Character recognition method and system
By introducing a contextual word meaning scoring model into the character recognition model, and combining character recognition and semantic analysis, the problem of low character recognition accuracy in existing technologies is solved, and higher recognition accuracy is achieved.
Patent Information
- Application Number
- CN202511126788.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-12
- Publication Date
- 2025-11-21
AI Technical Summary
Existing character recognition technologies have low accuracy when dealing with complex shapes and multi-character sentences, ignoring the semantic relationships within the sentences.
The target image of the sentence to be recognized is input into the trained character recognition model to obtain the initial recognition result. Then, the result character is input into the context word sense scoring model. Combining word sense score and result probability, the character recognition result is updated.
By combining character recognition and contextual semantic analysis, the accuracy of character recognition is improved, especially in cases of complex shapes and multi-character sentences.
Smart Images
Figure CN120997836A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of character recognition, and particularly relates to a character recognition method and system. BACKGROUND
[0002] Character recognition is a technology for automatically recognizing and analyzing text characters in a graphic image by using a computer. However, in the prior art, character recognition is only performed according to the shape of a to-be-recognized character in a graphic image, and when the shape is complex, the accuracy of recognition is greatly reduced. When a sentence in a graphic image involves many to-be-recognized characters, the to-be-recognized characters are also likely to be complex, and in this case, the shape of the to-be-recognized character is simply used for recognition, and the semantic association of the sentence is ignored. Therefore, the prior art has the technical defect of low accuracy of character recognition. SUMMARY
[0003] Therefore, the present application aims to provide a character recognition method and system which can overcome the shortcomings of the prior art.
[0004] To achieve the above object, the technical scheme adopted by the present application is as follows:
[0005] A character recognition method comprises the following steps:
[0006] inputting a target image including a to-be-recognized sentence into a trained character recognition model to obtain an initial recognition result; the to-be-recognized sentence comprises a plurality of to-be-recognized characters; the initial recognition result comprises a result character with the highest result probability corresponding to each to-be-recognized character;
[0007] inputting the plurality of result characters into a trained context word sense scoring model to obtain a word sense score of each result character;
[0008] obtaining a target result character of each to-be-recognized character according to the word sense score and the result probability.
[0009] Compared with the prior art, the present application has the following beneficial effects:
[0010] In the present application, a target image including a to-be-recognized sentence is inputted into a trained character recognition model to obtain an initial recognition result, wherein the initial recognition result comprises a result character with the highest result probability corresponding to each to-be-recognized character, and then the plurality of result characters are inputted into a trained context word sense scoring model to obtain a word sense score of each result character, and finally a target result character of each to-be-recognized character is obtained according to the word sense score and the result probability. The target result character of the to-be-recognized character is obtained by combining character recognition and context semantic scoring, and the character recognition of the sentence is performed in the dimensions of character recognition and semantic analysis, so that the accuracy of character recognition can be improved.
[0011] As an implementation form, the step of inputting the plurality of result characters into the trained contextual word sense scoring model to obtain the word sense score of each result character comprises:
[0012] According to the plurality of result characters and the character order of the corresponding to-be-recognized character, a predicted sentence is obtained;
[0013] The predicted sentence is input into the trained contextual word sense scoring model to obtain the word sense score of each result character.
[0014] According to the character order, the predicted sentence is constructed and then input into the contextual word sense scoring model, so that the word sense score of the result character can be accurately obtained according to the character order.
[0015] As an implementation form, the step of obtaining the target result character of each to-be-recognized character according to the word sense score and the result probability comprises:
[0016] According to the word sense score, the result probability of the corresponding result character is updated;
[0017] According to the updated result probability and the result probability of the candidate recognition result of the to-be-recognized character, the result character with the highest probability is updated.
[0018] If the result character before and after the update is the same, the target result character of the corresponding to-be-recognized character is determined.
[0019] According to the word sense score and the result probability of the result character, the target result character can be accurately obtained through the implementation form.
