A calculation problem correction method, a storage medium and an apparatus

Abstract

The application discloses a calculation question correcting method, a storage medium and equipment, and through the recognition and calculation of the calculation question stems in the template images in the un-answered state, the correct answers of the calculation questions can be accurately obtained, manual intervention can be effectively reduced, manpower cost can be saved, and the accuracy of the automatic correction of the calculation questions can be ensured. In addition, through the matching of the question regions, the corresponding relationship of each calculation question in the template images and the to-be-corrected images can be accurately established.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent correction, in particular to a calculation question correction method, a storage medium and equipment. BACKGROUND

[0002] In recent years, through the use of artificial intelligence technology, especially text recognition technology, automatic correction of homework has been realized, greatly improving the work efficiency of teachers and freeing them from the tedious task of homework correction, so that they can devote themselves better to teaching work.

[0003] Currently, the correction of calculation questions is mainly based on text recognition results, that is, the content of the calculation question stem is detected, and the content of the stem is calculated to obtain the first answer of the question, then the student's answer is recognized as the second answer of the question, and when the first answer and the second answer are consistent, the correction result is correct, and when they are inconsistent, the correction result is incorrect.

[0004] However, due to the fact that students often fill in, divide, and perform other operations in the stem area during the answering process, the calculation question correction based on text recognition results often has errors in recognizing the content of the stem or cannot recognize it, resulting in errors in calculating the first answer or being unable to calculate it, which seriously affects the accuracy of the calculation question correction based on text recognition results. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application aims to provide a calculation question correction method, a storage medium and equipment.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] A calculation question correction method, comprising the following steps:

[0008] S1, obtaining a template image corresponding to a to-be-corrected test paper or teaching aid page; the template image is an original image of the test paper or teaching aid page to be corrected in an unanswerable state;

[0009] S2, obtaining a to-be-corrected test paper or teaching aid page image;

[0010] S3, aligning the features of the to-be-corrected test paper or teaching aid page image and the template image to correct the to-be-corrected test paper or teaching aid page image;

[0011] S4, respectively, on the template image and the image of the to-be-corrected test paper or teaching aid page, a calculation question is extracted, and the question information of the calculation question of the template image is obtained as first information, and the question information of the calculation question of the to-be-corrected test paper or teaching aid page image is obtained as second information; the question information in the first information includes the stem of the calculation question and the question area coordinates of the calculation question, and the question information in the second information includes the question area coordinates of the calculation question;

[0012] S5, the stem of each calculation question extracted from the template image is calculated to obtain a calculation result and is respectively the first answer of each calculation question, and then the first answer of each calculation question is associated with the question area coordinates thereof in the template image;

[0013] S6, in the question area of each calculation question extracted from the to-be-corrected test paper or teaching aid page image, a handwritten answer result is recognized as the second answer of each calculation question, and the second answer of each calculation question is associated with the question area coordinates thereof in the to-be-corrected test paper or teaching aid page image;

[0014] S7, the question area coordinates in the second information are matched with the question area coordinates in the first information, so that each calculation question in the template image is corresponded to each calculation question in the to-be-corrected test paper or teaching aid page image;

[0015] S8, the first answer of each calculation question in the template image is compared with the second answer of the corresponding calculation question in the to-be-corrected test paper or teaching aid page image, when the first answer and the second answer are consistent, the correction result is correct, when they are inconsistent, the correction result is wrong.

[0016] Further, in step S4, a calculation question target detection model is used to detect the question area of each calculation question on the template image and the to-be-corrected test paper or teaching aid page image, and the question area of each calculation question is positioned and extracted to obtain a question area image and question area coordinates; for the template image, the question area image is further subjected to text recognition to obtain the stem of the calculation question.

[0017] Further, in step S7, when a question area in the second information overlaps with a question area in the first information, it is considered that the corresponding two calculation questions correspond.

