WO2020073497A1

WO2020073497A1 - Chinese language training image generation method and apparatus, computer device, and storage medium

Info

Publication number: WO2020073497A1
Application number: PCT/CN2018/122993
Authority: WO
Inventors: 黄泽浩
Original assignee: 平安科技（深圳）有限公司
Priority date: 2018-10-11
Filing date: 2018-12-24
Publication date: 2020-04-16
Also published as: CN109255826B; CN109255826A

Abstract

Disclosed are a Chinese language training image generation method and apparatus, a computer device, and a storage medium. The Chinese language training image generation method comprises: obtaining a training image generation request, wherein the training image generation request comprises a scene application requirement; based on the scene application requirement, obtaining an original background image from a precreated background image library; obtaining Chinese characters from a precreated Chinese character library; performing transparent processing on the original background image to obtain a first transparent image; adding the Chinese characters to the first transparent image to obtain a second transparent image, and labeling the second transparent image with Chinese characters to obtain a text file corresponding to the second transparent image; adding a noise point to the second transparent image to obtain a third transparent image, performing superposition on the third transparent image and the original background image to obtain an image to be trained, and storing the image to be trained in association with the text file. This process does not require collection of a training image manually, thereby improving the efficiency.

Description

Chinese training image generation method, device, computer equipment and storage medium

This patent application is based on the Chinese invention patent application filed on October 11, 2018 with the application number 201811182135.9 and titled "Chinese Training Image Generation Method, Device, Computer Equipment, and Storage Media", and claims its priority.

Technical field

The present application relates to the field of image recognition technology, and in particular, to a Chinese training image generation method, device, computer equipment, and storage medium.

Background technique

With the rapid development of the information age, artificial intelligence technology has also been gradually applied to various practical scenarios. Among them, OCR (Optical Character Recognition, optical character recognition) technology is currently the most commonly used technology for analyzing and recognizing image files to obtain text and layout information. However, when the OCR recognition technology is used to train the image recognition model, it is necessary to manually collect the training images and mark them to form a training set, and then perform model training based on the marked training set, which is time-consuming and labor-intensive.

Summary of the invention

Embodiments of the present application provide a method, device, computer equipment, and storage medium for generating a Chinese training image, to solve the problem that in the current image recognition model training process, training images need to be collected manually and labeled to form a training set, which is time-consuming and labor The problem of high cost.

A Chinese training image generation method, including:

Obtain a training image generation request, which includes scene application requirements;

Based on the scene application requirements, obtain the original background images corresponding to the scene application requirements from the pre-created background image library; obtain the Chinese characters corresponding to the scene application requirements from the pre-created Chinese character library ;

Performing transparency processing on the original background image to obtain a first transparent image;

Filling the first transparent image with the Chinese character, obtaining a second transparent image, using the Chinese character to mark the second transparent image, and obtaining a text file corresponding to the second transparent image;

Adding noise to the second transparent image, obtaining a third transparent image, superimposing the third transparent image and the original background image, obtaining an image to be trained, and associating the image to be trained with the text file storage.

A Chinese training image generation device, including:

A training image generation request acquisition module, configured to acquire a training image generation request, where the training image generation request includes scene application requirements;

The scene application requirement processing module is used to obtain the original background image corresponding to the scene application requirement from the pre-created background image library based on the scene application requirement; to obtain the original background image from the pre-created Chinese character library Chinese characters corresponding to scene application requirements;

A first transparent image acquisition module, configured to perform transparency processing on the original background image to obtain a first transparent image;

A second transparent image acquisition module, configured to fill the Chinese character on the first transparent image, acquire a second transparent image, use the Chinese character to mark the second transparent image, and acquire the second transparent image The text file corresponding to the two transparent images;

The image-to-be-trained acquisition module is used to add noise to the second transparent image, obtain a third transparent image, perform superposition processing on the third transparent image and the original background image, obtain the image to be trained, and convert the The training image is stored in association with the text file.

A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the following steps are implemented:

One or more non-volatile readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps:

The training image generation request acquisition module is used to obtain a training image generation request, and the training image generation request includes scene application requirements;

The details of one or more embodiments of the present application are set forth in the following drawings and description, and other features and advantages of the present application will become apparent from the description, drawings, and claims.

BRIEF DESCRIPTION

In order to more clearly explain the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application For those of ordinary skill in the art, without paying creative labor, other drawings can also be obtained based on these drawings.

