Image recognition method, device and equipment for a review system, and storage medium

Abstract

The application provides an image recognition method and device for a paper marking system, computer equipment and a storage medium. The method comprises the following steps: obtaining a candidate character set and reference answer information of a target subject; the candidate character set comprises all candidate characters; deleting part of the candidate characters in the candidate character set according to the reference answer information to obtain a target character set; adjusting an initial recognition model according to the target character set to generate a first target recognition model; and recognizing a to-be-recognized answer image through the first target recognition model to obtain a first recognition result. The embodiment of the application can reduce the recognition range of characters, thereby significantly improving the image recognition accuracy and recognition efficiency of test paper marking, and further improving the marking efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to an image recognition method and device for a paper marking system, a computer device and a storage medium. BACKGROUND

[0002] With the rapid development of education informatization and artificial intelligence technology, intelligent paper marking systems gradually replace traditional manual marking methods and are widely used in various types of examinations, significantly improving the marking efficiency.

[0003] Intelligent paper marking systems use optical character recognition (OCR) technology to automatically recognize and score handwritten answers on examination answer sheets or answer papers. However, the OCR system usually needs to recognize and compare against a complete set of Chinese characters during the answer recognition process. Due to the large number of Chinese characters, the recognition range is too large, and in cases of non-standard writing, blurred handwriting, and obvious font changes, the recognition error rate increases significantly, reducing the accuracy and reliability of the recognition results and seriously affecting the objectivity and accuracy of the marking. SUMMARY

[0004] Therefore, the present application proposes an image recognition method and device for a paper marking system, a computer device and a storage medium to solve the problem of increased recognition error rate due to an excessively large recognition range in related technologies.

[0005] The first aspect of the present application proposes an image recognition method for a paper marking system, which includes:

[0006] obtaining a candidate character set and reference answer information of a target subject; the candidate character set includes all candidate characters;

[0007] deleting part of the candidate characters in the candidate character set according to the reference answer information to obtain a target character set;

[0008] adjusting an initial recognition model according to the target character set to generate a first target recognition model;

[0009] recognizing the answer image to be recognized through the first target recognition model to obtain a first recognition result.

[0010] In the present application, the reference answer information is a first character set corresponding to the target subject; the first character set includes a plurality of first characters; the first character is a letter, a symbol or a number; and part of the candidate characters in the candidate character set is deleted through the reference answer information to obtain a target character set, including:

[0011] For any first character in the first character set, a plurality of matching degrees of the first character and a plurality of candidate characters in the candidate character set are calculated;

[0012] At least one target character corresponding to the first character is filtered from the plurality of candidate characters according to the plurality of matching degrees; the target character refers to a candidate character in the plurality of candidate characters whose matching degree is higher than a preset threshold.

[0013] A plurality of second character sets are generated according to a plurality of target characters corresponding to a plurality of first characters; each second character set includes at least one target character corresponding to a corresponding first character; and the target character set is one or more of the plurality of second character sets.

[0014] The embodiments of the present application filter at least one target character corresponding to the first character from the plurality of candidate characters through a plurality of matching degrees; the target character refers to a candidate character in the plurality of candidate characters whose matching degree is higher than a preset threshold; and a plurality of second character sets are generated according to a plurality of target characters corresponding to a plurality of first characters, which can greatly reduce the character recognition range of the first target recognition model, thereby reducing the recognition error rate, and further significantly improving the reliability and efficiency of the intelligent marking system in the actual examination environment, reducing the artificial review workload, and improving the overall marking speed.

[0015] In the embodiments of the present application, the initial recognition model is adjusted according to the target character set to generate a first target recognition model, including:

[0016] The target character set is adjusted according to the test taker information to obtain a third character set; the test taker information includes Chinese character mastery level information and test taker experience information; and the number of characters in the third character set is less than the number of characters in the target character set.

[0017] The initial recognition model is optimized according to the third character set to generate the first target recognition model.

