802 is used to determine the parameters of the training foreground primitive based on the product value and the third binarized image.

[0320] In some embodiments of the present invention, the determination unit 2802 is used to determine the blank center position of the second binarized image by performing calculations through a preset blank detection algorithm based on the second binarized image; and determine the parameters of the first foreground primitive by convolution scanning processing based on the blank center positions of the first binarized sub-image and the second binarized image.

[0321] In some embodiments of the present invention, the determining unit 2802 is used to select a blank area as an initial scanning position based on the second binary image; perform calculations based on the initial scanning position and a preset field strength calculation criterion to determine the field strength of each cell in the second binary image;

[0322] The acquisition unit 2801 is used to perform calculation based on the field intensity of each cell in the second binarized image by a preset gradient calculation rule to obtain the field intensity gradient of each cell;

[0323] The determination unit 2802 is used to perform iterative calculations based on the field intensity gradient of each cell until the maximum field intensity is obtained, and determine the cell corresponding to the maximum field intensity; and use the cell corresponding to the maximum field intensity as the blank center position.

[0324] In some embodiments of the present invention, the acquisition unit 2801 is used to perform data management based on the multiple foreground primitives by establishing a quadtree to obtain a data management structure;

[0325] The determination unit 2802 is used to search and determine adjacent graphics elements of the current foreground graphics element based on the data management structure; determine the type of the current foreground graphics element according to the types of the adjacent graphics elements; and determine the first foreground graphics element based on the type of the current foreground graphics element.

[0326] In some embodiments of the present invention, the determining unit 2802 is used to determine a local feature map of the second feature map through feature extraction processing based on the second feature map of the first background image element;

[0327] The acquisition unit 2801 is used to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map by processing the local feature map of the first feature map and the second feature map through the target detection special layer;

[0328] The fixed-dimensional feature vector is obtained by fusion processing based on the first fixed-dimensional vector and the second fixed-dimensional vector.

[0329] In some embodiments of the present invention, the acquisition unit 2801 is used to perform detection based on the second feature map of the first background image element by using a preset blank detection algorithm to obtain the blank center position of the second feature map; perform surround scanning processing based on the second feature map and the blank center position of the second feature map to obtain feature points of the second feature map; and obtain the local feature map of the second feature map by performing window interception processing based on the feature points of the second feature map.

[0330] In some embodiments of the present invention, the acquisition unit 2801 is used to obtain a first fixed feature point of the first feature map and a second fixed feature point of the local feature map of the second feature map by processing through a bilinear interpolation method based on the first feature map and the local feature map of the second feature map;

[0331] The determining unit 2802 is used to determine the first fixed dimensional vector of the first feature map and the second fixed dimensional vector of the second feature map based on the first fixed feature points and the second fixed feature points.

[0332] In some embodiments of the present invention, the acquisition unit 2801 is used to process the fixed-dimensional feature vector through a preset regression prediction network to obtain regression parameters; wherein the regression parameters include position parameters, size parameters and angle parameters; based on the angle parameters, the angle of the first foreground primitive prediction is obtained; based on the regression parameters, regression calculation is performed to obtain the predicted position and size of the first foreground primitive; based on the predicted angle of the first foreground primitive, the predicted position and size of the first foreground primitive, typeset to obtain the preliminary word cloud image.

[0333] In some embodiments of the present invention, the word cloud image generating device further includes a establishing unit 2803; wherein,

[0334] The establishing unit 2803 is used to perform spatial color clustering based on the word cloud image to be generated and establish a feature vector;

[0335] The acquisition unit 2801 is used to perform clustering based on the feature vector of the word cloud image to be generated to obtain a color center point; based on the color center point and the preliminary word cloud image, match them using the nearest match principle to obtain a matching result; based on the matching result, color the preliminary word cloud image to obtain the word cloud image.

[0336] In some embodiments of the present invention, the preliminary word cloud image includes a plurality of word cloud primitives;

[0337] In some embodiments of the present invention, the acquisition unit 2801 is used to perform matching based on the color center point and the multiple word cloud primitives by the nearest matching principle, and obtain a matching result if the word cloud primitive matches a unique color center;

[0338] The determining unit 2802 is used to perform matching based on the color center point and the multiple word cloud primitives through the nearest matching principle. If the word cloud primitive matches at least two color centers, a matching result is determined through a maximum weight matching algorithm.

