A method, system and medium for dynamic generation and adaptive optical mark recognition of an answer sheet based on group paper data driving

Abstract

The application relates to a test card dynamic generation and adaptive optical mark recognition method, system and medium based on group paper data driving, and relates to the technical field of image recognition. The test card dynamic generation and adaptive optical mark recognition method comprises the following steps: extracting test questions according to a test subject outline, generating a test subject test paper, combining a test paper structure, and constructing a corresponding test card; scanning and cleaning the test card after the test is completed to obtain a standard test card image, and combining a preset test card template to position an objective question card filling area and a subjective question answering area; recognizing the objective question card filling area, extracting a filling option, combining a correct option code sequence, and calculating an objective question score; correcting the subjective question answering area according to a correction circulation mechanism, calculating a subjective question score, combining the objective question score, and generating a student score table; and automatically generating a test card PDF according to the prepared test paper, without the need to set an OMR recognition template, so that the teacher operation cost and the template configuration error rate are reduced.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method, system and medium for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data. Background Technology

[0002] The OMR answer sheet recognition system is designed for postgraduate entrance examination education and training, as well as other standardized assessment scenarios. It can be applied to web backends, teacher-side management systems, image scanning and uploading terminals, student mini-programs, and grade reporting systems. The system integrates with existing question bank and test paper generation systems, automatically generating answer sheet PDFs from pre-assembled test papers. Students can fill in the answer sheets during offline exams. Teachers collect the answer sheets, scan them in batches using a regular scanner, and upload them to the system. The system automatically recognizes the answers to objective questions and calculates the scores, while also supporting manual grading of subjective questions by teachers. Students can view their grade reports through the mini-program and can view the stem, options, answer, and explanation for each question based on the question bank and test paper data.

[0003] In the future, it can be extended to multiple standardized examination scenarios such as civil service exams, CET-4 and CET-6, overseas study exams, professional qualification exams, school exams, and corporate internal assessments, to achieve an integrated marking service that links offline exams, online recognition, automatic scoring, manual grading, score analysis, and question explanation.

[0004] Existing patents disclose an intelligent, traceable marking system based on image recognition, comprising: a customized answer sheet module for generating answer sheet information; an image processing module for scanning the QR code to obtain the exam paper number, and obtaining the corresponding coordinate information description table and exam paper question information based on the exam paper number, comparing the actual coordinate positions of the marker points in the scanned image with the coordinate positions of the marker points recorded in the positioning information table; and a data processing module for obtaining the recognition result table of all participating students, thereby obtaining the answer information of the questions, calculating the scores for objective and subjective questions respectively, obtaining the score and ranking of each student's single exam paper, and storing it in the calculation result table. This invention allows for quick understanding of students' knowledge mastery through reports, facilitating targeted instruction, simplifying the marking process, and significantly reducing the learning cost for teachers.

[0005] The existing technical solutions mentioned above have the following drawbacks: 1. Traditional answer sheet recognition software usually requires teachers to format the answer sheets separately after the test paper is compiled and manually configure the OMR recognition template. The operation is complicated and it is easy to have problems such as inconsistencies between the question number, the number of options, and the layout position and the actual test paper, which increases the cost of use for teachers and the probability of errors. 2. Traditional marking systems rely on dedicated scanners, dedicated marking machines, or specified answer sheet paper, resulting in high equipment procurement and maintenance costs, making them unsuitable for ordinary educational institutions, training institutions, or flexible examination scenarios. For users who already own ordinary scanners, traditional systems lack compatibility. 3. Traditional OMR systems often require the creation of fixed templates for each type of answer sheet, with users manually selecting recognition areas, setting question number ranges, and option rules. Once the test paper structure, number of questions, or layout changes, the template needs to be reconfigured, making it difficult to adapt to teaching scenarios involving frequent test paper creation. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this application is to provide a method, system, and medium for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data. By integrating a question bank system, test paper generation system, automatic generation of answer sheet PDFs, uploading of images from ordinary scanners, OMR recognition, automatic scoring of objective questions, manual grading of subjective questions, and functions for students to query scores and view question analysis via a mini-program, this application solves the problems of complex answer sheet production, cumbersome template configuration, strong equipment dependence, and limited result feedback in traditional answer sheet recognition software. This enables an efficient, low-cost, traceable, and reviewable intelligent marking process in offline examination scenarios.

[0007] This was achieved using the following technical solutions: In a first aspect, this application provides a method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data, including: Based on the examination subject syllabus, test questions are extracted to generate examination papers for the examination subjects, and corresponding answer sheets are constructed in combination with the structure of the test papers. Scan and clean the answer sheets after the exam to obtain a standard answer sheet image, and use the preset answer sheet template to locate the areas for marking objective questions and answering subjective questions; Identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence; The subjective question answering area is graded according to the grading and transfer mechanism, the subjective question scores are calculated, and the objective question scores are combined to generate a student grade sheet.

[0008] By adopting the above technical solution, the answer sheet area is located based on scanning and template matching algorithms. Optical markers are used to identify the filling options for objective questions and verify the standard answer codes for scoring. At the same time, the subjective question area is scored in conjunction with the grading process. Finally, the scores are merged to generate student grade sheets, which realizes the automation, standardization and efficiency of the examination grading process, and significantly improves grading efficiency and score accuracy.

[0009] This application further includes: extracting test questions based on the examination subject syllabus, generating an examination paper for the examination subject, and constructing a corresponding answer sheet based on the examination paper structure, including: The exam syllabus was analyzed, and the weight of knowledge points in each chapter and the distribution of question types were extracted. The test questions are extracted based on the weight of the knowledge points in each chapter, resulting in a candidate test question set; Based on the distribution of question types and their historical usage frequency, stratified sampling was performed on the candidate question set to obtain the required questions for each question type. Based on the exam paper structure, the required questions for each question type are arranged and rendered to generate the exam paper for the exam subjects; Match the answer sheet template type according to the test paper structure to obtain a unique answer sheet template; The answer sheet template is divided according to the question type order of the test paper, and corresponding question number identifiers, objective question filling dot matrix and subjective question answer space are generated to obtain the answer sheet.