[0018] Further, in step S7, if a question area in the second information overlaps with multiple question areas in the first information, the intersection-over-union of the question area in the second information and each question area in the first information is calculated, and the question area in the first information corresponding to the maximum intersection-over-union calculated is corresponded to the question area in the second information.

[0019] Intersection and union ratio = M (A intersection B) / M (A union B), A is a certain subject area in the second information, B is a certain subject area in the first information, A intersection B represents the overlapping area of A and B, A union B represents the area after A and B are combined, and M represents the area.

[0020] The application further provides a computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and the computer program is executed by a processor to implement the method.

[0021] The application further provides a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method.

[0022] The application has the advantages that the application can accurately obtain the correct answers of the calculation questions by recognizing and calculating the stems of the calculation questions in the template images in the unanswered state, effectively reduces manual intervention, saves labor cost, and guarantees the accuracy of automatic correction of the calculation questions. In addition, the application can accurately establish the corresponding relationship of each calculation question in the template images and the images to be corrected by matching the subject areas. DETAILED DESCRIPTION

[0023] The application will be further described below. It should be noted that the embodiment is based on the technical solution, and detailed implementation and specific operation processes are given, but the protection scope of the application is not limited to the embodiment.

[0024] The embodiment provides a calculation question correction method, comprising the following steps:

[0025] S1, a template image corresponding to a test paper or a teaching aid page to be corrected is obtained; the template image is an original image of the test paper or the teaching aid page to be corrected in an unanswered state. The template image can be an electronic image directly obtained after typesetting on a computer, or can be obtained by scanning.

[0026] S2, a test paper or a teaching aid page image to be corrected is obtained. The test paper or the teaching aid page image to be corrected is an image of a test paper or a teaching aid page completed by a student, and can be obtained by photographing, scanning or the like.

[0027] S3, the test paper or the teaching aid page image to be corrected is aligned with the template image, and the test paper or the teaching aid page image to be corrected is corrected. After the feature alignment, the test paper or the teaching aid page image to be corrected and the template image can realize that the subject area ranges are basically consistent.

[0028] S4, respectively, on the template image and the to-be-corrected test paper or the teaching aid page image, a calculation question is extracted, and the question information of the calculation question of the template image is obtained as first information, and the question information of the calculation question of the to-be-corrected test paper or the teaching aid page image is obtained as second information; the question information in the first information includes a stem of the calculation question and a question region coordinate of the calculation question, and the question information in the second information includes the question region coordinate of the calculation question.

[0029] Specifically, in step S4, a calculation question target detection model is used to detect the question region of each calculation question on the template image and the to-be-corrected test paper or the teaching aid page image, and the question region of each calculation question is positioned and extracted respectively to obtain a question region image and a question region coordinate; for the template image, the question region image is further subjected to text recognition to obtain the stem of the calculation question. The calculation question target detection model can use a YOLOv model based on deep learning and be pre-trained and optimized.

[0030] S5, the stem of each calculation question extracted from the template image is calculated to obtain a calculation result and is used as a first answer of each calculation question, and then the first answer of each calculation question is associated with the question region coordinate thereof in the template image.

[0031] S6, in the question region of each calculation question extracted from the to-be-corrected test paper or the teaching aid page image, a handwritten answer result is recognized as a second answer of each calculation question, and the second answer of each calculation question is associated with the question region coordinate thereof in the to-be-corrected test paper or the teaching aid page image.

[0032] In order to improve the recognition accuracy of the handwritten answer result, a constraint condition can be added to the handwritten answer result, for example, the handwritten answer result should be located on the right side of the “=” in the question region.

[0033] S7, the question region coordinate in the second information is matched with the question region coordinate in the first information, so that each calculation question in the template image is corresponded to each calculation question in the to-be-corrected test paper or the teaching aid page image.

[0034] Specifically, when a certain question region in the second information overlaps with a certain question region in the first information, it can be considered that the corresponding two calculation questions correspond to each other.