1 is a schematic diagram of an application environment of a method for generating Chinese training images in an embodiment of the present application;

2 is a flowchart of a method for generating Chinese training images in an embodiment of the present application;

FIG. 3 is a specific flowchart of step S20 in FIG. 2;

FIG. 4 is a specific flowchart of step S30 in FIG. 2;

FIG. 5 is a specific flowchart of step S40 in FIG. 2;

6 is a schematic diagram of a Chinese training image generation device in an embodiment of the present application;

7 is a schematic diagram of a computer device in an embodiment of the present application.

detailed description

The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of this application.

The Chinese training image generation method provided in this application can be applied in the application environment as shown in FIG. 1. The Chinese training image generation method can be applied in a Chinese training image generation tool for automatically generating Chinese training images, saving manual data collection and Mark the time to improve efficiency. Among them, Chinese training image generation tools include servers and computer equipment. Among them, the computer equipment communicates with the server through the network. The computer device may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server can be implemented with an independent server.

In an embodiment, as shown in FIG. 2, a method for generating a Chinese training image is provided. The method is applied to the server in FIG. 1 as an example for illustration, which includes the following steps:

S10: Obtain a training image generation request, which includes scene application requirements.

The training image generation request is a request for triggering a Chinese training image generation tool to generate a training image. The scene application requirement refers to the requirement to generate training images of the application in a specific scene. Specifically, a list of scene types is displayed on the display interface of the Chinese training image generation tool. The list of scene types includes Hong Kong ID cards, second-generation ID cards, airline boarding passes (such as Air China) and the front of each bank card (Such as: Industrial and Commercial Bank of China) and other scene types are convenient for users to choose. Determine the scene application requirements according to the scene type selected by the user, so that the server obtains the training image generation request and obtains the scene application requirements according to the training image generation request.

S20: Based on the scene application requirements, obtain the original background images corresponding to the scene application requirements from the pre-created background image library, and obtain the Chinese characters corresponding to the scene application requirements from the pre-created Chinese character library.

Among them, the background image library is an image library of background images uploaded or produced in advance by taking pictures. The background image library includes scene images and non-scene images. The scene image refers to the background image applied in a specific scene. The scene images include, but are not limited to, the Hong Kong ID card image, the second-generation ID card image, the airline boarding pass images, and the bank card front images provided by this embodiment. Non-scene images refer to background images applied to non-specific scenes, such as background images of different colors. The original background image is based on the scene application requirements, and the server obtains the background image corresponding to the scene application requirements (such as the second-generation ID card) from the pre-created background image library.

The Chinese character library includes commonly used Chinese first-level character libraries, hundred family names and traditional character libraries. If you want to generate a Hong Kong identity card, you can obtain the corresponding traditional Chinese characters from the traditional Chinese character library; if you do not need to generate traditional Chinese characters, you can obtain the corresponding Chinese characters from the Chinese first-level character library. Understandably, Chinese characters include traditional characters or simplified characters. The Chinese characters are based on the scene application requirements, and the Chinese characters corresponding to the application requirements are obtained from the pre-created Chinese character library.

In this embodiment, if the scene usage requirement is a second-generation ID card, the server obtains the original background image corresponding to the scene application demand, that is, the ID card background image, from the pre-created background image library based on the scene application demand, A good Chinese character library can obtain Chinese characters (such as names) corresponding to the application requirements of the scene. This process does not require manual collection of original background images and editing of Chinese characters, saving time and providing technical support for the subsequent generation of training images.

S30: Transparency the original background image to obtain the first transparent image.

Specifically, in order to highlight the subsequent effect of adding noise to the image, the background image needs to be transparentized first to obtain the original transparent image. Transparency processing includes but is not limited to using pillow library technology for processing. Among them, Pillow is an image processing library (PIL: Python Image Library) in the Python tool, which provides extensive file format support and powerful image processing capabilities. It mainly provides image storage, image display, format conversion, and basic image processing operations. The interface can be directly called, which is simple to implement and can effectively save the time of repeated development.

S40: Fill the first transparent image with Chinese characters, obtain a second transparent image, mark the second transparent image with Chinese characters, and obtain a text file corresponding to the second transparent image.

The second transparent image refers to a transparent image filled with Chinese characters corresponding to scene application requirements in the first transparent image. The text file refers to a label file corresponding to the second transparent image. When generating a training image, N (N is a positive integer greater than 1 and can be specified by the user) original background images are obtained. The server uses the pillow library technology to randomly select the first transparent image corresponding to the original background image to convert the selected Chinese The word is filled on the selected first transparent image to obtain the second transparent image. At the same time, the server will use the selected Chinese characters to mark the second transparent image to obtain the text file corresponding to the second transparent image. This process does not require Manual labeling, you can automatically label.

S50: Add noise to the second transparent image, obtain the third transparent image, superimpose the third transparent image and the original background image, obtain the image to be trained, and store the image to be trained in association with the text file.