[0010] By adopting the above technical solution, the weight of knowledge points and the distribution of question types are extracted by analyzing the syllabus. The test questions are extracted and arranged and rendered to generate the test paper using a stratified sampling strategy. At the same time, the answer sheet template is automatically matched according to the test paper structure, and the question numbers, answer sheet grids and subjective question answering spaces are divided to achieve the synchronous generation of test paper and answer sheet, which significantly improves the efficiency and standardization of test paper assembly.

[0011] This application further includes: scanning and cleaning the answer sheets after the exam to obtain a standard answer sheet image, and using a pre-set answer sheet template to locate the areas for marking objective questions and answering subjective questions, including: The answer sheets were scanned one by one after the exam according to the scanning resolution to generate the original answer sheet images; The original answer sheet image is subjected to orientation recognition, the position of the orientation marker is extracted, and the image orientation is determined. If the image orientation is inconsistent with the preset standard image orientation, the original answer sheet image is rotated and corrected to obtain the corrected answer sheet image; The reference point is detected on the corrected answer sheet image to obtain the actual coordinates of the reference point, and the image tilt angle is calculated by combining the theoretical coordinates of the reference point; A rotation matrix is ​​constructed based on the image tilt angle, and an affine transformation is performed on the corrected answer sheet image to obtain a horizontal answer sheet image; Edge detection and segmentation are performed on the horizontal answer sheet image to determine the effective content image of the answer sheet; The valid content image of the answer sheet is binarized to obtain a grayscale image of the answer sheet content; The grayscale image of the answer sheet content is denoised and smoothed to obtain a standard answer sheet image, and the actual positioning markers are extracted. The preset answer sheet template is tested, and the template positioning marks, coordinates of the objective question filling area, and coordinates of the subjective question answering area are extracted. Match the template positioning marks with the actual positioning marks, and calculate the image scaling ratio and coordinate offset; Based on the coordinates of the objective question filling area, combined with the image scaling and coordinate offset, locate the objective question filling area; The answer area for subjective questions is located by combining the coordinates of the answer area with the image scaling and coordinate offset.

[0012] By adopting the above technical solution, after scanning to generate the original answer sheet image, standardization processes such as orientation recognition, rotation correction, reference point positioning, affine transformation, edge cropping, binarization, and noise reduction are performed in sequence to obtain a standard answer sheet image and accurately locate the objective question marking area and the subjective question answering area. This achieves automated geometric correction and noise suppression of the answer sheet image, significantly improving the accuracy and robustness of subsequent recognition and scoring.

[0013] This application is further configured to: identify the marked area for objective questions, extract the marked options, combine the correct option coding sequence, and calculate the objective question score, including: Map the coordinates of the areas marked on the answer sheet for objective questions to obtain the areas for filling in the marks; The filling point area is divided according to the question number to obtain the question filling points; The number of black pixels in the question's fill-in points is counted, and the black pixel density is calculated. If the black pixel density is greater than the preset filling judgment threshold, it is determined to be a valid filling option, and the number of filling options is recorded; otherwise, it is determined to be an invalid filling option. If the number of options filled in is 1, the option status is marked as normal single selection; If the number of filled-in options is greater than 1, the option status is marked as multiple selections pending verification; If the number of options filled in is 0, it is determined that the question has been missed in the option status marker. The valid fill options are converted according to the preset option arrangement order to form fill option values; The filling option values ​​and option status markers are combined according to the question number identifier to generate a filling result sequence; The answer sequence is compared with the answer sequence based on the correct answer code sequence for the objective question, and the score for the objective question is calculated by combining the answer sequence with the question score.

[0014] By adopting the above technical solution, the marking area is extracted through coordinate mapping and question number segmentation, the black pixel density is calculated to determine the valid marking, and the single-choice, multiple-choice, pending verification, or missing marking status is automatically distinguished according to the number of options (1 / multiple / 0). The marked values ​​are converted into codes and compared with the standard answers one by one. The objective question scores are accumulated by combining the question scores, realizing the fully automated, high-precision identification and abnormal status marking of objective question scoring, which significantly improves the efficiency, accuracy and flexibility of marking.

[0015] This application is further configured to: grade the subjective question answer area according to the grading and transfer mechanism, calculate the subjective question scores, combine them with the objective question scores, and generate a student grade sheet, including: The subjective question answering area is divided according to the grading and circulation mechanism, generating several task packages to be graded; the grading and circulation mechanism includes a question-based circulation mode, a student-based circulation mode, a class-based circulation mode, and a task-based circulation mode; The task package to be graded is parsed based on the grading interface to extract the answer area image and question content; The answer area image is extracted and matched based on the subject character library to obtain the subjective question answer text field; Based on the content of the question, the standard answer is matched, and the questions are divided and assigned values ​​according to keywords to obtain several score items; Approximate matching is performed on the scoring items, and then matched with the subjective question response fields to obtain the initial subjective question scores; The initial subjective question scores are verified. If the verification is successful, the initial subjective question scores are determined to be the subjective question scores. If not, the scoring items will be revised, and the subjective question responses will be re-scored until the verification is successful; The scores of subjective questions from different marking and circulation mechanisms are combined according to the score difference control rules to obtain the final subjective question score. The final scores for subjective and objective questions are summarized based on the students' exam numbers, and then arranged according to the question number to generate a student grade sheet.

[0016] By adopting the above technical solution, subjective questions are divided into task packages based on the grading workflow mechanism (by question / student / class / task). The answer text is extracted by OCR and character library matching. The initial score is obtained by approximate comparison by combining keyword division and score assignment. After multiple rounds of verification and correction to ensure the accuracy of the score, the scores under different workflow modes are finally synthesized according to the score difference control rules. After summarizing the objective question scores, a student grade sheet is generated, which improves the automation, flexibility and reliability of subjective question grading.