[0035] More specifically, if a certain question region in the second information overlaps with multiple question regions in the first information, the intersection-over-union of the certain question region in the second information and each question region in the first information is calculated respectively, and the certain question region in the second information is corresponded to the question region in the first information corresponding to the maximum intersection-over-union calculated.

[0036] Intersection and union ratio = M (A∩B) / M (A∪B), A is a certain subject area in the second information, B is a certain subject area in the first information, A∩B represents the overlapping area of A and B, A∪B represents the area after A and B are combined, and M represents the area.

[0037] S8, compare the first answer of each calculation question in the template image with the second answer of the corresponding calculation question in the to-be-corrected test paper or the teaching aid page image, when the first answer is consistent with the second answer, the correction result is correct, and when the first answer is inconsistent with the second answer, the correction result is incorrect.

[0038] For those skilled in the art, various corresponding changes and modifications can be given according to the above technical solutions and concepts, and all these changes and modifications should be included in the protection scope of the claims of the present application.

Claims

1. A method of correcting a calculation problem, characterized by, The method comprises the following steps: S1, obtaining a template image corresponding to a to-be-corrected test paper or teaching aid page; the template image is an original image of the test paper or teaching aid page in a non-answered state; S2, obtaining a to-be-corrected test paper or teaching aid page image; S3, aligning features of the to-be-corrected test paper or teaching aid page image and the template image to correct the to-be-corrected test paper or teaching aid page image; S4, extracting calculation questions from the template image and the to-be-corrected test paper or teaching aid page image respectively, obtaining question information of the calculation questions in the template image as first information and question information of the calculation questions in the to-be-corrected test paper or teaching aid page image as second information; the question information in the first information comprises a question stem of a calculation question and a question region coordinate of the calculation question, and the question information in the second information comprises a question region coordinate of a calculation question; S5, calculating the question stem of each calculation question extracted from the template image to obtain a calculation result as a first answer of each calculation question, and then associating the first answer of each calculation question with the question region coordinate of the calculation question in the template image; S6, recognizing a hand-written answer result in the question region of each calculation question extracted from the to-be-corrected test paper or teaching aid page image as a second answer of each calculation question, and associating the second answer of each calculation question with the question region coordinate of the calculation question in the to-be-corrected test paper or teaching aid page image; S7, matching the question region coordinate in the second information with the question region coordinate in the first information to correspond each calculation question in the template image with each calculation question in the to-be-corrected test paper or teaching aid page image; S8, comparing the first answer of each calculation question in the template image with the second answer of the corresponding calculation question in the to-be-corrected test paper or teaching aid page image, and when the first answer is consistent with the second answer, the correction result is correct, and when the first answer is inconsistent with the second answer, the correction result is incorrect.

2. The method of claim 1, wherein, In step S4, a calculation question target detection model is used to detect the question region of each calculation question on the template image and the to-be-corrected test paper or teaching aid page image, and the question region of each calculation question is positioned and extracted to obtain a question region image and a question region coordinate; for the template image, the question region image is further subjected to text recognition to obtain the question stem of the calculation question.

3. The method of claim 1, wherein, In step S7, when a question region in the second information overlaps with a question region in the first information, the corresponding two calculation questions are considered to be corresponding.

4. The method of claim 3, wherein, In step S7, if a question region in the second information overlaps with multiple question regions in the first information, the intersection-over-union of the question region in the second information and each question region in the first information is calculated, and the question region in the first information corresponding to the maximum intersection-over-union calculated is considered to correspond to the question region in the second information. Intersection-over-union = M(A∩B) / M(A∪B), A is a question region in the second information, B is a question region in the first information, A∩B represents an overlapping region of A and B, A∪B represents a combined region of A and B, and M represents an area.

5. A computer readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method in any one of claims 1-4.

6. A computer device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method in any one of claims 1-4.