The third transparent image is a transparent image after adding noise to the second transparent image. Specifically, the server randomly selects a predetermined proportion of pixels that need to increase noise, so as to randomly increase noise on the proportion of pixels, so as to increase the robustness of the training image. Then, the third transparent image and the original background image are superimposed to obtain the image to be trained, and the image to be trained is stored in association with the text file to form a training sample, so that the training sample is used for model training, and the step of manual collection is omitted ,Improve efficiency. Superimposition processing refers to the process of superimposing the third transparent image and the original background image into one image, so as to obtain the image to be trained. In this embodiment, the imadd function is used to superimpose the third transparent image and the original background image to obtain the image to be trained. The imadd function is a function in computer language used to superimpose images.

The types of noise include but are not limited to reflection, interference lines, interference color point, tilt angle (including two tilt methods, each tilt method includes three tilt angles: 0.5, 1 and 1.5), dilation, corrosion and Gaussian blur etc. . Taking the example of increasing interference color points, randomly selecting pixels of a preset ratio and setting the selected pixels to black can complete the purpose of increasing noise. Among them, the preset ratio is the ratio automatically recommended by the Chinese training image mixing generation tool according to the experience value, and supports user changes. There are two ways to change it: one is to change the proportion of pixels with increased noise; the other is to change and increase The number of pixels of noise. In this embodiment, the processing of expansion and corrosion is judged according to the font to be generated, taking the generation of a Hong Kong identity card as an example. For conventional fonts, expansion processing can be selected due to the thinner lines of the conventional font, while for bold fonts, Because the lines are thicker, corrosion treatment can be selected to enhance the clarity of the training image.

In this embodiment, the server first obtains the training image generation request, so as to obtain the original background image corresponding to the scene application requirement from the pre-created background image library based on the scene application requirements in the training image generation request, and create The Chinese characters corresponding to the application requirements of the scene are obtained from the Chinese character library in this database. This process does not require manual collection of original background images and editing of Chinese characters, saving time. Transparency is performed on the original background image to obtain the first transparent image to highlight the effect of increasing noise in subsequent images. Then, fill the first transparent image with Chinese characters to obtain the second transparent image, and at the same time, use the Chinese characters to label the second transparent image to obtain the text file corresponding to the second transparent image. This process does not require manual labeling, that is, The second transparent image can be automatically annotated. Finally, add noise to the second transparent image, obtain the third transparent image, superimpose the third transparent image and the original background image to obtain the image to be trained, and increase the authenticity of the image to be trained to improve the subsequent use of the image to be trained The recognition accuracy of the model obtained by training. Finally, the training image is associated and stored with the text file to form a training sample, so that the training sample is used for training without manual collection, and the efficiency is improved.

In one embodiment, as shown in FIG. 3, in step S20, that is, based on the scene application requirements, the original background image corresponding to the scene application requirements is obtained from the pre-created background image library, and the pre-created Chinese character library is obtained. Obtain the Chinese characters corresponding to the application requirements of the scene, including the following steps:

S21: If the scene application requirement is the first application requirement, the original background image corresponding to the first application requirement is obtained from the background image library. The original background image includes the scene field, and based on the scene field, according to the preset generation rule, the Chinese characters The Chinese characters corresponding to the scene fields are obtained from the library.

Among them, the first application requirement refers to generating training images that are applied in specific scenarios, such as second-generation ID card images and bank card front images. Specifically, if the scene application requirement is the first application requirement, the original background image corresponding to the first application requirement is obtained from the background image library. The original background image includes a scene field (such as a name), based on the scene field, according to the preset Generate rules to get the Chinese characters corresponding to the scene field from the Chinese character library. The preset generation rule is a rule set in advance for generating attribute values corresponding to each scene field. For example: if the first application requirement is the second-generation ID card image, the server will obtain the second-generation ID card image from the background image library as the original background image based on the first application requirement. Scene fields such as year, month, date, address and ID number. Based on the scene fields, the Chinese characters corresponding to each scene field are obtained from the Chinese character library according to the preset generation rules. This process requires no human intervention and saves labor costs.

For the scene field of name, since the names of ethnic minority groups currently contain long characters, the preset generation rule of the name field in this embodiment is limited to 10 characters.

For the scene field of gender, it can only be randomly obtained from male / female, so the corresponding preset generation rule is one of the two characters male / female.

For the date of birth, the default generation rules are set according to the date format.

For residential addresses, address data crawled from the existing address database by web crawlers can be used. These address data basically conform to their corresponding preset generation rules.

The preset generation rules for ID card numbers are as follows: Since the ID card number structure has a fixed format, the ID number is a combination code of features, consisting of a 17-digit numeric body code and a check code. The arrangement order from left to right is: six-digit address code, eight-digit birth date code, three-digit sequence code and one-digit check code.

