TWI655617B - Test paper automatic correction system and method thereof - Google Patents

Abstract

An automatic test paper correction method includes: detecting at least one test paper image on an test paper by using an image detection module, and the test paper image includes several pixels; using an edge detection module in each pixel of the test paper image The group captures several neighboring pixels; calculates a local intensity average value and a local signal energy variation of each neighboring pixel to obtain a grayscale image; converts the test paper image from the grayscale image to one or two Valued image; and use the binarized image to identify several data to be compared in the test paper image, and then use a standard data to compare and characterize the data to be compared to produce at least one batch of correction results.

Description

Test paper automatic correction system and method

The invention relates to a system and method for automatically reviewing test papers [review]; in particular, to a system and method for automatically reviewing test papers using image recognition [image recognition].

For example, the system and method for revising test papers are used, for example: Patent for Invention of the Republic of China Patent Announcement No. TW-I239200 "Recognition System of Documents, Recognition Method and Method for Obtaining Examination Information", which discloses a document Acquisition identification system. The document acquisition and identification system includes an image acquisition unit and an image identification unit.

The aforementioned image capturing unit No. TW-I239200 captures a single image of a document with a single photograph. The image recognition unit is electrically connected to the image capture unit, and the image recognition unit is suitable for recognizing the single image, and according to a result obtained by the recognition, provides the relevant auxiliary learning information required by the reader.

The aforementioned image recognition method No. TW-I239200 includes steps: 1. Steps of estimating light surface parameters and removing light effects on the single image; 2. Steps of character cutting; 3. Steps of character regularization ; Fourth, the steps of feature comparison. The document fetch recognition system is used to obtain information on a question on the test paper. After retrieving a tag code image of a question on the test paper, the tag code image is recognized to obtain a tag code character; then, according to The marked code character provides a piece of related information.

However, the above-mentioned document acquisition and recognition method No. TW-I239200 has complex recognition operations in Step 1, Step 2, Step 3 and Step 4. Industry and the shortcomings of being unable to accurately identify test papers. Therefore, the aforementioned image recognition method of No. TW-I239200 needs to further simplify its overall recognition steps or improve its recognition accuracy.

Another conventional test paper revision system, for example: the new patent of "Recognition Test Paper System" of the Republic of China Patent Announcement No. TW-M441891, which discloses a recognition test paper system. The identification test paper system includes a handheld device, a camera tool, a calibration system, a network transmission module, a network and a server.

The aforementioned handheld device No. TW-M441891 is used for recognizing image information into a document data, and the camera tool is embedded in the handheld device, and the camera tool is used to shoot a test paper image. The calibration system is embedded in the handheld device, and the calibration system is used to adjust the position of the test paper image. The network transmission module is embedded in the handheld device, and the network transmission module is used to transmit the text data and receive a score information. The server is used to store a standard answer of the test paper and several text data uploaded by the handheld device, and compare the standard answer of the test paper with the text data uploaded by the handheld device, so as to produce a score for review, and then the score Return to the handheld device.

The aforementioned network No. TW-M441891 is used to connect the handheld device and the server, and the handheld device can drive the camera tool to shoot a test paper, and then generate an image data, and then drive the calibration system to adjust the After recognizing the location of the image data, and then recognizing the image data as the text data, using the network transmission module to upload the text data to the server, and then receiving the results returned by the server through the network transmission module, Finally, the result is displayed.

However, the aforementioned image recognition method No. TW-M441891 needs to use the camera tool of the handheld device to capture one test paper image one by one, so it has a recognition efficiency that cannot improve the identification operation, and it needs to use the correction system to adjust the test paper image Location, so the aforementioned The image recognition method of TW-M441891 needs to further simplify its overall recognition operation or improve its recognition efficiency.

Obviously, the aforementioned method of revising the test papers of the aforementioned ROC Patent Publications No. TW-I239200 and No. TW-M441891 inevitably has a potential need to provide an improved method of identification. The aforementioned patents of the Republic of China Patent Announcements No. TW-I239200 and No. TW-M441891 are only for reference of the technical background of the present invention and description of the current state of technological development, and they are not intended to limit the scope of the present invention.