[0018] In the embodiments of the present application, the first recognition result includes a plurality of fourth characters; after obtaining the first recognition result, the method further includes:

[0019] For any one of the plurality of fourth characters, the confidence of the fourth character is calculated;

[0020] It is judged whether the confidence is less than a confidence threshold;

[0021] If the confidence is less than the confidence threshold, the fourth character is marked as a misrecognized character;

[0022] Generate a feedback report according to the misrecognized character and context information of the misrecognized character, and send the feedback report to the review terminal.

[0023] In the embodiments of the present application, the plurality of matching degrees of the first character and the plurality of candidate characters in the candidate character set are calculated, including:

[0024] Calculate the similarity between the semantic feature vector of the first character and the semantic feature vector of the plurality of candidate characters; and take the similarity as the matching degree.

[0025] In the embodiments of the present application, before the first target recognition model is used to recognize the to-be-recognized answer image, the method further includes:

[0026] Obtain an initial answer image;

[0027] Perform denoising processing on the initial answer image to obtain a first image; the denoising processing refers to replacing each pixel value in the initial answer image with a median value in its neighborhood;

[0028] Perform handwriting correction processing on the first image to obtain a second image; the handwriting correction processing refers to removing the broken or blurred parts of the character strokes in the first image;

[0029] Perform character contour sharpening processing on the second image to obtain a third image; the character contour sharpening processing refers to enhancing the edge information and contour information of the second image.

[0030] In the embodiments of the present application, the method further includes:

[0031] For any first character, calculate the term frequency of the first character in the reference answer information and extract the semantic feature vector of the first character;

[0032] Determine the recognition weight of the first character according to the term frequency of the first character and context information; the context information is determined according to the semantic feature vector of the first character and the semantic feature vector of the adjacent first character;

[0033] Adjust the initial recognition model according to the plurality of recognition weights of the plurality of first characters in the first character set to generate a second target recognition model;

[0034] Recognize the to-be-recognized answer image by using the second target recognition model to obtain a second recognition result;

[0035] Determine a final recognition result according to the first recognition result and the second recognition result.

[0036] The embodiment of the present application determines the recognition weight of the first character through the word frequency of the first character and the context information, adjusts the initial recognition model according to the multiple recognition weights of the multiple first characters in the first character set, generates a second target recognition model, can greatly reduce the recognition range of the second target recognition model, thereby reducing the recognition error rate, and further significantly improving the reliability and efficiency of the intelligent marking system in the actual examination environment, reducing the artificial review workload, and improving the overall marking speed.

[0037] The embodiment of the second aspect of the present application provides an image recognition device for a marking system, comprising:

[0038] An information acquisition module is configured to acquire a candidate character set and reference answer information of a target subject; the candidate character set comprises all candidate characters;

[0039] A target character set determination module is configured to delete part of the candidate characters in the candidate character set according to the reference answer information to obtain a target character set;

[0040] A model adjustment module is configured to adjust an initial recognition model according to the target character set to generate a first target recognition model;

[0041] An identification module is configured to identify a to-be-identified answer image through the first target recognition model to obtain a first identification result.

[0042] The embodiment of the third aspect of the present application provides a computer device, which comprises a memory and a processor, the memory and the processor are in communication connection with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the image recognition method for a marking system according to the first aspect.

[0043] The embodiment of the fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores computer instructions, and the computer instructions are used to make a computer execute the image recognition method for a marking system according to the first aspect.

[0044] The additional aspects and advantages of the present application will be partially given in the following description, partially become obvious from the following description, or be known by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are intended to only illustrate preferred embodiments and are not considered limiting of the present application. Moreover, like reference numerals designate similar parts throughout the several views in the drawings. In the drawings:

[0046] Figure 1 A flowchart of an image recognition method for a paper marking system is shown according to an embodiment of the present application;

[0047] Figure 2 A structural diagram of an image recognition device for a paper marking system is shown according to an embodiment of the present application;

[0048] Figure 3 A structural diagram of a computer device is shown according to an embodiment of the present application;

[0049] Figure 4 A structural diagram of a computer device is shown according to an embodiment of the present application; DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present application will be described herein below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thoroughly and completely understood, and will fully convey the scope of the application to those skilled in the art.