The address code (first six digits) indicates the administrative division code of the county (city, flag, district) where the permanent residence of the encoding object is located, and shall be implemented in accordance with the provisions of GB / T2260. In this case, the region and the region code will be associated first, and then the region and the corresponding region code will be randomly obtained. 7-14 digits are the year and month of birth, randomly generated according to the date format. 15 to 17 bits are sequential codes, which are generated according to the random number generation method. The last digit check code is generated according to the check code rule. The date of birth code indicates the year, month, and day of birth of the coding object, and is implemented in accordance with the provisions of GB / T7408. There is no separator between the year, month, and day codes. The sequence code indicates the sequence number assigned to people born in the same year, month, and day within the area identified by the same address code. The odd number of the sequence code is assigned to men, and the even number is assigned to women.

The verification code acquisition process includes the following steps:

1) The weighted sum formula of the seventeen digit body code S = Sum (Ai * Wi), i = 0, ..., 16, first sum the weights of the first 17 digits, where Ai: indicates the i-th position The digital value of the ID number on the Wi; Wi: indicates the weighting factor at the i-th position Wi: 7 9 9 10 8 5 4 2 1 3 6 7 9 9 10 8 8 4 2

2) Modulus calculation: Y = mod (S, 11).

3) Get the corresponding check code Y: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 by taking the modulus, and the check code: 1, 0, X, 9, 8, 76, 5, 4, 3, 2 * /

For example, the calculation method of the eighteenth digit (check code) is: 1. Multiply the 17 digits of the previous ID number by different coefficients. The coefficients from the first place to the seventeenth place are: 7 9 10 10 5 8 8 4 2 6 3 3 7 9 10 8 5. 2. Add the result of multiplying the 17 digits and the coefficient. 3. Divide the sum by 11, and see what is the remainder? 4. The remainder can only have 11 numbers of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. The number of the last ID card corresponding to it is 1 0 0 X 9 9 8 8 5 5 4 3 2. 5. If the remainder is 2, the 18th digit of the ID will be the Roman numeral X. If the remainder is 10, the last number of the ID card is 2.

S22: If the scene application requirement is the second application requirement, the original background image is randomly obtained from the background image library, and the Chinese characters are randomly obtained from the Chinese character library.

Among them, the second application requirement refers to generating training images that are applied in non-specific scenarios, such as Chinese character images. Because this type of Chinese character image is only used to train the OCR Chinese character recognition model in non-specific scenes, the original background image can be randomly obtained directly from the background image library, and the corresponding Chinese character can be randomly obtained from the Chinese character library, which is simple and convenient.

In this embodiment, if the scene application requirement is the first application requirement, the original background image corresponding to the first application requirement is obtained from the background image library, so that based on the scene field in the original background image, a preset generation rule is used, Obtain the Chinese characters corresponding to the scene field from the Chinese character library without manual intervention, saving labor costs. If the scene application requirement is the second application requirement, it is simple and convenient to randomly obtain the original background image from the direct background image library and randomly obtain the Chinese characters from the Chinese character library.

In an embodiment, the scene field includes a name field; the Chinese character library includes a hundred family names and a Chinese first-level character library. In step S21, the Chinese characters corresponding to the scene fields are obtained from the Chinese character library according to the preset generation rules based on the scene fields, which specifically includes:

Based on the name field, the surnames are obtained sequentially or randomly from the hundred family surnames, and the Chinese characters are obtained sequentially or randomly from the first-class Chinese character library.

In this embodiment, there are two generation rules for the name field. One is to obtain the surnames from the surnames of the surnames in the order of the surnames, and then obtain the Chinese characters in sequence from the first-class Chinese character library and splice the surnames with the Chinese characters , You can obtain the Chinese characters corresponding to the name field, and improve the efficiency of obtaining the attribute value corresponding to the name field. Or, randomly select the surnames from the surnames as the surnames corresponding to the name field, randomly select the Chinese characters from the first-class Chinese character library, and stitch the selected surnames with the Chinese characters to obtain the Chinese characters corresponding to the name field Get the diversity of attribute values corresponding to the name field.

Further, in practical applications, the corresponding surnames can also be selected according to the proportion of the number of various surnames currently counted by relevant agencies, and the Chinese characters can be selected from commonly used Chinese characters and randomly combined to ensure the diversity of their combinations. It can improve the authenticity and reliability of the image recognition model obtained by training with the obtained training images.