[0017] This application further specifies: extracting and matching the answer area image based on the subject character library to obtain the subjective question answer text field, including: The image of the answer area is filtered and denoised to obtain a noise-free image of the answer area. The image of the noiseless answer area is horizontally projected and segmented to extract the characters of the answer area lines; The characters in the answer area are vertically projected and segmented to obtain several character image blocks; The character image blocks are normalized in size and their features are extracted to obtain character feature vectors; Based on the subject character library, character feature vectors are matched to calculate character similarity; If the character similarity is greater than the preset character matching threshold, the current character is determined to be a candidate character; If the number of candidate characters is one, then it is determined to be the target character; If there are multiple candidate characters, they are combined into a string sequence, and the context constraint is checked using a language model to filter the target characters. The target characters are combined to obtain the subjective question response field.

[0018] By adopting the above technical solution, character image blocks are extracted through filtering and denoising and projection segmentation. After size normalization and feature extraction, they are matched with the subject character library for similarity. Target characters are then filtered using context verification and finally combined into the subjective question answer field. This achieves high-precision automatic recognition and extraction of handwritten text, significantly improving the efficiency and accuracy of subjective question grading.

[0019] Secondly, this application also provides a dynamic generation and adaptive optical marker recognition system for answer sheets based on test paper data, which adopts the following technical solution: A dynamic answer sheet generation and adaptive optical marker recognition system based on test paper data, comprising the following methods for implementing dynamic answer sheet generation and adaptive optical marker recognition: The question paper generation data terminal is used to extract questions based on the examination subject syllabus, generate examination papers for the examination subjects, and construct corresponding answer sheets based on the examination paper structure. The answer sheet generation terminal is used to scan and clean the answer sheets after the exam to obtain a standard answer sheet image, and, in conjunction with a preset answer sheet template, locate the areas for marking objective questions and answering subjective questions. The image recognition processing unit is used to identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence. The scoring report is used to grade the subjective question answer area according to the grading process, calculate the subjective question scores, and combine them with the objective question scores to generate a student grade sheet.

[0020] By adopting the above technical solutions, the test questions are extracted by analyzing the syllabus and the test papers and corresponding answer sheets are generated simultaneously. Image preprocessing and template matching algorithms are used to accurately locate the marking and answering areas. Combined with OCR recognition and keyword matching, the objective questions are automatically scored and the subjective questions are intelligently graded. Multiple transfer modes and score difference verification are used to ensure the consistency of scoring. Finally, the student grade sheets are generated. This significantly improves the automation, processing efficiency and scoring reliability of the entire marking process.

[0021] Thirdly, this application also provides an electronic device, comprising: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement any of the methods in the above scheme.

[0022] Fourthly, this application also provides a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the answer sheet dynamic generation and adaptive optical marker recognition method based on test paper data driven as described above.

[0023] In summary, the beneficial technical effects of this application are as follows: The system automatically generates answer sheets in PDF format from pre-assembled test papers, eliminating the need for teachers to manually format answer sheets or set OMR recognition templates. This significantly reduces teacher workload and template configuration error rates. Furthermore, it integrates the question bank, test paper generation, answer sheets, grading, score reports, and question explanations, ensuring consistency in answer sheet layout, question numbers, question types, number of options, standard answers, and explanation data. This avoids the problem of separate test papers and answer sheets in traditional systems. It can use most ordinary scanners on the market to scan answer sheets in batches, eliminating the need to purchase dedicated marking machines or scanning equipment, thus reducing hardware costs and facilitating rapid deployment by training institutions and schools; it automatically collects students' answer sheet marking results through image preprocessing and OMR recognition, and automatically scores objective questions in combination with standard answers, improving marking efficiency and reducing manual data entry and scoring workload; It also supports the display of subjective question areas and manual grading, making it suitable for the complete examination scenario that includes both objective and subjective questions, rather than being limited to grading purely objective questions; through positioning point recognition, image correction, smear threshold judgment, anomaly marking and manual review mechanisms, it improves the recognition stability of scanned images under different devices, different brightness, and different tilt angles; By unifying the management of exam results, question data, student answer data, incorrect question data, and analysis data, a data foundation can be provided for subsequent teaching analysis, class learning statistics, analysis of knowledge weaknesses, and personalized tutoring. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the dynamic generation of answer sheets and the adaptive optical marker recognition method in this application; Figure 2 This is a flowchart of the context-aware speech generation process in this application; Figure 3 This is a schematic diagram of the portrait completion-driven dialogue path in this application; Figure 4 This is a schematic diagram of the answer sheet dynamic generation and adaptive optical marker recognition system in this application. Detailed Implementation

[0025] The present application will be further described in detail below with reference to the accompanying drawings.

[0026] Reference Figure 1 This application discloses a method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data, comprising: S1: Extract test questions based on the examination subject syllabus, generate examination papers for the examination subjects, and construct corresponding answer sheets based on the structure of the examination papers; S2: Scan and clean the answer sheet after the exam to obtain a standard answer sheet image, and locate the objective question marking area and the subjective question answering area using the preset answer sheet template; S3: Identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence; S4: Based on the marking and circulation mechanism, mark the subjective question answer area, calculate the subjective question score, and combine it with the objective question score to generate a student grade sheet.

[0027] In this embodiment, teachers select questions in the question bank system or generate exam papers through the test paper generation system. The system records data such as the test paper questions, question types, question numbers, number of options, scores, standard answers, and explanations.

[0028] The system automatically generates an answer sheet PDF based on the test paper compilation results. The answer sheet PDF contains areas for marking objective questions, answering subjective questions, positioning marks, exam identification, and student identification, all corresponding to the test paper structure.