[0020] As an implementation form, the step of updating the result probability of the corresponding result character according to the word sense score comprises:
[0021] According to the product of the semantic score and the result probability, the updated result probability is obtained.
[0022] Through the implementation form, the updated result probability can be accurately obtained.
[0023] As an implementation form, after the step of updating the result character with the highest probability according to the updated result probability and the result probability of the candidate recognition result of the to-be-recognized character, the step comprises:
[0024] If the result character before and after the update is different, the updated result character and the updated result character and / or the target result character of other to-be-recognized characters are input into the trained contextual word sense scoring model to obtain a new word sense score.
[0025] By the embodiment, when the result characters before and after the update are different, the new word sense score is obtained according to the context word sense scoring model, and the new word sense score can be obtained for updating the corresponding result probability when the result characters change.
[0026] A character recognition system comprises:
[0027] An initial recognition result acquisition module is configured to input a target image comprising a to-be-recognized sentence into a trained character recognition model to obtain an initial recognition result; the to-be-recognized sentence comprises a plurality of to-be-recognized characters; and the initial recognition result comprises a result character with the highest result probability corresponding to each to-be-recognized character.
[0028] A word sense score acquisition module is configured to input the plurality of result characters into a trained context word sense scoring model to obtain a word sense score of each result character.
[0029] A target result character acquisition module is configured to obtain a target result character of each to-be-recognized character according to the word sense score and the result probability.
[0030] Compared with the prior art, the character recognition system has the following advantages:
[0031] The character recognition system comprises an initial recognition result acquisition module, a word sense score acquisition module, and a target result character acquisition module. The initial recognition result acquisition module is configured to input a target image comprising a to-be-recognized sentence into a trained character recognition model to obtain an initial recognition result; the to-be-recognized sentence comprises a plurality of to-be-recognized characters; and the initial recognition result comprises a result character with the highest result probability corresponding to each to-be-recognized character. The word sense score acquisition module is configured to input the plurality of result characters into a trained context word sense scoring model to obtain a word sense score of each result character. The target result character acquisition module is configured to obtain a target result character of each to-be-recognized character according to the word sense score and the result probability. The character recognition system combines character recognition and context semantic scoring to obtain a target result character of each to-be-recognized character, and integrates character recognition and semantic analysis to recognize the characters in the sentence, thereby improving the accuracy of character recognition.
[0032] As an implementation form, the word sense score acquisition module comprises:
[0033] A predicted sentence acquisition module is configured to obtain a predicted sentence according to the plurality of result characters and the character order of the corresponding to-be-recognized characters.
[0034] A score acquisition module is configured to input the predicted sentence into a trained context word sense scoring model to obtain a word sense score of each result character.
[0035] According to the embodiment, the predicted sentence is constructed according to the character order and then input into the context word sense scoring model, so that the word sense score of the result character can be accurately obtained according to the character order.
[0036] As an implementation form, the target result character obtaining module comprises:
[0037] a result probability updating module configured to update a result probability of the result character according to the semantic score;
[0038] a result character updating module configured to update a result character with the highest probability according to the updated result probability and a result probability of a candidate recognition result of the to-be-recognized character;
[0039] a first processing module configured to determine the target result character as the target result character of the to-be-recognized character if the result character before and after the updating is the same.
[0040] According to the semantic score of the result character and the result probability, the target result character can be accurately obtained.
[0041] As an implementation form, the result probability updating module is configured to perform the following steps:
[0042] The updated result probability is obtained according to the product of the semantic score and the result probability.
[0043] According to the implementation form, the updated result probability can be accurately obtained.
[0044] As an implementation form, the character recognition method further comprises:
[0045] a second processing module configured to input the result character after the updating and the result character after the updating and / or the target result character of other to-be-recognized characters to the trained context semantic score model to obtain a new semantic score if the result character before and after the updating is different.