[0051] It should be noted that, unless otherwise specified, technical terms or scientific terms used in the present application should be understood as their common meanings to those skilled in the art to which the present application pertains.

[0052] The technical scenario related to the embodiments of the present application is described below.

[0053] Most existing OCR technologies use a general character recognition dictionary for recognition. Such a general dictionary covers all commonly used Chinese characters and symbols, and thus cannot accurately match specific exam questions. In particular, in the application of the education field, exam questions are usually fixed and explicit, and the Chinese characters and characters appearing in the questions generally have strong regularity. In view of the deficiencies of the prior art, the present application proposes a technical solution capable of automatically narrowing the range of the OCR recognition dictionary and optimizing the weight of the characters according to the reference answers or the range of the student's mastery of Chinese characters, thereby improving the accuracy of handwriting recognition in an intelligent paper marking system.

[0054] According to an embodiment of the present application, an image recognition method for a paper marking system is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0055] In this embodiment, an image recognition method for a marking system is provided, which can be used in an intelligent marking system. Figure 1 It is a flowchart of an image recognition method for a marking system according to an embodiment of the present application, as Figure 1 shown. This process includes the following steps:

[0056] Step S101, obtain a candidate character set and reference answer information for the target subject.

[0057] Among them, the candidate character set includes all candidate characters; the reference answer information is the first character set corresponding to the target subject; the first character set includes multiple first characters; the first character is a character, symbol or number.

[0058] In the embodiment of the present application, the target subject includes but is not limited to Chinese, mathematics, English, science comprehensive related subjects (physics, chemistry, biology) and liberal arts comprehensive related subjects (politics, history, geography). The standard reference answers for the corresponding examination subjects can be obtained. Among them, the multiple first characters included in the first character set can be understood as the examination answers of multiple examination questions in a test paper. For example, if the answer to a fill-in-the-blank question is "A bosom friend afar brings a distant land near", then "A bosom friend afar brings a distant land near" can be used as a first character.

[0059] Step S102, delete some candidate characters in the candidate character set according to the reference answer information to obtain a target character set.

[0060] In some specific embodiments, the above step S102 includes steps S1021 - step S1023:

[0061] Step S1021, for any first character in the first character set, calculate multiple matching degrees between the first character and multiple candidate characters in the candidate character set.

[0062] In the embodiment of the present application, the matching degree between the first character and the candidate character can be calculated in the following way: first determine the semantic feature vector of the first character and the semantic feature vector of each candidate character, and then use cosine similarity or Euclidean distance to calculate the matching degree between the semantic feature vector of each candidate character and the semantic feature vector of the first character, that is:

[0063]

[0064] Among them, represents the semantic feature vector of the candidate character, represents the semantic feature vector of the first character.

[0065] Step S1022, screen out at least one target character corresponding to the first character from the multiple candidate characters according to the multiple matching degrees.

[0066] In the embodiments of the present application, the target character refers to a candidate character in the plurality of candidate characters whose matching degree is higher than a preset threshold. The preset threshold can be set according to the situation, which is not limited here.

[0067] In the embodiments of the present application, the top-K filtering mechanism can be used to select the top N characters from the candidate set according to the set threshold.

[0068] Step S1023, generating a plurality of second character sets according to the plurality of target characters corresponding to the plurality of first characters.

[0069] In the embodiments of the present application, each second character set includes at least one target character corresponding to the corresponding first character; and the target character set is one or more of the plurality of second character sets.

[0070] Step S103, adjusting the initial recognition model according to the target character set to generate a first target recognition model.

[0071] In the embodiments of the present application, the target character set can be adjusted according to the test taker information to obtain a third character set. The test taker information includes Chinese character mastery level information and test taker experience information; and the number of characters in the third character set is less than the number of characters in the target character set. Thus, the initial recognition model is optimized according to the third character set to generate the first target recognition model.

[0072] In the embodiments of the present application, the test taker information can also include the student's region and school.