[0029] Teachers download and print the answer sheets (PDF) and organize students for an offline exam. Students fill in their answers to objective questions on the answer sheets and write their answers to subjective questions in the designated areas. After the exam, teachers collect the answer sheets and scan them in batches using a regular scanner, without the need for specialized marking equipment. Teachers then upload the scanned answer sheet images or PDF files in batches to the Yantu OMR marking system.

[0030] The system preprocesses the uploaded images, including orientation recognition, tilt correction, cropping, binarization, noise reduction, and location point recognition. Based on the answer sheet layout parameters and location points, the system matches the scanned image with the coordinates of a standard answer sheet to locate the areas for marking objective questions and answering subjective questions. The system performs OMR recognition on the objective question marking areas to determine the selected answer for each question and automatically calculates the objective question score based on the question bank and the standard answers in the test paper generation system.

[0031] The system presents subjective questions to teachers, who then manually grade and enter scores online. The system supports grading by question, student, class, or task. It aggregates scores for both objective and subjective questions, generating a student's total score, individual question scores, incorrect answer records, and a performance report. Students can view their report via a mini-program, clicking on question numbers to view the question stem, options, correct answer, and explanation.

[0032] Preferably, step S1 includes: The exam syllabus was analyzed, and the weight of knowledge points in each chapter and the distribution of question types were extracted. The test questions are extracted based on the weight of the knowledge points in each chapter, resulting in a candidate test question set; Based on the distribution of question types and their historical usage frequency, stratified sampling was performed on the candidate question set to obtain the required questions for each question type. Based on the exam paper structure, the required questions for each question type are arranged and rendered to generate the exam paper for the exam subjects; Match the answer sheet template type according to the test paper structure to obtain a unique answer sheet template; The answer sheet template is divided according to the question type order of the test paper, and corresponding question number identifiers, objective question filling dot matrix and subjective question answer space are generated to obtain the answer sheet.

[0033] In this embodiment, the corresponding exam syllabus is loaded according to the current exam subject. The syllabus includes the exam scope, the weight of knowledge points in each chapter, the distribution of question types (multiple choice, true / false, fill-in-the-blank, short answer, essay, etc.), the number of questions for each type, and the score allocation. Based on the knowledge point scope in the subject syllabus, all questions that match the current exam subject are selected from the question bank and loaded into memory according to question type. Based on the question type distribution requirements in the syllabus, the number of questions to be extracted for each question type and their corresponding score weights are determined. According to the extraction rules for each question type, questions for each question type are extracted sequentially from the corresponding question bank based on principles such as balanced knowledge point coverage, reasonable difficulty distribution, and historical usage frequency, using random sampling or stratified sampling algorithms.

[0034] All extracted questions are arranged according to the pre-set test paper structure, with each question type arranged sequentially. Based on the weight of each question's score, a complete test paper is generated, including the question text, options (multiple-choice questions), answer key, and score allocation. The assembled test paper is then formatted and rendered according to a standard exam paper template, generating an exam paper file (PDF format) for the exam subject, labeled with the exam subject name, exam duration, total score, and instructions.

[0035] The structural information is extracted from the generated exam paper, including the number of questions for each question type, the score for each question type, the number of options for multiple-choice questions, and the required answer area for subjective questions. Based on the exam paper structure, a pre-defined answer sheet template type is matched. Answer sheet types include general standard cards, cards with answer areas for subjective questions, and mixed question type cards. Following the question type order, the answer sheet is divided into multiple-choice, true / false, fill-in-the-blank, and subjective question answer areas. Each area generates a corresponding question number identifier, and multiple-choice and true / false questions generate a corresponding number of fill-in dots.

[0036] For objective question types such as multiple choice and true / false questions, the system generates standardized dot matrix markings in the corresponding locations on the answer sheet, with each dot corresponding to an option and labeled with the question number. For subjective question types such as short answer and essay questions, the system draws blank answer areas in the corresponding locations on the answer sheet, labeled with the question number and reserved space for answering. A candidate information entry area is generated at the top of the answer sheet or in a fixed location, including fields such as candidate name, examination number, examination room number, and seat number. All areas are formatted and rendered according to the standard answer sheet template, generating an answer sheet file (PDF format) containing positioning markers and a barcode area for subsequent automatic grading and scanning.

[0037] Preferably, step S2 includes: The answer sheets were scanned one by one after the exam according to the scanning resolution to generate the original answer sheet images; The original answer sheet image is subjected to orientation recognition, the position of the orientation marker is extracted, and the image orientation is determined. If the image orientation is inconsistent with the preset standard image orientation, the original answer sheet image is rotated and corrected to obtain the corrected answer sheet image; The reference point is detected on the corrected answer sheet image to obtain the actual coordinates of the reference point, and the image tilt angle is calculated by combining the theoretical coordinates of the reference point; A rotation matrix is ​​constructed based on the image tilt angle, and an affine transformation is performed on the corrected answer sheet image to obtain a horizontal answer sheet image; Edge detection and segmentation are performed on the horizontal answer sheet image to determine the effective content image of the answer sheet; The valid content image of the answer sheet is binarized to obtain a grayscale image of the answer sheet content; The grayscale image of the answer sheet content is denoised and smoothed to obtain a standard answer sheet image, and the actual positioning markers are extracted. The preset answer sheet template is tested, and the template positioning marks, coordinates of the objective question filling area, and coordinates of the subjective question answering area are extracted. Match the template positioning marks with the actual positioning marks, and calculate the image scaling ratio and coordinate offset; Based on the coordinates of the objective question filling area, combined with the image scaling and coordinate offset, locate the objective question filling area; The answer area for subjective questions is located by combining the coordinates of the answer area with the image scaling and coordinate offset.