[0046] According to the implementation form, when the result character before and after the updating is different, the new semantic score is obtained according to the context semantic score model, and the new semantic score can be obtained to update the corresponding result probability when the result character changes.
[0047] In order to better understand and implement, the application is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS
[0048] Figure 1 a flowchart of a character recognition method according to an embodiment of the application;
[0049] Figure 2 a schematic diagram of module connection of a character recognition system according to an embodiment of the application;
[0050] 10, initial recognition result obtaining module; 20, semantic score obtaining module; 30, target result character obtaining module. DETAILED DESCRIPTION
[0051] In order to make the objects, technical solutions and advantages of the present application clearer, the following will further describe the embodiments of the present application with reference to the accompanying drawings.
[0052] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.
[0053] When the following description refers to the accompanying drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not necessarily describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances. The singular forms "a", "an" and "the" used in the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. The word "if" used herein can be interpreted as "when" or "when" or "in response to determining".
[0054] In addition, in the description of the present application, "multiple" means two or more, unless otherwise specified. The "and / or" describes the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents that the associated objects before and after are in an "or" relationship.
[0055] Please refer to Figure 1 which is a flowchart of the character recognition method of the first embodiment of the present application, comprising:
[0056] S1: inputting a target image including a to-be-recognized sentence to a trained character recognition model to obtain an initial recognition result; the to-be-recognized sentence includes a plurality of to-be-recognized characters; the initial recognition result includes a result character with the highest probability corresponding to each to-be-recognized character.
[0057] The character recognition model is one of the core technologies in the field of computer vision and is widely used in scenarios such as CAPTCHA recognition, document digitization, and scene text extraction. The character recognition model used in the present application can be trained based on a character recognition training sample to obtain a convolutional neural network (CNN) or a convolutional recurrent neural network (CRNN). The character recognition training sample includes an image sample and an identified character of a character to be recognized in the image sample. The identified character can be identified by a human. In the above neural network, the convolutional neural network constructs a hierarchical feature extraction structure through a convolutional layer, a pooling layer, and a nonlinear activation function, and is suitable for character classification tasks in images; the convolutional recurrent neural network combines the feature extraction capability of the CNN and the sequence modeling capability of the recurrent neural network (RNN), and is suitable for non-fixed-length text sequence recognition.
[0058] S2: inputting the plurality of result characters into the trained context word sense scoring model to obtain a word sense score of each of the result characters.
[0059] The context word sense scoring model is a technical framework for dynamically evaluating the semantic importance and accuracy of words in different contexts. The core of the model is to quantify the adaptability of word semantics by analyzing the context relationship. The model is usually implemented in combination with the following technical features:
[0060] Contextual dynamic representation, using contextualized word representation methods (such as recurrent neural network language models, self-attention networks), generating dynamic semantic vectors according to the specific context of the word, breaking through the limitations of traditional static word vectors. For example, a model based on the Transformer architecture captures long-range dependencies through a self-attention mechanism.
[0061] Semantic association modeling, using sequence modeling techniques (such as N-gram models, neural probabilistic language models) to establish the probability association between words, and evaluating the reasonableness of word sense by calculating the co-occurrence probability in a specific context. In the framework of the continuous bag-of-words model, local semantic modeling is achieved through context word pair extraction.
[0062] Coherence scoring, detecting whether the generated word is logically coherent with the dialogue history or the text topic.
[0063] Ambiguity resolution capability, quantifying the semantic differentiation degree of polysemous words by comparing the differences in word vectors in different contexts.
[0064] Dynamic weight adjustment, based on the context window mechanism, giving higher semantic weight to the proximal context.
[0065] The context word sense scoring model can use a Transformer architecture, a hybrid neural network model, or a dynamic word vector model.
[0066] Transformer architecture dynamically captures long-distance semantic dependency by using multi-head self-attention mechanism, and preserves sequence order information through position encoding.