[0073] The embodiments of the present application further filter the second characters in the target character set according to the test taker information, which can eliminate characters irrelevant to the test taker information, thereby improving the OCR recognition accuracy of the test paper marking.

[0074] Step S104, identifying the to-be-recognized test image through the first target recognition model to obtain a first recognition result.

[0075] Before step S104, the method further includes steps a1-a4:

[0076] Step a1, obtaining an initial test image.

[0077] Step a2, performing denoising processing on the initial test image to obtain a first image.

[0078] In the embodiments of the present application, the denoising processing refers to replacing each pixel value in the initial answer image with the median value in its neighborhood, and other denoising methods can also be adopted, which are not limited herein.

[0079] Step a3, performing handwriting correction processing on the first image to obtain a second image.

[0080] In the embodiments of the present application, the handwriting correction processing refers to removing the broken or blurred parts of the character strokes in the first image.

[0081] Step a4, performing character contour clear processing on the second image to obtain a third image.

[0082] In the embodiments of the present application, the character contour clear processing refers to enhancing the edge information and contour information of the second image.

[0083] The embodiments of the present application can effectively improve the basic recognition ability of OCR by applying multiple image enhancement algorithms, including character contour clear, handwriting correction and denoising technology.

[0084] In some specific embodiments, the first recognition result includes a plurality of fourth characters; after obtaining the first recognition result, the method further includes steps b1-b4:

[0085] Step b1, calculating the confidence of any one of the fourth characters.

[0086] In the embodiments of the present application, calculating the confidence of each character is an important step to evaluate the reliability of the recognition result. The confidence is usually a value between 0 and 1, indicating the credibility of the model to the recognition result. In the OCR system based on deep learning, the model usually outputs the prediction probability of each character. These probabilities can be directly used as the confidence of the character.

[0087] Step b2, judging whether the confidence is less than a confidence threshold.

[0088] In the embodiments of the present application, the confidence threshold can be set according to the actual situation, which is not limited herein.

[0089] Step b3, if the confidence is less than the confidence threshold, marking the fourth character as a misrecognized character;

[0090] Step b4, generating a feedback report according to the misrecognized character and the context information of the misrecognized character, and sending the feedback report to the marking end.

[0091] This application embodiment analyzes the confidence level of each identified character, automatically marking low-confidence characters as suspected misidentifications and generating feedback reports for human examiners to quickly verify, thereby improving the accuracy of marking.

[0092] In some specific embodiments, the method further includes steps 201-205:

[0093] Step 201: For any first character, calculate the word frequency of the first character in the reference answer information and extract the semantic feature vector of the first character.

[0094] In this embodiment, each first character in the first character set is a letter, symbol, or number. For example, the reference answer information for a math exam is "Solution: The first character set extracted is [solution, :, x, ... 2 ,+,y, 2 [,=,2,5]. The extracted first character set can be used as a dedicated OCR recognition dictionary for the target subject exam.

[0095] In this embodiment, word frequency refers to the frequency of occurrence, which can be understood as the frequency of the first character appearing in the reference answer information. The semantic feature vector is used to characterize the semantic information of the first character.

[0096] In some specific embodiments, the word frequency of the first character can be calculated as follows: traverse the reference answer information, and for each first character, check whether the first character is already in the counter: if so, increment the count of the first character by one. If not, add the first character to the counter and set the count to 1.

[0097] In some specific embodiments, the semantic feature vector of the first character can be obtained in the following way:

[0098] Using semantic feature extraction methods in NLP (such as word embedding or sentence embedding), each character in the reference answer information is mapped to a high-dimensional vector space, thereby obtaining the semantic feature vector corresponding to each character. These vectors can capture the semantic information of the character.

[0099] In some specific embodiments, the specific method for obtaining the semantic feature vector of the first character is not limited; obtaining the semantic feature vector of the first character is sufficient. For example, the semantic feature vector corresponding to each character can also be calculated by combining the transformer model with commonly used models in traditional NLP.