[0339] Based on the word cloud image generation method of the above embodiment, the embodiment of the present invention also provides a word cloud image generation device, such as Fig.29 As shown, Fig.29 A schematic diagram of a word cloud image generation device provided by an embodiment of the present invention Figure 2 The device includes: a processor 2901 and a memory 2902; the memory 2902 stores one or more programs executable by the processor, and when one or more programs are executed, any word cloud image generation method of the above-mentioned embodiments is executed by the processor 2901.

[0340] An embodiment of the present invention provides a computer-readable storage medium, characterized in that the storage medium stores executable instructions, which, when executed, are used to cause a processor to execute the word cloud image generation method as described in the embodiment of the present invention.

[0341] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0342] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0343] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0344] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0345] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A method for generating a word cloud image, characterized in that: include: Acquire a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; Based on the first foreground image element and the first background image element, feature extraction is performed by a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element; Based on the second feature map of the first background image element, determining a local feature map of the second feature map through feature extraction processing; Based on the local feature maps of the first feature map and the second feature map, processing is performed through a target detection special layer to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map; Obtaining a fixed-dimensional feature vector through fusion processing based on the first fixed-dimensional vector and the second fixed-dimensional vector; Based on the fixed-dimensional feature vector, predictive typesetting is performed through a preset regression prediction network to obtain a preliminary word cloud image; The preliminary word cloud image is colorized to determine a generated word cloud image.

2. The method according to claim 1, characterized in that Before extracting features based on the foreground image element and the background image element by a preset convolutional neural network to obtain feature maps corresponding to the foreground image element and the background image element respectively, the method further includes: Acquire multiple training images, wherein each training image includes multiple foreground image elements and training background image elements; Based on the multiple foreground primitives, determining a training foreground primitive of a current primitive type through type selection processing; Based on the training foreground primitive and the training background primitive, determining the parameters of the training foreground primitive by convolution scanning; Based on the parameters of the training foreground primitives, initial word cloud images corresponding to the plurality of training images are obtained, and qualified images are screened out as training sample data; Based on the training sample data, the initial convolutional neural network is trained to obtain the preset convolutional neural network.

3. The method according to claim 2, characterized in that The determining the parameters of the training foreground primitives by convolution scanning based on the training foreground primitives and the training background primitives includes: Based on the training foreground image element, a first binary image of the training foreground image element is obtained through binarization processing; Based on the training background image element, a second binarized image of the training background image element is obtained through binarization processing; Based on the first binarized image and the second binarized image, parameters of the training foreground primitive are determined.

4. The method according to claim 3, characterized in that: The step of obtaining a first binary image of the training foreground image element through binarization based on the training foreground image element comprises: Based on the training foreground primitive, obtaining multi-size training foreground sub-primitives through scaling processing; Based on each foreground sub-pixel of the multi-size training foreground sub-pixel, obtain a multi-angle training foreground sub-pixel of each foreground sub-pixel through rotation processing; Based on the multi-angle training foreground sub-pixel, the first binarized image of the training foreground sub-pixel is obtained through binarization processing.

5. The method according to claim 2, characterized in that: The determining the parameters of the training foreground primitives by convolution scanning based on the training foreground primitives and the training background primitives includes: Based on the training foreground image element and the training background image element, a third binarized image of the training foreground image element and a fourth binarized image of the training background image element are obtained through binarization processing; Performing convolution scanning on the third binarized image on the fourth binarized image to obtain a product value at each step; Based on the product value and the third binarized image, parameters of the training foreground primitive are determined.

6. The method according to claim 3, characterized in that The step of determining the parameters of the training foreground primitive based on the first binarized image and the second binarized image includes: Based on the second binary image, a preset blank detection algorithm is used to perform calculations to determine a blank center position of the second binary image; Based on the blank center positions of the first binarized image and the second binarized image, parameters of the first foreground primitive are determined through convolution scanning processing.

7. The method according to claim 6, characterized in that The step of performing calculation based on the second binary image by using a preset blank detection algorithm to determine the blank center position of the second binary image includes: Based on the second binarized image, a blank area is selected as an initial scanning position; Based on the scanning initial position and the preset field strength calculation criterion, performing calculation to determine the field strength of each cell in the second binary image; Based on the field intensity of each cell in the second binarized image, a calculation is performed using a preset gradient calculation criterion to obtain the field intensity gradient of each cell; Based on the field intensity gradient of each cell, an iterative operation is performed until a maximum value of the field intensity is obtained, and a cell corresponding to the maximum value of the field intensity is determined; The cell corresponding to the maximum value of the field intensity is taken as the blank center position.