[0038] In this embodiment, the answer sheet scanning device is started, and the scanning resolution and color mode are set to ensure that the image clarity meets the requirements for subsequent recognition. After the exam, the answer sheets are sequentially placed into the paper feed slot of the scanning device. The device scans each sheet one by one and generates original answer sheet image files, which are then temporarily stored in the scanning buffer in sequence. Each scanned image undergoes a preliminary quality check to detect problems such as image blurring, overexposure, and severe skew. For images that do not meet the quality standards, the system marks them and prompts for rescanning. The color or grayscale scanned image is converted to a standard grayscale image, and the RGB three channels are converted to single-channel grayscale values. An adaptive thresholding algorithm is used to convert the grayscale image into a binary image, creating a clear contrast between the filled area and the background, facilitating subsequent area location and recognition.

[0039] The system detects positioning markers or baselines in the binary image, calculates the image tilt angle, and performs rotation correction to restore the answer sheet content to a horizontally and vertically orthogonal state. After correction, noise is removed from the image, irrelevant edge regions are eliminated, and the valid answer sheet area is retained to generate a standard answer sheet image.

[0040] Based on the exam subject and paper structure, the corresponding preset answer sheet template is loaded from the template library. The template includes the coordinates of the objective question filling area, the coordinates of the subjective question answer area, the location marker positions, and template size parameters. The system searches for location markers in a standard answer sheet image, matches the identified actual location markers with the location markers in the template, and calculates the scaling ratio and coordinate offset between the image and the template.

[0041] Based on the coordinate parameters of the objective question filling area in the template, combined with the image scaling and coordinate offset, the template coordinates are mapped to the actual coordinates of the standard answer sheet image, locating the positions of all objective question filling points one by one. Based on the coordinate parameters of the subjective question answer area in the template, the actual positions of all subjective question answer areas are located using the above mapping method.

[0042] Check if the number of objective question fill-in points and subjective question answer areas located matches the template, and whether the location of each area is within a reasonable range. If the verification passes, the area location is complete; if the verification fails, the system marks the current answer sheet as having an abnormal location and prompts for manual review.

[0043] Preferably, step S3 includes: Map the coordinates of the areas marked on the answer sheet for objective questions to obtain the areas for filling in the marks; The filling point area is divided according to the question number to obtain the question filling points; The number of black pixels in the question's fill-in points is counted, and the black pixel density is calculated. If the black pixel density is greater than the preset filling judgment threshold, it is determined to be a valid filling option, and the number of filling options is recorded; otherwise, it is determined to be an invalid filling option. If the number of options filled in is 1, the option status is marked as normal single selection; If the number of filled-in options is greater than 1, the option status is marked as multiple selections pending verification; If the number of options filled in is 0, it is determined that the question has been missed in the option status marker. The valid fill options are converted according to the preset option arrangement order to form fill option values; The filling option values ​​and option status markers are combined according to the question number identifier to generate a filling result sequence; The answer sequence is compared with the answer sequence based on the correct answer code sequence for the objective question, and the score for the objective question is calculated by combining the answer sequence with the question score.

[0044] In this embodiment, using the positioning point as a reference and combining the theoretical coordinate positions of each objective question's fill-in point in the preset answer sheet template, the template coordinates are mapped to the actual pixel coordinates of the current answer sheet image, and the coordinate range of the image area where each fill-in point is located is determined one by one. Based on the mapped position information, the fill-in area corresponding to each question number is cropped from the answer sheet image, and each area contains the fill-in points of all options for that question.

[0045] The system counts the number of black pixels within each option's fill-in area and calculates the black pixel density, which is the proportion of black pixels to the total number of pixels in that fill-in area. This black pixel density is compared to a preset fill-in threshold. If the density is higher than the threshold, the option is considered filled; otherwise, it is considered unfilled. The system counts the number of filled options for each question number. If the number is 1, the question is a normal single-choice question; if the number is greater than 1, the question is marked as an abnormal multiple-choice question; if the number is 0, it is marked as an unfilled option. Based on a preset option order (A, B, C, D, etc.), the options judged as filled under each question number are converted into corresponding option codes, forming the filled option value for that question. All filled option values ​​and abnormal status markers for all question numbers are combined in ascending order of question number to form a fill-in result sequence for subsequent comparison with the standard answer.

[0046] Based on the current exam subject and paper version, the corresponding objective question standard answer sequence, including the correct option code for each question number, is loaded from the question bank management system. The filled-in result sequence is compared with the standard answer sequence for each question: if the filled-in result matches the standard answer, the question is considered correct; if they differ or there are multiple selections / omissions, the question is considered incorrect. The number of correct questions is counted, and this number is multiplied by the score for each question type to obtain the basic objective question score. According to the preset scoring rules, points are deducted for multiple selections or omissions. The deducted score and the basic score together constitute the candidate's final objective question score.

[0047] Reference Figure 2 Preferably, step S4 includes: S41: Divide the answer area for subjective questions according to the grading and circulation mechanism, and generate several task packages to be graded; the grading and circulation mechanism includes the circulation mode by question, the circulation mode by student, the circulation mode by class, and the circulation mode by task; S42: Parse the task package to be graded based on the grading interface, and extract the answer area image and question content; S43: Extract and match the answer area image according to the subject character library to obtain the subjective question answer text field; S44: Match the standard answer content with the question content, divide and assign values ​​according to keywords, and obtain several score items; S45: Perform approximate matching on the score items and match them with the subjective question response fields to obtain the initial subjective question scores; S46: Verify the initial subjective question scores. If the verification is successful, the initial subjective question scores are determined to be the subjective question scores. If not, the scoring items will be revised, and the subjective question responses will be re-scored until the verification is successful; S47: Subjective question scores from different grading and circulation mechanisms are combined according to the score difference control rules to obtain the final subjective question score; S48: Summarize the final subjective and objective scores based on the student's exam number, arrange them according to the question number, and generate a student grade sheet.

[0048] In this embodiment, the following modes are used: Question-based distribution mode: The same subjective question is assigned to a fixed group of examiners, who then grade all students' questions on that question; Student-based distribution mode: All subjective questions for the same student are packaged and assigned to the same examiner, who then grades all of that student's subjective questions; Class-based distribution mode: All subjective questions for the same class are assigned to one examiner, who then grades all subjective questions for all students in that class; Task-based distribution mode: A certain number of subjective question tasks are evenly distributed among multiple examiners, with each teacher receiving several independent task packages. The questions within each task package are not limited by question type or class.