[0100] Before step S202, the method further comprises: determining context information of the first character according to the semantic feature vector of the first character and the semantic feature vector of the adjacent first character, in particular in the following manner:

[0101] calculating similarity between the semantic feature vector of the first character and the semantic feature vector of the adjacent first character; and taking the similarity as the context information of the first character.

[0102] In the embodiments of the present application, the similarity calculation can adopt cosine similarity, which measures the similarity of two vectors in direction, and the value is closer to 1, indicating that the similarity is higher. The calculated similarity is taken as the context information of the first character, which is used to adjust the recognition weight of each first character in OCR recognition. If the similarity of two first characters is higher, it means that they are more related in semantics, which can be used as context information to help the OCR system better understand the text.

[0103] The above embodiments are described by way of example:

[0104] Suppose there is "x 2 +y 2 =z 2 " in the reference answer information;

[0105] The semantic feature vector of character x is extracted, and the similarity between the semantic feature vector of character x and the semantic feature vector of adjacent character "y 2 " is calculated; when the similarity value is greater than the similarity threshold, it means that "x" is often followed by "y 2 ", so the recognition weight of character x can be increased.

[0106] Specific recognition scenario: when the OCR system recognizes a fuzzy character, if it can be "x", and it is followed by a superscript symbol "^", then the recognition weight of "x" is increased.

[0107] If the OCR system recognizes a fuzzy character that can be "y" or "z", and it is preceded by a "+" symbol, then the recognition weight of "y" or "z" is also increased.

[0108] Step 202, determining the recognition weight of the first character according to the word frequency of the first character and the context information.

[0109] The context information is determined according to the semantic feature vector of the first character and the semantic feature vector of the adjacent first character. The adjacent first character can be understood as the left character or the right character of the first character.

[0110] In some specific embodiments, the recognition weight of the first character can be calculated in the following manner:

[0111] According to the word frequency, the first weight coefficient corresponding to the word frequency, the context information, and the second weight coefficient corresponding to the context information, the recognition weight of the first character is calculated. As shown in the following formula:

[0112] a1*A+b1*B=C

[0113] Wherein, A represents the word frequency of the first character, a1 represents the first weight coefficient corresponding to the word frequency A, B represents the context information of the first character, b1 represents the second weight coefficient corresponding to the context information B, and C represents the recognition weight of the first character. The first weight coefficient and the second weight coefficient can be set according to actual conditions, which are not limited here.

[0114] Step 203, adjusting the initial recognition model according to the plurality of recognition weights of the plurality of first characters in the first character set to generate a second target recognition model.

[0115] In the embodiments of the present application, in the OCR (Optical Character Recognition) system, by adjusting the recognition weight of the character in the OCR recognition dictionary (i.e. the first character set), the recognition performance of the model can be optimized to better adapt to specific scenarios or needs.

[0116] In the embodiments of the present application, the first character set with weight is used to fine-tune the initial OCR model. During the fine-tuning process, the initial OCR model will adjust the recognition tendency of different characters according to the weight (for example: modify the hyperparameters in the configuration file, such as learning rate and batch size, to adapt to the fine-tuning process), so as to improve the recognition accuracy of the target character.

[0117] Step 204, recognizing the to-be-recognized answer image through the second target recognition model to obtain a second recognition result.

[0118] In the embodiments of the present application, the first target recognition model and the second target recognition model include but are not limited to Tesseract, EasyOCR, and PaddleOCR. The first target recognition model and the second target recognition model are any one of the plurality of recognition models, and the first target recognition model and the second target recognition model are not the same.

[0119] Step 205, determining a final recognition result according to the first recognition result and the second recognition result.

[0120] In the embodiments of the present application, the first recognition result and the second recognition result are preliminarily integrated. Voting mechanism or integrated learning algorithm can be used, such as weighted average or rule-based fusion.

[0121] The embodiment of the application can combine the recognition advantages of multiple recognition sub-models by using the mixed model integration technology, thereby significantly improving the accuracy and robustness of OCR recognition. This method is particularly suitable for processing complex scenes, multi-language environments or low-quality images.