[0067] The mixed neural network model includes RNN / LSTM and CNN+Attention. RNN / LSTM: uses a recurrent structure to model temporal dependencies, and passes historical context information through hidden states, which is suitable for local semantic coherence analysis. CNN+Attention: the convolutional layer extracts local word sequence features, and combines attention mechanism to strengthen key word weights, which is commonly used for multi-granularity semantic evaluation.
[0068] The dynamic word vector model can generate context-sensitive word vectors based on the bidirectional LSTM architecture of ELMo, and solve polysemy ambiguity problems through multi-layer representation fusion.
[0069] S3: obtaining a target result character of each of the to-be-recognized characters according to the word sense score and the result probability.
[0070] Compared with the prior art, the target image including the to-be-recognized sentence is input into the trained character recognition model to obtain an initial recognition result, wherein the initial recognition result includes a result character with the highest result probability corresponding to each to-be-recognized character, and then the plurality of result characters are input into the trained context word sense scoring model to obtain a word sense score of each of the result characters, and then a target result character of each of the to-be-recognized characters is obtained according to the word sense score and the result probability, which combines character recognition and context semantic scoring to obtain the target result character of the to-be-recognized character, and integrates the character recognition dimension and the semantic analysis dimension to perform sentence character recognition, thereby improving the accuracy of character recognition.
[0071] In one possible embodiment, the step S2 of inputting the plurality of result characters into the trained context word sense scoring model to obtain a word sense score of each of the result characters comprises:
[0072] S21: obtaining a predicted sentence according to the plurality of result characters and the character order of the corresponding to-be-recognized characters.
[0073] The character order includes from left to right and from top to bottom. The priority of the character order can be determined according to the up-down position relationship of the plurality of to-be-recognized characters. For example, in the case of punctuation, the up-down positions of the plurality of to-be-recognized characters are corresponding, indicating that the plurality of to-be-recognized characters are read in the up-down order, and the priority of the character order is from top to bottom greater than from left to right. In the case of punctuation, the up-down positions of the plurality of to-be-recognized characters are not corresponding, indicating that the plurality of to-be-recognized characters are read in the left-right order, and the priority of the character order is from left to right greater than from top to bottom.
[0074] S22: inputting the predicted sentence into the trained context word sense scoring model to obtain a word sense score of each result character.
[0075] According to the character order, the predicted sentence is constructed and then input into the context word sense scoring model, so that the word sense score of the result character can be accurately obtained according to the character order.
[0076] In a feasible embodiment, the step S3 of obtaining the target result character of each to-be-recognized character according to the word sense score and the result probability comprises:
[0077] S31: updating the result probability of the corresponding result character according to the word sense score;
[0078] S32: updating the result character with the highest probability according to the updated result probability and the result probability of the candidate recognition result of the to-be-recognized character;
[0079] S33: if the result character before and after the update is the same, determining the corresponding to-be-recognized character as the target result character.
[0080] According to the word sense score and the result probability of the result character, the target result character can be accurately obtained.
[0081] In a feasible embodiment, the step S31 of updating the result probability of the corresponding result character according to the word sense score comprises:
[0082] The updated result probability is obtained according to the product of the semantic score and the result probability.
[0083] According to the embodiment, the updated result probability can be accurately obtained.
[0084] In a feasible embodiment, after the step S32 of updating the result character with the highest probability according to the updated result probability and the result probability of the candidate recognition result of the to-be-recognized character, the step comprises:
[0085] S34: If the result character before and after the update is different, input the updated result character, and the updated result character and / or target result character of other to-be-recognized characters into the trained context word sense scoring model to obtain a new word sense score.
[0086] By the embodiment, when the result character before and after the update is different, the new word sense score is obtained by continuing to obtain the new word sense score according to the context word sense scoring model, so that the new word sense score can be obtained for updating the corresponding result probability when the result character is changed.