It should be noted that the Chinese character library also includes a traditional character library. If you want to generate a Hong Kong identity card, you do not need to obtain simplified Chinese characters from the Chinese first-level character library. You can directly obtain the corresponding traditional characters from the traditional character library. For the family names, the family names used in step S21 are in simplified Chinese characters. If you want to generate a Hong Kong identity card, you can obtain the family names from the family names in traditional Chinese characters. Splicing, you can get the Chinese characters corresponding to the name field.

In an embodiment, as shown in FIG. 4, in step S30, the original background image is transparentized to obtain the first transparent image, which specifically includes the following steps:

S31: Perform mode conversion on the original background image to obtain a mode image, and the mode image includes color parameters.

Among them, the mode image refers to a true color image mode with transparency (abbreviated as RGBA mode). It should be noted that the image mode of the original background image itself is the RGB mode (that is, the color image mode). Specifically, the following method PIL.Image.new (mode, size, color = 0) can be used to convert the image mode of the original background image to RGBA mode, where the mode parameter defines some attributes of pixels in the image, such as Transparency true color RGBA. The size parameter specifies the length and width of the image in pixels. The color parameter is the color parameter used to define the background color of the image (ie, the original background image). Among them, RGBA mode is Red (red), Green (green), Blue (blue) and Alpha color space mode, that is, transparency.

S32: Set the color parameter of the mode image to empty, and obtain the first transparent image.

Specifically, when the image mode is the RGBA mode, if the color parameters of the mode image are not specified, the server defaults to a transparent background, and the first transparent image is obtained, which is simple to implement and improves the efficiency of generating training images.

In this embodiment, the server first converts the original background image to obtain a pattern image with transparency. By setting the color parameter in the pattern image to empty, the first transparent image is obtained, which is simple to implement and improves the efficiency of generating training images .

In an embodiment, as shown in FIG. 5, in step S40, the Chinese characters are filled on the first transparent image to obtain the second transparent image, which specifically includes the following steps:

S41: Obtain the attribute parameters corresponding to the Chinese characters.

Among them, the attribute parameters corresponding to the Chinese characters include the position, content, color and font of the Chinese characters to be filled in the first transparent image. This attribute parameter is set in advance according to different scene application requirements. Understandably, if the scenario application requirement is the first application requirement, the setting is performed according to the actual application scenario. For example, if the first application requires a second-generation ID card, the attribute parameters corresponding to the Chinese characters are set according to the text attributes in the actual ID card image to fit the reality and improve the authenticity and reliability of the training image. For example, if the scene application requirement is the second application requirement, you can randomly obtain the attribute parameters corresponding to the Chinese characters. For example, if you want to generate a Chinese character image, you can randomly select the corresponding font from the pre-stored fonts (such as Kai and Song), or Can be customized by the user. The text content, text color, and text position can also be randomly obtained by the server or customized by the user, which improves the practicality of the Chinese training image generation tool.

S42: Apply the attribute parameters to the text filling function to fill the first transparent image with Chinese characters and obtain the second transparent image.

Specifically, based on the setting of the attribute parameters, the server applies the attribute parameters to the text filling function based on the image processing technology (namely, the pillow library technology) to fill the Chinese characters on the first transparent image and obtain the second transparent image. Specifically, the server uses the following text filling function "draw.text ((40,10), u, font = myfont, fill = fillcolor)" to fill the first transparent image with Chinese text based on the attribute parameters to obtain transparency image. Understandably, "(40,10), u, font = myfont, fill = fillcolor" means attribute parameters; draw.text () means text fill function. Among them, the first parameter (40,10) represents the text position; the second parameter u represents the text content; the third parameter font represents the text font, and the fourth parameter fill represents the text color. The server automatically fills in the above sentence to obtain the second transparent image without manual intervention, and realizes the purpose of automatically generating the training image.

In this embodiment, the server obtains the attribute parameters corresponding to the Chinese characters, so that based on the attribute parameters, the image processing interface provided by the pillow library technology is used to fill the Chinese characters on the first transparent image and obtain the second transparent image. Manual intervention to achieve the purpose of automatically generating training images.

It should be understood that the size of the sequence numbers of the steps in the above embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

In an embodiment, a Chinese training image generating device is provided, and the Chinese training image generating device corresponds one-to-one to the Chinese training image generating method in the foregoing embodiment. As shown in FIG. 6, the Chinese training image generation device includes a training image generation request acquisition module 10, a scene application demand processing module 20, a first transparent image acquisition module 30, a second transparent image acquisition module 40, and a to-be-trained image acquisition module 50 . The detailed description of each functional module is as follows:

The training image generation request obtaining module 10 is used to obtain a training image generation request, and the training image generation request includes scene application requirements.

The scene application requirement processing module 20 is used to obtain the original background image corresponding to the scene application requirement from the pre-created background image library based on the scene application requirement; to obtain the scene application requirement corresponding to the scene application requirement from the pre-created Chinese character library Chinese characters.