8. The method according to claim 2, characterized in that: The determining of the first foreground image primitive by type selection processing based on the multiple foreground image primitives includes: Based on the multiple foreground primitives, data management is performed by establishing a quadtree to obtain a data management structure; Based on the data management structure, searching is performed to determine adjacent graphics elements of the current foreground graphics element; Determining the type of the current foreground image element according to the type of the adjacent image element; The first foreground primitive is determined based on the type of the current foreground primitive.

9. The method according to claim 1, characterized in that: The determining of a local feature map of the second feature map based on the second feature map of the first background image element by feature extraction processing includes: Based on the second feature map of the first background image element, a preset blank detection algorithm is used to detect and obtain a blank center position of the second feature map; Based on the second feature map and the blank center position of the second feature map, a surround scanning process is performed to obtain feature points of the second feature map; Based on the feature points of the second feature map, the local feature map of the second feature map is obtained by performing window clipping processing.

10. The method according to claim 1, characterized in that The local feature map based on the first feature map and the second feature map is processed by the target detection special layer to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map, including: Based on the first feature map and the local feature map of the second feature map, a first fixed feature point of the first feature map and a second fixed feature point of the local feature map of the second feature map are obtained by processing through a bilinear interpolation method; Based on the first fixed feature points and the second fixed feature points, the first fixed dimensional vector of the first feature map and the second fixed dimensional vector of the second feature map are determined.

11. The method according to any one of claims 1 to 10, characterized in that: Based on the fixed-dimensional feature vector, a preset regression prediction network is used to perform predictive typesetting to obtain a preliminary word cloud image, including: Based on the fixed-dimensional feature vector, a preset regression prediction network is used to process the fixed-dimensional feature vector to obtain regression parameters, wherein the regression parameters include position parameters, size parameters and angle parameters. Based on the angle parameter, obtaining a predicted angle of the first foreground primitive; Based on the regression parameters, a regression calculation is performed to obtain a predicted position and size of the first foreground primitive; Based on the predicted angle of the first foreground image element, the predicted position and size of the first foreground image element, typesetting is performed to obtain the preliminary word cloud image.

12. The method according to any one of claims 1 to 10, characterized in that: The coloring process is performed on the preliminary word cloud image to determine the generated word cloud image, including: Based on the word cloud image to be generated, spatial color clustering is performed to establish a feature vector; Perform clustering based on the feature vector of the word cloud image to be generated to obtain a color center point; Based on the color center point and the preliminary word cloud image, matching is performed by the nearest matching principle to obtain a matching result; Based on the matching result, the preliminary word cloud image is colored to obtain the word cloud image.

13. The method according to claim 12, characterized in that The preliminary word cloud image includes a plurality of word cloud graphics elements; The matching result is obtained by matching the color center point and the preliminary word cloud image according to the nearest matching principle, including: Based on the color center point and the multiple word cloud primitives, matching is performed by the nearest matching principle, and if the word cloud primitive matches a unique color center, a matching result is obtained; Based on the color center point and the multiple word cloud primitives, matching is performed using a nearest matching principle. If the word cloud primitive matches at least two color centers, a matching result is determined using a maximum weight matching algorithm.

14. A word cloud image generating device, characterized in that: It includes an acquisition unit and a determination unit; wherein, The acquisition unit is used to acquire a word cloud image to be generated; wherein the word cloud image to be generated includes a first foreground image element and a first background image element; based on the first foreground image element and the first background image element, feature extraction is performed through a preset convolutional neural network to obtain a first feature map of the first foreground image element and a second feature map of the first background image element; The determining unit is used to determine a local feature map of the second feature map through feature extraction processing based on the second feature map of the first background image element; based on the local feature maps of the first feature map and the second feature map, perform processing through a target detection special layer to obtain a first fixed-dimensional vector of the first feature map and a second fixed-dimensional vector of the second feature map; based on the first fixed-dimensional vector and the second fixed-dimensional vector, perform fusion processing to obtain a fixed-dimensional feature vector; The acquisition unit is used to perform predictive typesetting based on the fixed-dimensional feature vector through a preset regression prediction network to obtain a preliminary word cloud image; The determining unit is used to perform coloring processing on the preliminary word cloud image to determine the generated word cloud image.

15. A word cloud image generating device, characterized in that: include: A memory for storing executable instructions; A processor, for implementing the method for generating a word cloud image according to any one of claims 1 to 13 when executing executable instructions stored in the memory.

16. A computer-readable storage medium, characterized in that: The storage medium stores executable instructions, which, when executed, cause the processor to execute the method for generating a word cloud image as described in any one of claims 1 to 13.

Images