[0122] In some specific embodiments, in an actual system, in addition to the first character set and the second character set, if a situation where the top1 / top5 character confidence of the first / second character set recognition is not high is encountered, which is relatively rare, the full character set processing is used to ensure the accuracy of recognition. Although in this case, the examination accuracy is generally not affected, because the student's answer is probably wrong, only to improve the OCR recognition rate and improve the user experience.

[0123] Corresponding to the implementation manners of the image recognition method for the examination system, the embodiment of the application further provides an image recognition device for an examination system, which is used to execute the image recognition method for the examination system described in the above embodiments. As shown in the figure, Figure 2 The image recognition device for the examination system includes:

[0124] An information acquisition module is configured to acquire a candidate character set and reference answer information of a target subject; the candidate character set includes all candidate characters;

[0125] A target character set determination module is configured to delete part of the candidate characters in the candidate character set according to the reference answer information, and obtain a target character set;

[0126] A model adjustment module is configured to adjust an initial recognition model according to the target character set, and generate a first target recognition model;

[0127] A recognition module is configured to recognize a to-be-recognized answer image by using the first target recognition model, and obtain a first recognition result.

[0128] Optionally, the reference answer information is a first character set corresponding to the target subject; the first character set includes a plurality of first characters; the first character is a letter, a symbol or a number; the target character set determination module is further configured to, for any first character in the first character set, calculate a plurality of matching degrees of the first character and a plurality of candidate characters in the candidate character set; select at least one target character corresponding to the first character from the plurality of candidate characters according to the plurality of matching degrees; the target character refers to a candidate character in the plurality of candidate characters whose matching degree is higher than a preset threshold; generate a plurality of second character sets according to a plurality of target characters corresponding to a plurality of first characters; each second character set includes at least one target character corresponding to a corresponding first character; and the target character set is one or more of the plurality of second character sets.

[0129] Optionally, the model adjustment module is further configured to adjust the target character set to obtain a third character set according to answerer information; the answerer information includes Chinese character mastery level information and answerer experience information; the number of characters in the third character set is less than the number of characters in the target character set; and the first target recognition model is generated by performing model optimization on the initial recognition model according to the third character set.

[0130] Optionally, the first recognition result includes a plurality of fourth characters; and the device further includes:

[0131] The confidence degree judgment module is configured to, after obtaining the first recognition result, calculate a confidence degree of any one of the plurality of fourth characters; judge whether the confidence degree is less than a confidence degree threshold; if the confidence degree is less than the confidence degree threshold, mark the fourth character as a misrecognized character; generate a feedback report according to the misrecognized character and context information of the misrecognized character, and send the feedback report to the paper reading end.

[0132] Optionally, the device further includes an image processing module configured to, before recognizing the to-be-recognized answer image by using the first target recognition model, acquire an initial answer image; perform denoising processing on the initial answer image to obtain a first image; the denoising processing refers to replacing each pixel value in the initial answer image with a median value in a neighborhood of the pixel value; perform handwriting correction processing on the first image to obtain a second image; the handwriting correction processing refers to removing a broken part or a blurred part of a character stroke in the first image; and perform character contour sharpening processing on the second image to obtain a third image; the character contour sharpening processing refers to enhancing edge information and contour information of the second image.

[0133] Optionally, the device further comprises: a final recognition result determination module configured to, for any first character, calculate the word frequency of the first character in the reference answer information and extract a semantic feature vector of the first character; determine a recognition weight of the first character according to the word frequency of the first character and context information, wherein the context information is determined according to the semantic feature vector of the first character and the semantic feature vectors of adjacent first characters; adjust the initial recognition model according to the recognition weights of the plurality of first characters in the first character set to generate a second target recognition model; recognize the to-be-recognized answer image through the second target recognition model to obtain a second recognition result; and determine a final recognition result according to the first recognition result and the second recognition result.