The first transparent image acquisition module 30 is configured to perform transparency processing on the original background image to acquire the first transparent image.

The second transparent image obtaining module 40 is used to fill Chinese characters on the first transparent image, obtain the second transparent image, mark the second transparent image with Chinese characters, and obtain a text file corresponding to the second transparent image.

The image-to-be-trained acquisition module 50 is used to add noise to the second transparent image, obtain the third transparent image, superimpose the third transparent image and the original background image, obtain the image to be trained, and store the image to be trained in association with the text file .

Specifically, the scene application requirement processing module includes a first processing unit and a second processing unit.

The first processing unit is used to obtain an original background image corresponding to the first application requirement from the background image library if the scene application requirement is the first application requirement, the original background image includes a scene field; based on the scene field, the preset Generate rules to get the Chinese characters corresponding to the scene field from the Chinese character library.

The second processing unit is configured to randomly obtain original background images from the background image library and randomly obtain Chinese characters from the Chinese character library if the scene application requirement is the second application requirement.

Specifically, the first processing unit is specifically: based on the name field, sequentially or randomly obtain the surnames from the hundred family surnames, and sequentially or randomly obtain the Chinese characters from the first-class Chinese character library; Corresponding Chinese characters.

Specifically, the first transparent image acquisition module includes an image mode conversion unit and a first transparent image acquisition unit.

The image mode conversion unit is used for mode conversion of the original background image to obtain a mode image; the mode image includes color parameters.

The first transparent image acquisition unit is configured to set the color parameter of the pattern image to be empty and acquire the first transparent image.

Specifically, the second transparent image acquisition module includes an attribute parameter acquisition unit and a second transparent image acquisition unit.

The attribute parameter obtaining unit is used to obtain attribute parameters corresponding to Chinese characters.

The second transparent image obtaining unit is used to apply attribute parameters to the text filling function to fill the first transparent image with Chinese characters and obtain the second transparent image.

For the specific definition of the Chinese training image generating device, please refer to the above definition of the Chinese training image generating method, which will not be repeated here. Each module in the above-mentioned Chinese training image generating device may be implemented in whole or in part by software, hardware, and a combination thereof. The above modules may be embedded in the hardware or independent of the processor in the computer device, or may be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure may be as shown in FIG. 7. The computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer device is used to store data generated or acquired during the execution of the Chinese training image generation method, such as the image to be trained. The network interface of the computer device is used to communicate with external terminals through a network connection. When the computer readable instructions are executed by the processor to implement a Chinese training image generation method.

In one embodiment, a computer device is provided, which includes a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor. The processor implements the computer-readable instructions to implement the The steps of the Chinese training image generation method, such as steps S10-S50 shown in FIG. 2, or the steps shown in FIGS. 3 to 5. Or, the processor implements the functions of each module / unit in the embodiment of the Chinese training image generating device when executing the computer-readable instructions, for example, the function of each module / unit shown in FIG. 6, to avoid repetition, it will not be repeated here .

In an embodiment, a computer-readable storage medium is provided, and the computer-readable storage medium stores computer-readable instructions, which when executed by a processor, implements the steps of the method for generating a Chinese training image in the foregoing embodiments For example, steps S10-S50 shown in FIG. 2 or steps shown in FIG. 3 to FIG. 5, in order to avoid repetition, they will not be repeated here. Alternatively, when the computer-readable instructions are executed by the processor, the functions of each module / unit in the above embodiment of the Chinese training image generation device, such as the functions of each module / unit shown in FIG. 6, are implemented. To avoid repetition, here No longer.

A person of ordinary skill in the art may understand that all or part of the process in the method of the foregoing embodiments may be completed by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions may be stored in a non-volatile computer In the readable storage medium, when the computer-readable instructions are executed, they may include the processes of the foregoing method embodiments. Wherein, any reference to the memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Those skilled in the art can clearly understand that, for convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example for illustration. In practical applications, the above-mentioned functions can be allocated by different functional units, Module completion means that the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still implement the foregoing The technical solutions described in the examples are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application. Within the scope of protection of this application.