In view of this, the present invention provides an automatic test paper correction system and method in order to meet the above-mentioned needs, which uses an edge detector or an edge detection module to capture the pixels of a test paper image to be identified Several neighbor pixels, and calculate a local intensity average value and a local signal energy variation of each neighbor pixel to obtain a grayscale image, and convert the test paper image from the grayscale image to a binarization Image, and then use the binarized image to identify several pieces of data to be compared in the test paper image, and the data to be compared is located in at least one test paper mark area or at least one answer area, and then perform text recognition to improve the practice The technical shortcomings of the examination paper correction system and its methods.

The main objective of the preferred embodiment of the present invention is to provide an automatic examination paper correction system and method thereof, which uses an edge detector or an edge detection module to capture a plurality of pixels to be identified in a test paper image Neighbor pixels, and calculate a local intensity average value and a local signal energy variation of each neighbor pixel to obtain a grayscale image, and convert the test paper image from a grayscale image to a binary image, Then, the binary image is used to discriminate several pieces of data for comparison of the test paper image, and the data for comparison is located in at least one test paper mark area or at least one answer area, and then performs text recognition operation to achieve automatic correction of the test paper And the purpose of improving the efficiency of automatic correction.

In order to achieve the above object, the automatic test paper correcting system of the preferred embodiment of the present invention includes: an image detection module for detecting at least one test paper image or inputting at least one test paper image on a test paper, and the test paper image It includes several pixels to be identified; an edge detection module that uses the edge detection module to capture several neighboring pixels in each of the pixels to be identified in the test paper image; and a data processing module, It is used to calculate a local intensity average value and a local signal energy variation of each of the neighboring pixels in order to obtain a gray-scale image; wherein the test paper image is converted from the gray-scale image into a binary image and reused The binarized image is used to identify several pieces of data to be compared for the test paper image, and then a standard data is used for comparison and character recognition with the data to be compared to generate at least one batch of correction results.

In the preferred embodiment of the present invention, the pixel to be identified is set as a central pixel.

In the preferred embodiment of the present invention, the neighboring pixels are a 3x3 neighboring pixel array.

The edge detection module of the preferred embodiment of the present invention is used to detect a strong edge structure or a weak edge structure.

In the preferred embodiment of the present invention, the pixel to be recognized is located in a predetermined recognition area.

The predetermined recognition area of the preferred embodiment of the present invention includes at least one test paper marking area or at least one answering area.

In order to achieve the above objective, the method for automatically revising test papers according to a preferred embodiment of the present invention includes: using an image detection module to detect at least one test paper image or input at least one test paper image on a test paper, and the test paper image includes Several pixels to be identified; an edge detection module is used to capture several neighboring pixels in each of the pixels to be identified in the test paper image; a local intensity average and a local of each of the neighboring pixels are calculated Signal energy variation in order to obtain a gray-scale image; convert the test paper image from the gray-scale image to a binary image; and use the binary image to identify several data to be compared for the test paper image and reuse A standard data is compared with the data to be compared and the text is recognized to produce at least one batch of correction results.

The preferred embodiment of the present invention further includes detecting the front of one test paper or the back of a test paper.

The test paper of the preferred embodiment of the present invention is selected from a paper test paper, a recycled paper test paper or a writable test paper.

The image detection module of the preferred embodiment of the present invention is selected from an image pickup module or an image scanning module, so as to capture or scan the test paper.

The test paper of the preferred embodiment of the present invention is transported by an automatic transport device.

In the preferred embodiment of the present invention, the binary image is generated from a threshold image.

In the preferred embodiment of the present invention, on the automatic correction of test papers, after finding several pieces of data for comparison of the test paper images, a recognition module is used to identify the pieces of data for comparison of the test paper images to obtain several The text of the data to be compared is answered.

The recognition module of the preferred embodiment of the present invention is a text recognition module.

1‧‧‧Image detection module

10‧‧‧ digital camera

11‧‧‧Image scanning device

2‧‧‧Edge detection module

3‧‧‧Data processing module

4‧‧‧Data transmission module

40‧‧‧Data wireless transmission module

41‧‧‧ cloud server

5‧‧‧The first test paper automatic conveying device

51‧‧‧Second test paper automatic conveying device

6‧‧‧ The first test paper output

61‧‧‧The second test paper output

100‧‧‧ Examination Paper Automatic Correction System

Fig. 1: Block diagram of the automatic examination paper correcting system of the first preferred embodiment of the present invention.