[0134] The image recognition device for the marking system provided by the above embodiments of the present application and the image recognition method for the marking system provided by the embodiments of the present application have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0135] The present application also provides a computer device for executing the above image recognition method for the marking system. Please refer to Figure 3 , which shows a schematic diagram of a computer device provided by some embodiments of the present application. As shown in Figure 3 , the computer device 3 comprises a processor 300, a memory 301, a bus 302 and a communication interface 303, wherein the processor 300, the communication interface 303 and the memory 301 are connected through the bus 302; the memory 301 stores a computer program capable of running on the processor 300, and the processor 300 runs the computer program to execute the image recognition method for the marking system provided by the above embodiments of the present application.

[0136] Among them, the memory 301 can contain a high-speed random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory. The communication between the system network element and at least one other network element is realized through at least one communication interface 303 (which can be wired or wireless), and the Internet, wide area network, local network, metropolitan area network, etc. can be used.

[0137] The bus 302 can be an ISA bus, a PCI bus, an EISA bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. The memory 301 is configured to store a program, and the processor 300 executes the program after receiving an execution instruction. The image recognition method for a marking system disclosed in the foregoing embodiments can be applied to the processor 300 or implemented by the processor 300.

[0138] The processor 300 can be an integrated circuit chip with a processing capability of signals. In the implementation process, each step of the foregoing method can be completed by an integrated logic circuit of hardware in the processor 300 or an instruction in the form of software. The processor 300 described above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), and the like; or can be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a ready programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. Each method, step, and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware coding processor for execution, or a combination of hardware and software modules in the coding processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, or other mature storage media in the art. The storage medium is located in the memory 301, and the processor 300 reads information in the memory 301 and combines the hardware to complete the steps of the foregoing method.

[0139] The computer device provided in the embodiments of the present application and the image recognition method for a marking system provided in the embodiments of the present application have the same beneficial effects as the method adopted, run, or implemented by the computer device.

[0140] The embodiments of the present application also provide a computer readable storage medium corresponding to the image recognition method for a marking system provided in the foregoing embodiments. Please refer to Figure 4 The computer readable storage medium shown in the figure is an optical disc 30, and a computer program (i.e., a program product) is stored on the optical disc 30. When the computer program is run by a processor, the image recognition method for a marking system provided in any of the foregoing embodiments is executed.

[0141] It should be noted that examples of the computer-readable storage medium can also 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 optical, magnetic storage mediums, and the like, which are not listed one by one here.

[0142] The computer-readable storage medium provided by the above embodiments of the present application has the same beneficial effects as the method adopted, run or implemented by the application program stored therein, in terms of the same inventive concept.

[0143] It should be noted that:

[0144] In the specification provided herein, a large number of specific details are described. However, it can be understood that the embodiments of the present application can be practiced without these specific details. In some examples, well-known structures and techniques are not shown in detail in order not to obscure the understanding of the present specification.

[0145] Similarly, it should be appreciated that the individual features of the application described in the above description of example embodiments of the application are sometimes grouped together in a single embodiment, a drawing figure, or a description of a related art, for the purpose of streamlining the present description. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed application requires more features than are explicitly recited in each claim. Rather, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following, in this application are hereby expressly incorporated into this detailed description, with each claim acting as a separate embodiment of the application.

[0146] Furthermore, those skilled in the art will recognize that, although some of the embodiments described herein comprise other embodiments, not all embodiments comprising certain features require the particular combination of features described. For example, method embodiments of the present application can provide product, service, and / or processor embodiments with none, some or all of the features described herein. Thus, features discussed and / or shown in one embodiment can be used, alone or in any combination, in another embodiment.

[0147] The above descriptions are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any changes or substitutions easily conceived by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An image recognition method for a marking system, characterized in that, The method includes: Obtain a candidate character set and reference answer information for the target subject; the candidate character set includes all candidate characters; the reference answer information is a first character set corresponding to the target subject; the first character set includes multiple first characters; Based on the reference answer information, some candidate characters in the candidate character set are deleted to obtain the target character set; The initial recognition model is adjusted based on the target character set to generate the first target recognition model; The first target recognition model is used to recognize the answer image to be recognized, and a first recognition result is obtained; For any first character, calculate the word frequency of the first character in the reference answer information and extract the semantic feature vector of the first character; The recognition weight of the first character is determined based on the word frequency of the first character and the context information; the context information is determined based on the semantic feature vector of the first character and the semantic feature vectors of adjacent first characters. Based on the multiple recognition weights of multiple first characters in the first character set, the initial recognition model is adjusted to generate a second target recognition model; The second target recognition model is used to recognize the answer image to be recognized, and a second recognition result is obtained. The final identification result is determined based on the first identification result and the second identification result.