Claims

A Chinese training image generation method, which is characterized by including:

Obtain a training image generation request, which includes scene application requirements;

Based on the scene application requirements, obtain the original background images corresponding to the scene application requirements from the pre-created background image library; obtain the Chinese characters corresponding to the scene application requirements from the pre-created Chinese character library ;

Performing transparency processing on the original background image to obtain a first transparent image;

Filling the first transparent image with the Chinese character, obtaining a second transparent image, using the Chinese character to mark the second transparent image, and obtaining a text file corresponding to the second transparent image;

Adding noise to the second transparent image, obtaining a third transparent image, superimposing the third transparent image and the original background image, obtaining an image to be trained, and associating the image to be trained with the text file storage.
The Chinese training image generation method according to claim 1, wherein the original background image corresponding to the scene application requirement is obtained from a background image library created in advance based on the scene application requirement; Obtain Chinese characters corresponding to the application requirements of the scene from the pre-created Chinese character library, including:

If the scene application requirement is the first application requirement, an original background image corresponding to the first application requirement is obtained from the background image library, the original background image includes a scene field; based on the scene field, Obtain the Chinese characters corresponding to the scene field from the Chinese character library according to a preset generation rule;

If the scene application requirement is the second application requirement, the original background image is randomly obtained from the background image library, and the Chinese characters are randomly obtained from the Chinese character library.
The method for generating a Chinese training image according to claim 2, wherein the scene field includes a name field; the Chinese character library includes a hundred family names and a Chinese first-level character library;

Based on the scene field, according to a preset generation rule, obtaining the Chinese character corresponding to the scene field from the Chinese character library includes:

Based on the name field, sequentially or randomly obtaining surnames from the hundred family surnames, and sequentially or randomly obtaining Chinese characters from the first-class Chinese character library;

Join the surname and the Chinese character to obtain the Chinese character corresponding to the scene field.
The method for generating a Chinese training image according to claim 1, wherein the process of transparentizing the original background image to obtain the first transparent image includes:

Performing mode conversion on the original background image to obtain a mode image; the mode image includes color parameters;

Set the color parameter of the mode image to null to obtain the first transparent image.
The method for generating a Chinese training image according to claim 1, wherein the filling the Chinese character on the first transparent image and obtaining the second transparent image includes:

Obtain the attribute parameters corresponding to the Chinese characters;

The attribute parameter is applied to a text filling function to fill the Chinese text on the first transparent image and obtain a second transparent image.
A Chinese training image generating device, characterized in that it includes:

A training image generation request acquisition module, configured to acquire a training image generation request, where the training image generation request includes scene application requirements;

The scene application requirement processing module is used to obtain the original background image corresponding to the scene application requirement from the pre-created background image library based on the scene application requirement; to obtain the original background image from the pre-created Chinese character library Chinese characters corresponding to scene application requirements;

A first transparent image acquisition module, configured to perform transparency processing on the original background image to obtain a first transparent image;

A second transparent image acquisition module, configured to fill the Chinese character on the first transparent image, acquire a second transparent image, use the Chinese character to mark the second transparent image, and acquire the second transparent image The text file corresponding to the two transparent images;

The image-to-be-trained acquisition module is used to add noise to the second transparent image, obtain a third transparent image, perform superposition processing on the third transparent image and the original background image, obtain the image to be trained, and convert the The training image is stored in association with the text file.
The Chinese training image generating device according to claim 6, wherein the scene application requirement processing module includes:

A first processing unit, configured to obtain an original background image corresponding to the first application requirement from the background image library if the scene application requirement is the first application requirement, the original background image including a scene field ; Based on the scene field, according to a preset generation rule, obtain the Chinese character corresponding to the scene field from the Chinese character library;

The second processing unit is configured to randomly obtain original background images from the background image library and randomly obtain Chinese characters from the Chinese character library if the scene application requirement is the second application requirement.
The Chinese training image generation device according to claim 6, wherein the second transparent image acquisition module includes:

An attribute parameter obtaining unit, configured to obtain attribute parameters corresponding to the Chinese characters;

The second transparent image obtaining unit is configured to apply the attribute parameter to a text filling function to fill the Chinese character on the first transparent image and obtain a second transparent image.
The Chinese training image generation device according to claim 6, wherein the first transparent image acquisition module includes:

An image mode conversion unit, configured to perform mode conversion on the original background image to obtain a mode image; the mode image includes color parameters;

The first transparent image acquisition unit is configured to set the color parameter of the pattern image to be empty and acquire the first transparent image.
The Chinese training image generating device according to claim 7, wherein the scene field includes a name field; the Chinese character library includes a hundred family names and a Chinese first-level character library;

The first processing unit is specifically: based on the name field, sequentially or randomly obtaining surnames from the hundred family surnames, sequentially or randomly obtaining Chinese characters from the first-class Chinese character library; The Chinese characters are stitched together to obtain the Chinese characters corresponding to the scene field.
A computer device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that, when the processor executes the computer-readable instructions, it is implemented as follows step:

Obtain a training image generation request, which includes scene application requirements;