Figure 2: Schematic diagram of a method for automatically revising test papers according to a preferred embodiment of the present invention.

Fig. 3: Block diagram of an automatic examination paper correcting system of the second preferred embodiment of the present invention.

Fig. 4: A schematic diagram of the automatic test paper correcting system of the third preferred embodiment of the present invention combined with the first automatic test paper conveying device.

Figure 5: A schematic diagram of the automatic test paper correction system of the fourth preferred embodiment of the present invention combined with the second automatic test paper conveying device.

In order to fully understand the present invention, preferred embodiments will be exemplified below and described in detail in conjunction with the accompanying drawings, and they are not intended to limit the present invention.

First, the automatic test paper correction system and method of the preferred embodiment of the present invention can be applied to various test papers, various test papers, various paper test papers, various paper card test papers, various writeable test papers, or other various test papers; The automatic test paper correction system of the preferred embodiment of the present invention can be appropriately selected and combined with various test paper conveying devices, such as: reel conveying device, negative pressure suction conveying device or other various paper conveying devices, but it is not intended to limit the application of the present invention range.

FIG. 1 shows a block diagram of an automatic examination paper correcting system of the first preferred embodiment of the present invention. Please refer to FIG. 1, for example, the test paper automatic correction system 100 of the first preferred embodiment of the present invention includes an image detection module 1, an edge detection module [edge detecting module] 2, a data Data processing module [data processing module] 3 and a data transmission module [data transmission module] 4, and the image detection module 1, edge detection module 2, data processing module 3 and data transmission module 4 Appropriate integration forms a single device, and configures to form an automatic examination paper correction device.

Please refer to FIG. 1 again. For example, the image detection module 1 is selected from an image capturing module or an image scanning module, so as to capture or scan at least one test paper or a series of test papers. In addition, the image detection module 1 is appropriately arranged at a predetermined position of the automatic test paper correcting device, so that the image detection module 1 can take or scan the test paper.

Please refer to Figure 1 again. For example, the test paper is selected from a paper test paper, a recycled paper test paper, a writable test paper or a similar material test paper [thin paper], and the test paper can be selected as the general The test paper, and the test paper includes a front of the test paper or a back of the test paper, and the front of the test paper or the back of the test paper can be used for answering, or a front label or a back label is appropriately configured.

The automatic test paper correction system of the preferred embodiment of the present invention is suitable for users [for example: candidates] who may cause stains, ink stains, alteration or graffiti on the front or back of the test paper when writing or answering, or even possible Accidentally rubbing or pulling the test paper causes wrinkles in the test paper itself, so the test paper needs to be properly identified.

Please refer to FIG. 1 again. For example, the edge detection module 2 [or edge detector] can be connected to the image detection by wired [cable] or wireless [wireless] according to various needs. Module 1 so that the edge detection module 2 can receive the data output by the image detection module 1 and perform appropriate subsequent related processing on the edge detection module 2.

Please refer to FIG. 1 again. For example, the data processing module 3 can be connected to the edge detection module 2 in a wired or wireless manner according to various needs, so that the data processing module 3 can receive the The data output by the edge detection module 2 is subjected to appropriate subsequent processing in the data processing module 3.

Please refer to FIG. 1 again. For example, the data transmission module 4 can be connected separately or simultaneously in a wired or wireless manner according to various needs. Communicate with the edge detection module 2 or the data processing module 3, so that the data transmission module 4 can receive the data output from the edge detection module 2 or the data processing module 3, and Group 4 performs appropriate follow-up processing.

FIG. 2 shows a schematic flow chart of a method for automatically revising test papers according to a preferred embodiment of the present invention. Please refer to FIG. 1 and FIG. 2, the method for automatically revising test papers according to the preferred embodiment of the present invention includes step S1: First, for example, the image detection module 1 is used to detect one or a series of the test papers At least one test paper image [ W x H ] or input at least one test paper image, and the test paper image includes several pixels to be recognized. In addition, the pixel to be recognized is located in a predetermined recognition area, and the predetermined recognition area includes at least one test paper marking area or at least one answering area.