[0087] In the present application, as long as there is a to-be-recognized character whose result character before and after the update is different, step S34 is performed until the result characters before and after the update of all to-be-recognized characters are the same. It should be noted that the result probability of all result characters is limited by the number of updates to prevent the result probability of the result character from being continuously reduced due to the increase of the number of updates. The number of updates can be set by the user, for example, set to 1, that is, the result probability of the result character can only be updated once, or can be determined according to the word sense score of the result character of the predicted sentence obtained for the first time, for example, the number of result characters whose word sense score is less than a preset score threshold in the predicted sentence obtained for the first time is the number of updates, wherein the score threshold is set by the user.
[0088] The present application also discloses a character recognition system, comprising:
[0089] An initial recognition result acquisition module 10 is configured to input a target image including a to-be-recognized sentence into a trained character recognition model to obtain an initial recognition result; the to-be-recognized sentence includes a plurality of to-be-recognized characters; and the initial recognition result includes a result character with the highest result probability corresponding to each to-be-recognized character.
[0090] A word sense score acquisition module 20 is configured to input the plurality of result characters into a trained context word sense scoring model to obtain a word sense score of each result character.
[0091] A target result character acquisition module 30 is configured to obtain a target result character of each to-be-recognized character according to the word sense score and the result probability.
[0092] Compared with the prior art, the target image including to-be-recognized characters is input into a trained character recognition model to obtain an initial recognition result, wherein the initial recognition result includes a result character with the highest result probability corresponding to each to-be-recognized character, and then the plurality of result characters are input into a trained context word sense scoring model to obtain a word sense score of each result character, and then a target result character of each to-be-recognized character is obtained according to the word sense score and the result probability, so that the target result character of the to-be-recognized character is obtained by combining character recognition and context semantic scoring, and the character recognition of the sentence is performed in the dimensions of character recognition and semantic analysis, thereby improving the accuracy of character recognition.
[0093] In one possible implementation, the word sense score obtaining module 20 includes:
[0094] A predicted sentence obtaining module is configured to obtain a predicted sentence according to the plurality of result characters and the character order of the to-be-recognized characters.
[0095] A score obtaining module is configured to input the predicted sentence into a trained context word sense scoring model to obtain a word sense score of each result character.
[0096] According to the above implementation, the predicted sentence is constructed according to the character order and then input into the context word sense scoring model, so that the word sense score of the result character can be accurately obtained according to the character order.
[0097] In one possible implementation, the target result character obtaining module 30 includes:
[0098] A result probability updating module is configured to update the result probability of the corresponding result character according to the word sense score.
[0099] A result character updating module is configured to update the result character with the highest probability according to the updated result probability and the result probability of the candidate recognition result of the to-be-recognized character.
[0100] A first processing module is configured to determine the target result character of the corresponding to-be-recognized character if the result character before and after the update is the same.
[0101] According to the above implementation, the target result character can be accurately obtained by combining the word sense score and the result probability of the result character.
[0102] In one possible implementation, the result probability updating module is configured to perform the following steps:
[0103] The updated result probability is obtained according to the product of the semantic score and the result probability.
[0104] Through the embodiment, the updated result probability can be accurately obtained.
[0105] In one possible implementation, the method further comprises:
[0106] The second processing module is configured to input the updated result character and the updated result characters of the other to-be-recognized characters and / or the target result character into the trained context word sense scoring model to obtain new word sense scores.
[0107] Through the embodiment, when the result character before and after the update is different, the new word sense scores are obtained according to the context word sense scoring model, so that the new word sense scores can be obtained for updating the corresponding result probability when the result character changes.
[0108] The device embodiments described above are merely illustrative, wherein the components illustrated as separate components can or can not be physically separate, and the components illustrated as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the present application. Those skilled in the art can understand and implement without creative labor.
[0109] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0110] The present application is described with reference to the flowcharts and / or block diagrams according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks, can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1apparatus for performing a selected function in a block or a plurality of blocks. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow Figure 1 flow or flows and / or blocks Figure 1 selected function in a block or a plurality of blocks.
[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 flow or flows and / or blocks Figure 1 selected function in a block or a plurality of blocks.