Based on the scene application requirements, obtain the original background images corresponding to the scene application requirements from the pre-created background image library; obtain the Chinese characters corresponding to the scene application requirements from the pre-created Chinese character library ;

Performing transparency processing on the original background image to obtain a first transparent image;

Filling the first transparent image with the Chinese character, obtaining a second transparent image, using the Chinese character to mark the second transparent image, and obtaining a text file corresponding to the second transparent image;

Adding noise to the second transparent image, obtaining a third transparent image, superimposing the third transparent image and the original background image, obtaining an image to be trained, and associating the image to be trained with the text file storage.
The computer device according to claim 11, characterized in that, based on the scene application requirements, the original background image corresponding to the scene application requirements is obtained from a pre-created background image library; The Chinese characters in the Chinese character library corresponding to the application requirements of the scene include:

If the scene application requirement is the first application requirement, an original background image corresponding to the first application requirement is obtained from the background image library, the original background image includes a scene field; based on the scene field, Obtain the Chinese characters corresponding to the scene field from the Chinese character library according to a preset generation rule;

If the scene application requirement is the second application requirement, the original background image is randomly obtained from the background image library, and the Chinese characters are randomly obtained from the Chinese character library.
The computer device according to claim 12, characterized in that the scene field includes a name field; the Chinese character library includes a hundred family names and a Chinese first-level character library;

Based on the scene field, according to a preset generation rule, obtaining the Chinese character corresponding to the scene field from the Chinese character library includes:

Based on the name field, sequentially or randomly obtaining surnames from the hundred family surnames, and sequentially or randomly obtaining Chinese characters from the first-class Chinese character library;

Join the surname and the Chinese character to obtain the Chinese character corresponding to the scene field.
The computer device according to claim 11, wherein the transparentizing the original background image to obtain the first transparent image includes:

Performing mode conversion on the original background image to obtain a mode image; the mode image includes color parameters;

Set the color parameter of the mode image to null to obtain the first transparent image.
The computer device according to claim 11, wherein the filling of the Chinese characters on the first transparent image to obtain the second transparent image includes:

Obtain the attribute parameters corresponding to the Chinese characters;

The attribute parameter is applied to a text filling function to fill the Chinese text on the first transparent image and obtain a second transparent image.
A non-volatile storage medium that stores computer-readable instructions, characterized in that the computer-readable instructions are executed by a processor to implement the following steps:

Obtain a training image generation request, which includes scene application requirements;

Based on the scene application requirements, obtain the original background images corresponding to the scene application requirements from the pre-created background image library; obtain the Chinese characters corresponding to the scene application requirements from the pre-created Chinese character library ;

Performing transparency processing on the original background image to obtain a first transparent image;

Filling the first transparent image with the Chinese character, obtaining a second transparent image, using the Chinese character to mark the second transparent image, and obtaining a text file corresponding to the second transparent image;

Adding noise to the second transparent image, obtaining a third transparent image, superimposing the third transparent image and the original background image, obtaining an image to be trained, and associating the image to be trained with the text file storage.
The non-volatile readable storage medium according to claim 16, wherein the original background corresponding to the scene application requirement is obtained from a pre-created background image library based on the scene application requirement Image; obtain Chinese characters corresponding to the application requirements of the scene from the pre-created Chinese character library, including:

If the scene application requirement is the first application requirement, an original background image corresponding to the first application requirement is obtained from the background image library, the original background image includes a scene field; based on the scene field, Obtain the Chinese characters corresponding to the scene field from the Chinese character library according to a preset generation rule;

If the scene application requirement is the second application requirement, the original background image is randomly obtained from the background image library, and the Chinese characters are randomly obtained from the Chinese character library.
The non-volatile readable storage medium according to claim 17, wherein the scene field includes a name field; the Chinese character library includes a hundred family names and a Chinese first-level character library;

Based on the scene field, according to a preset generation rule, obtaining the Chinese character corresponding to the scene field from the Chinese character library includes:

Based on the name field, sequentially or randomly obtaining surnames from the hundred family surnames, and sequentially or randomly obtaining Chinese characters from the first-class Chinese character library;

Join the surname and the Chinese character to obtain the Chinese character corresponding to the scene field.
The non-volatile readable storage medium according to claim 16, wherein the performing a transparent process on the original background image to obtain the first transparent image includes:

Performing mode conversion on the original background image to obtain a mode image; the mode image includes color parameters;

Set the color parameter of the mode image to null to obtain the first transparent image.
The non-volatile readable storage medium according to claim 16, wherein the filling the Chinese character on the first transparent image and obtaining the second transparent image includes:

Obtain the attribute parameters corresponding to the Chinese characters;

The attribute parameter is applied to a text filling function to fill the Chinese text on the first transparent image and obtain a second transparent image.