2. The method according to claim 1, characterized in that, The first character is a letter, symbol, or number; By deleting some candidate characters from the candidate character set using the reference answer information, a target character set is obtained, including: For any first character in the first character set, calculate multiple matching degrees between the first character and multiple candidate characters in the candidate character set; Based on the multiple matching degrees, at least one target character corresponding to the first character is selected from the multiple candidate characters; the target character refers to the candidate character among the multiple candidate characters whose matching degree is higher than a preset threshold; Multiple sets of second characters are generated based on multiple target characters corresponding to multiple first characters; each set of second characters includes at least one target character corresponding to a corresponding first character; the set of target characters is one or more of the multiple sets of second characters.

3. The method according to claim 1 or 2, characterized in that, The initial recognition model is adjusted based on the target character set to generate a first target recognition model, including: The target character set is adjusted based on the respondent's information to obtain a third character set; the number of characters in the third character set is less than the number of characters in the target character set; The initial recognition model is optimized based on the third character set to generate the first target recognition model.

4. The method according to claim 1 or 2, characterized in that, The first recognition result includes multiple fourth characters; After obtaining the first identification result, the method further includes: For any one of the plurality of fourth characters, calculate the confidence level of the fourth character; Determine whether the confidence level is less than the confidence threshold; If the confidence level is less than the confidence threshold, the fourth character is marked as a misidentified character. A feedback report is generated based on the misidentified character and its context information, and then the feedback report is sent to the marking terminal.

5. The method according to claim 2, characterized in that, Calculating multiple matching degrees between the first character and multiple candidate characters in the candidate character set, including: Calculate the similarity between the semantic feature vector of the first character and the semantic feature vectors of multiple candidate characters; use the similarity as the matching degree.

6. The method according to claim 1 or 2, characterized in that, Before recognizing the answer image to be recognized using the first target recognition model, the method further includes: Obtain the initial answer image; The initial answer image is denoised to obtain a first image; the denoising process refers to replacing each pixel value in the initial answer image with the median value of its neighborhood. The first image is subjected to handwriting correction processing to obtain the second image; the handwriting correction processing refers to removing broken or blurred parts of the character strokes in the first image; The second image is subjected to character outline sharpening processing to obtain the third image; the character outline sharpening processing refers to enhancing the edge information and outline information of the second image.

7. An image recognition device for a marking system, characterized in that, The device includes: The information acquisition module is used to acquire a candidate character set and reference answer information for the target subject; the candidate character set includes all candidate characters; the reference answer information is a first character set corresponding to the target subject; the first character set includes multiple first characters; The target character set determination module is used to delete some candidate characters in the candidate character set according to the reference answer information to obtain the target character set; The model adjustment module is used to adjust the initial recognition model according to the target character set to generate a first target recognition model; The recognition module is used to recognize the answer image to be recognized through the first target recognition model and obtain a first recognition result; For any first character, calculate the word frequency of the first character in the reference answer information and extract the semantic feature vector of the first character; The recognition weight of the first character is determined based on the word frequency of the first character and the context information; the context information is determined based on the semantic feature vector of the first character and the semantic feature vectors of adjacent first characters. Based on the multiple recognition weights of multiple first characters in the first character set, the initial recognition model is adjusted to generate a second target recognition model; The second target recognition model is used to recognize the answer image to be recognized, and a second recognition result is obtained. The final identification result is determined based on the first identification result and the second identification result.

8. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the image recognition method for a marking system as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the image recognition method for a marking system as described in any one of claims 1 to 6.

Images