[0049] Based on the selected workflow mode, the subjective question answering area to be graded is divided into several task packages according to the corresponding allocation rules, and the task packages are distributed to the grading queues of the corresponding graders. After logging into the system, graders can only see the task packages to be graded assigned to them.

[0050] After entering the grading interface, the system displays the answer area image and question information for each subjective question in the order of the task packages. Students score the currently displayed subjective question according to the grading criteria and the content of the answer, enter the score, and submit. The system supports multiple scoring systems (100-point, 10-point, 5-point, etc.). According to preset rules, the same question is assigned to multiple graders for independent grading, and each grader's score and corresponding grader ID are recorded. Scores are then combined after all grading is completed. Each grading record is linked and stored with the current student ID, question number, grading timestamp, and grader ID to ensure traceability throughout the grading process.

[0051] If the current question uses a single-marking mode, the system will directly use the teacher's submitted score as the subjective score for that question. If the current question uses a double-marking or multi-marking mode, the system will synthesize the scores according to preset score difference control rules after two or more teachers have completed their independent marking. If the score difference is within the allowable range, the average score will be taken as the final score; if the score difference exceeds the threshold, an arbitration mechanism will be triggered, and the marking team leader will re-mark the question and give an arbitration score.

[0052] The scores for all objective and subjective questions for the same candidate are summarized separately according to question type, resulting in sub-scores for each question type and for subjective questions. The objective and subjective scores are then added together to obtain the candidate's total score for the exam subject, generating a detailed score report for the candidate in the current subject, including objective question scores, subjective question scores, and the total score.

[0053] After all candidates' scores are compiled, a structured score sheet is generated. The score sheet contains a list of candidates sorted by student ID or examination ID. Each candidate's score details include objective question scores, subjective question scores, total score, and detailed scores for each question type. It also supports automatic descending sorting by total score.

[0054] Reference Figure 3 Preferably, step S43 includes: The image of the answer area is filtered and denoised to obtain a noise-free image of the answer area. The image of the noiseless answer area is horizontally projected and segmented to extract the characters of the answer area lines; The characters in the answer area are vertically projected and segmented to obtain several character image blocks; The character image blocks are normalized in size and their features are extracted to obtain character feature vectors; Based on the subject character library, character feature vectors are matched to calculate character similarity; If the character similarity is greater than the preset character matching threshold, the current character is determined to be a candidate character; If the number of candidate characters is one, then it is determined to be the target character; If there are multiple candidate characters, they are combined into a string sequence, and the context constraint is checked using a language model to filter the target characters. The target characters are combined to obtain the subjective question response field.

[0055] In this embodiment, based on the already located coordinates of the subjective question answering area, the answering area image block corresponding to each subjective question is cropped from the standard answer sheet image; the answering area image is preprocessed, including converting the color or grayscale image into a binary image, removing noise and interference pixels from the image, and using a horizontal projection method or a deep learning-based target detection method to detect and segment each line of text.

[0056] Each line of text is segmented using vertical projection analysis. The number of black pixels in each column is counted, and the segmentation points between characters are determined based on the trough positions of the projection values. The line image is then further segmented into independent character image blocks. For cases where characters are connected, the system employs connected component analysis or deep learning-based character detection methods for segmentation.

[0057] Each segmented character image block is uniformly scaled to a fixed size to align all character images in terms of size, facilitating subsequent standardized comparison of feature vectors. Feature vectors representing the shape features of the characters are extracted from the normalized character image blocks. Vectors representing the character shape, such as histogram of oriented gradients or stroke direction features, are used, and each character image block is converted into a fixed-length feature vector.

[0058] Load the character library corresponding to the current exam subject from the configuration parameters. The character library contains sample data of printed and handwritten characters, stored as key-value pairs of character encoding and corresponding feature vectors, and includes character usage frequency weights for priority determination when processing fuzzy matching.

[0059] The extracted character feature vectors are compared with the feature vectors of each character in the character library one by one to calculate the similarity score between each character image block and each candidate character in the character library. Candidate characters are sorted from highest to lowest based on their similarity scores, and the character with the highest score exceeding a preset matching threshold is selected as the recognition result for the current character image block. If multiple candidate characters have similar scores and all exceed the threshold, the system temporarily stores multiple candidate results for further disambiguation through subsequent contextual verification.

[0060] The continuously identified character sequences are combined into strings, and a language model is used to perform contextual constraint verification on the candidate string sequences. For character positions with multiple candidates, the system selects the character that best fits the contextual meaning based on the contextual context and the semantic rationality of adjacent character matching. The system analyzes the coherence of the identified strings in terms of grammatical structure and lexical collocation. Sequences that do not conform to common collocations or have obvious semantic anomalies are marked as low-confidence areas for manual review. The character sequences that pass the contextual verification are combined into complete strings, which serve as the response text field for the current subjective question answering area. At the same time, character positions or characters with an identification confidence level below the threshold are marked as "pending review" for subsequent manual correction.

[0061] Reference Figure 4 A dynamic answer sheet generation and adaptive optical marker recognition system based on test paper data is applied to the dynamic answer sheet generation and adaptive optical marker recognition method, including: The question paper generation data terminal is used to extract questions based on the examination subject syllabus, generate examination papers for the examination subjects, and construct corresponding answer sheets based on the examination paper structure, including: In this embodiment, the question bank management module is used to store and manage question data, including information such as question stem, options, standard answer, explanation, question type, score, knowledge point, difficulty, subject, and exam type. This module provides basic data for test paper generation, answer sheet generation, scoring, and grade reporting.

[0062] The test paper generation and management module is used by teachers or the system to generate test papers according to examination requirements, determining the question order, question type structure, score settings, and the division between objective and subjective questions. The test paper generation results serve as the basis for answer sheet generation and subsequent marking and identification.