[0112] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.
[0113] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory. The memory can also include non-volatile memory, such as read only memory (ROM), EPROM, and / or flash memory. The memory can be a memory of a computer readable medium. The memory can be a memory of a computer readable storage medium.
[0114] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.
[0115] It should also be noted that the terms "comprising", "comprises" or other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.
[0116] The above embodiments are only used to illustrate the present application, but not to limit it. Instead of the above, various modifications and changes can be made to the application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall fall into the scope of the claims of the application.
Claims
1. A character recognition method, characterized in that, include: The target image, including the sentence to be recognized, is input into the trained character recognition model to obtain the initial recognition result; The statement to be recognized includes multiple characters to be recognized; The initial recognition result includes the result character with the highest probability for each character to be recognized; The multiple result characters are input into a trained contextual word sense scoring model to obtain the word sense score for each result character; Based on the semantic score and the result probability, the target result character for each of the characters to be identified is obtained.
2. The character recognition method according to claim 1, characterized in that, The step of inputting the multiple result characters into a trained context-based word sense scoring model to obtain the word sense score for each result character includes: Based on the character order of the multiple result characters and the corresponding characters to be identified, a predicted statement is obtained; The predicted statement is input into a trained contextual semantic scoring model to obtain the semantic score of each of the resulting characters.
3. The character recognition method according to claim 1, characterized in that, The step of obtaining the target result character for each of the characters to be identified based on the word meaning score and the result probability includes: Update the result probability of the corresponding result character based on the word meaning score; Based on the updated result probabilities and the result probabilities of the candidate recognition results of the character to be recognized, update the result character with the highest probability; If the resulting characters before and after the update are the same, they are determined to be the target result characters of the corresponding characters to be identified.
4. The character recognition method according to claim 3, characterized in that, The step of updating the result probability of the corresponding result character based on the word meaning score includes: The updated result probability is obtained by multiplying the semantic score and the result probability.
5. The character recognition method according to claim 3, characterized in that, The step of updating the result character with the highest probability based on the updated result probability and the result probability of the candidate recognition results of the character to be recognized includes: If the result characters before and after the update are different, the updated result characters, along with the updated result characters of other characters to be identified and / or the target result characters, are input into the trained contextual word sense scoring model to obtain a new word sense score.
6. A character recognition system, characterized in that, include: The initial recognition result acquisition module is used to input the target image, including the sentence to be recognized, into the trained character recognition model to obtain the initial recognition result; The statement to be recognized includes multiple characters to be recognized; the initial recognition result includes the result character with the highest probability corresponding to each character to be recognized. The word sense scoring module is used to input the multiple result characters into a trained context word sense scoring model to obtain the word sense score of each result character; The target result character acquisition module is used to obtain the target result character for each of the characters to be identified based on the word meaning score and the result probability.
7. The character recognition system according to claim 6, characterized in that, The word meaning scoring acquisition module includes: The prediction statement acquisition module is used to obtain the prediction statement based on the character order of the multiple result characters and the corresponding characters to be identified; The scoring acquisition module is used to input the predicted statement into the trained contextual word sense scoring model to obtain the word sense score of each of the result characters.
8. The character recognition system according to claim 6, characterized in that, The target result character acquisition module includes: The result probability update module is used to update the result probability of the corresponding result character based on the word meaning score. The result character update module is used to update the result character with the highest probability based on the updated result probability and the result probability of the candidate recognition result of the character to be recognized; The first processing module is used to determine the target result character as the corresponding character to be identified if the result characters before and after the update are the same.
9. The character recognition system according to claim 8, characterized in that, The result probability update module is used to perform the following steps: The updated result probability is obtained by multiplying the semantic score and the result probability.
10. The character recognition system according to claim 8, characterized in that, Also includes: The second processing module is used to input the updated result characters, along with the updated result characters of other characters to be identified and / or the target result characters, into the trained contextual word sense scoring model to obtain a new word sense score.