[0063] The answer sheet generation terminal is used to scan and clean the answer sheets after the exam to obtain a standard answer sheet image, and, in conjunction with a preset answer sheet template, locate the areas for marking objective questions and answering subjective questions. In this embodiment, the automatic answer sheet PDF generation module automatically generates answer sheet PDFs based on the test paper compilation results, eliminating the need for teachers to manually format the answer sheets. The system automatically generates an answer sheet layout adapted to the current test paper based on information such as the number of questions, question types, number of options, student information area, location markers, and subjective question answering area. The answer sheet printing and offline answering module allows teachers to download and print the system-generated PDF answer sheet, and students complete the answer sheet marking and writing their answers in the offline exam. Objective questions are answered by filling in the answer choices, while subjective questions can be answered by writing their answers in the designated areas.

[0064] The image recognition processing unit is used to identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence. In this embodiment, the image acquisition module scans the answer sheets after the exam by collecting them and performing batch scanning using a regular scanner. This module is compatible with most commercially available scanners and does not require a dedicated marking machine or scanning equipment. The scanned results can be image files or image files split from a PDF, and batch uploading is supported.

[0065] The batch upload and task management module is used to receive answer sheet images uploaded by teachers, create marking tasks, and record upload batches, exam information, class information, student information, and scanned file status. The system can queue uploaded files and record the recognition status, abnormal status, and manual review status of each answer sheet.

[0066] The image preprocessing module is used to automatically process the scanned answer sheet image, including image orientation correction, tilt correction, boundary cropping, grayscale conversion, binarization, noise filtering, brightness and contrast adjustment, paper deformation correction, and positioning point recognition, in order to improve the accuracy of subsequent OMR recognition.

[0067] The answer sheet layout positioning module locates the answer sheet area in the uploaded image based on the layout parameters and positioning point information recorded when the answer sheet PDF was generated. The system maps the coordinates of the scanned image to the coordinates of the standard answer sheet template to determine the location of the student information area, the objective question marking area, and the subjective question answering area.

[0068] The OMR (Optical Marking and Recognition) module is used to identify the marking on answer sheets in the answer areas for objective questions. Based on the number of options and area coordinates for each question, the system detects the blackness, area, contour features, and threshold results of each option's marking to determine the student's chosen option. For multiple-choice questions, single-choice questions, unanswered questions, suspected multiple markings, and unclear corrections, the system can mark them and initiate either automatic scoring or manual review.

[0069] The student identity and exam paper matching module is used to bind scanned images to specific students, exams, and exam papers based on the admission ticket number, student ID, name area, or other identifying information on the answer sheet. The system can automatically identify and match student identities, or initiate a manual confirmation process if identification fails.

[0070] The objective question automatic scoring module automatically calculates the objective question score based on the standard answers, question type rules, scoring rules, and OMR recognition results from the test paper generation system. For different question types such as multiple choice, true / false, etc., scoring rules can be configured separately, such as full marks for correct answers, no marks for incorrect answers, and proportional marks for missed answers.

[0071] The scoring report is used to grade the subjective question answer area according to the grading process, calculate the subjective question scores, and combine them with the objective question scores to generate a student grade sheet.

[0072] In this embodiment, the subjective question manual grading module allows teachers to manually grade the subjective questions on the answer sheet. Teachers can view students' subjective question answers in the web backend and enter scores and comments. The system automatically sums the subjective question scores and objective question scores to generate a total score and grade report.

[0073] The grade summary and report generation module aggregates data such as automatic score recognition for objective questions, manual score recognition for subjective questions, total score, ranking, question score details, and incorrect question details to form a student grade report. The report can display the answer, correct answer, score, and explanation entry for each question.

[0074] The student mini-program query module allows students to view their exam scores, total scores, individual question scores, and incorrect answers. Since the system is integrated with a question bank and exam paper generation system, students can click on a question number in the score report to view the question stem, options, correct answer, and explanation, achieving a closed-loop learning process from score results to question review.

[0075] The anomaly handling and manual review module automatically marks issues such as low confidence levels, blurry images, missing answer sheet positioning points, mismatched student identities, improper marking, multiple markings, or suspected alterations as anomalies, and provides a manual review entry point to ensure accurate and reliable marking results.

[0076] An electronic device, comprising: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement any of the methods in the above scheme.

[0077] A storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the answer sheet dynamic generation and adaptive optical marker recognition method as described above.

[0078] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data, characterized in that, include: Based on the examination subject syllabus, test questions are extracted to generate examination papers for the examination subjects, and corresponding answer sheets are constructed in combination with the structure of the test papers. Scan and clean the answer sheets after the exam to obtain a standard answer sheet image, and use the preset answer sheet template to locate the areas for marking objective questions and answering subjective questions; Identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence; The subjective question answering area is graded according to the grading and transfer mechanism, the subjective question scores are calculated, and the objective question scores are combined to generate a student grade sheet.

2. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 1, characterized in that, The process of extracting test questions based on the examination syllabus, generating examination papers, and constructing corresponding answer sheets based on the paper structure includes: The exam syllabus was analyzed, and the weight of knowledge points in each chapter and the distribution of question types were extracted. The test questions are extracted based on the weight of the knowledge points in each chapter, resulting in a candidate test question set; Based on the distribution of question types and their historical usage frequency, stratified sampling was performed on the candidate question set to obtain the required questions for each question type. Based on the exam paper structure, the required questions for each question type are arranged and rendered to generate the exam paper for the exam subjects; Match the answer sheet template type according to the test paper structure to obtain a unique answer sheet template; The answer sheet template is divided according to the question type order of the test paper, and corresponding question number identifiers, objective question filling dot matrix and subjective question answer space are generated to obtain the answer sheet.

3. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 1, characterized in that, The process of scanning and cleaning the answer sheets after the exam to obtain a standard answer sheet image, and using a pre-set answer sheet template, locating the areas for marking objective questions and answering subjective questions, includes: The answer sheets were scanned one by one after the exam according to the scanning resolution to generate the original answer sheet images; The original answer sheet image is subjected to orientation recognition, the position of the orientation marker is extracted, and the image orientation is determined. If the image orientation is inconsistent with the preset standard image orientation, the original answer sheet image is rotated and corrected to obtain the corrected answer sheet image; The reference point is detected on the corrected answer sheet image to obtain the actual coordinates of the reference point, and the image tilt angle is calculated by combining the theoretical coordinates of the reference point; A rotation matrix is ​​constructed based on the image tilt angle, and an affine transformation is performed on the corrected answer sheet image to obtain a horizontal answer sheet image; Edge detection and segmentation are performed on the horizontal answer sheet image to determine the effective content image of the answer sheet; The valid content image of the answer sheet is binarized to obtain a grayscale image of the answer sheet content; The grayscale image of the answer sheet content is denoised and smoothed to obtain a standard answer sheet image, and the actual positioning markers are extracted.

4. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 3, characterized in that, The process of scanning and cleaning the answer sheets after the exam to obtain a standard answer sheet image, and using a preset answer sheet template to locate the areas for marking objective questions and answering subjective questions, also includes: The preset answer sheet template is tested, and the template positioning marks, coordinates of the objective question filling area, and coordinates of the subjective question answering area are extracted. Match the template positioning marks with the actual positioning marks, and calculate the image scaling ratio and coordinate offset; Based on the coordinates of the objective question filling area, combined with the image scaling and coordinate offset, locate the objective question filling area; The answer area for subjective questions is located by combining the coordinates of the answer area with the image scaling and coordinate offset.

5. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 1, characterized in that, The process of identifying the marked areas for objective questions, extracting the marked options, combining them with the correct option coding sequence, and calculating the objective question score includes: Map the coordinates of the areas marked on the answer sheet for objective questions to obtain the areas for filling in the marks; The filling point area is divided according to the question number to obtain the question filling points; The number of black pixels in the question's fill-in points is counted, and the black pixel density is calculated. If the black pixel density is greater than the preset filling judgment threshold, it is determined to be a valid filling option, and the number of filling options is recorded; otherwise, it is determined to be an invalid filling option. If the number of options filled in is 1, the option status is marked as normal single selection; If the number of filled-in options is greater than 1, the option status is marked as multiple selections pending verification; If the number of options filled in is 0, it is determined that the question has been missed in the option status marker. The valid fill options are converted according to the preset option arrangement order to form fill option values; The filling option values ​​and option status markers are combined according to the question number identifier to generate a filling result sequence; The answer sequence is compared with the answer sequence based on the correct answer code sequence for the objective question, and the score for the objective question is calculated by combining the answer sequence with the question score.

6. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 1, characterized in that, The grading and workflow mechanism includes workflow by question, workflow by student, workflow by class, and workflow by task.

7. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 1, characterized in that, The process of grading subjective questions according to the grading and transfer mechanism, calculating subjective question scores, and combining them with objective question scores to generate a student grade sheet includes: The subjective question answering area is divided according to the grading and circulation mechanism, generating several task packages to be graded; The task package to be graded is parsed based on the grading interface to extract the answer area image and question content; The answer area image is extracted and matched based on the subject character library to obtain the subjective question answer text field; Based on the content of the question, the standard answer is matched, and the questions are divided and assigned values ​​according to keywords to obtain several score items; Approximate matching is performed on the scoring items, and then matched with the subjective question response fields to obtain the initial subjective question scores; The initial subjective question scores are verified. If the verification is successful, the initial subjective question scores are determined to be the subjective question scores. If not, the scoring items will be revised, and the subjective question responses will be re-scored until the verification is successful; The scores of subjective questions from different marking and circulation mechanisms are combined according to the score difference control rules to obtain the final subjective question score. The final scores for subjective and objective questions are summarized based on the students' exam numbers, and then arranged according to the question number to generate a student grade sheet.

8. The method for dynamic generation of answer sheets and adaptive optical marker recognition based on test paper data as described in claim 7, characterized in that, The step of extracting and matching the answer area image based on the subject character library to obtain the subjective question answer text field includes: The image of the answer area is filtered and denoised to obtain a noise-free image of the answer area. The image of the noiseless answer area is horizontally projected and segmented to extract the characters of the answer area lines; The characters in the answer area are vertically projected and segmented to obtain several character image blocks; The character image blocks are normalized in size and their features are extracted to obtain character feature vectors; Based on the subject character library, character feature vectors are matched to calculate character similarity; If the character similarity is greater than the preset character matching threshold, the current character is determined to be a candidate character; If the number of candidate characters is one, then it is determined to be the target character; If there are multiple candidate characters, they are combined into a string sequence, and the context constraint is checked using a language model to filter the target characters. The target characters are combined to obtain the subjective question response field.

9. A dynamic answer sheet generation and adaptive optical marker recognition system based on test paper data, used to implement the dynamic answer sheet generation and adaptive optical marker recognition method as described in any one of claims 1-8, characterized in that, include: The question paper generation data terminal is used to extract questions based on the examination subject syllabus, generate examination papers for the examination subjects, and construct corresponding answer sheets based on the examination paper structure. The answer sheet generation terminal is used to scan and clean the answer sheets after the exam to obtain a standard answer sheet image, and, in conjunction with a preset answer sheet template, locate the areas for marking objective questions and answering subjective questions. The image recognition processing unit is used to identify the marked areas for objective questions, extract the marked options, and calculate the scores for objective questions by combining the correct option coding sequence. The scoring report is used to grade the subjective question answer area according to the grading process, calculate the subjective question scores, and combine them with the objective question scores to generate a student grade sheet.

10. A storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the answer sheet dynamic generation and adaptive optical marker recognition method as described in any one of claims 1 to 8.

Images