The method for automatically correcting test papers according to another preferred embodiment of the present invention further includes the steps of detecting the front side of one test paper or the back side of a test paper with appropriate technical means, so as to identify the front image of the test paper and the test paper image or predetermined identification of the back of the test paper region.

The method for automatically correcting test papers according to another preferred embodiment of the present invention further includes the steps of: transporting the test papers in an automatic or semi-automatic manner in a conveying system by appropriate technical means [using an automatic conveying device] to improve the efficiency of automatic correcting.

Referring again to FIGS. 1 and 2, the method for automatically revising test papers according to the preferred embodiment of the present invention includes step S2: Then, for example, the edge detection mode is used in each of the pixels to be recognized in the test paper image Group 2 [or select the appropriate operator, operator] captures several neighborhood pixels [neighborhood pixels]. For example, the pixel to be recognized is set as a single center pixel [single center pixel] I ( x , y ), and the neighboring pixels are selected as a 3x3 neighboring pixel array or other neighboring pixel arrays. The edge detection module is used to detect a strong edge structure or a weak edge structure.

Referring again to FIGS. 1 and 2, the method for automatically revising test papers according to the preferred embodiment of the present invention includes step S3: Then, for example, the data processing module 3 or other processing modules are used to calculate each neighbor A local intensity mean of one of the pixels in the domain and a local signal energy variation to obtain a grayscale image.

For example, the preferred embodiment of the present invention chooses to use the local intensity average and local signal energy variation as follows:

Where μ _l is the average local intensity, I ( x + i , y + j ) is the pixel, and ε is the local signal energy variation. N = 9 as normalizing constant

Please refer to Figures 1 and 2 again. For example, the preferred embodiment of the present invention selects and adopts gray scale conversion as follows:

Wherein W is the image width, H is the image height, μ _g is the gray scale threshold.

Referring again to FIGS. 1 and 2, the method for automatically revising test papers according to the preferred embodiment of the present invention includes step S4: Then, for example, the data processing module 3 or other processing modules are used to select the test paper image from The grayscale image is converted into a binary image [binary image], and the binary image is generated from a threshold image.

Please refer to Figures 1 and 2 again. For example, the preferred embodiment of the present invention selects and adopts the conversion into a binary image:

Where B ( x , y ) is a binary pixel.

Referring again to FIGS. 1 and 2, the method for automatically revising test papers according to the preferred embodiment of the present invention includes step S5: Then, for example, the data processing module 3 or other processing modules are selected to use the binarization. The image is used to identify several data to be compared for the image of the test paper, and then a standard data (for example, answer or student number) is compared with the data to be compared to produce at least one batch of correction results.

In the preferred embodiment of the present invention, on the automatic correction of test papers, after finding several pieces of data (edges) of the test paper images to be compared, a recognition module (for example, a text recognition module) is used to recognize the number of the test paper images Data to be compared, to obtain a number of answers to the text of the data to be compared [for example: A, B, C, D or 1, 2, 3, 4], in order to carry out subsequent automatic correction of test papers.

For example, the preferred embodiment of the present invention, after arithmetic processing, contrast defects obtained have underscored the boundary feature data classification clear out here using a fast nearest neighbor classification algorithms [k -NN, k -nearest neighbor], and further use the training database generated by Principal Component Analysis (PCA, Principal Component Analysis), that is, the principal component analysis training database to perform sample identification and discrimination operations.

For example, the preferred embodiment of the present invention uses PCA to select more discriminative features for image data as the basis for detection and discrimination, and the most important goal in the training process is to obtain test images that can be converted into identifying features The transformation matrix w and the feature space database P _ALL as the comparison similarity are as follows: P _ALL = w ^T X _ALL

X _ALL is composed of x ⁽¹⁾ , x ⁽²⁾ , ..., x ^{( n )} training data, and n is the number of image data.

Next, the preferred embodiment of the present invention calculates the respective average value for each feature, as follows:

among them Is the average value of each feature, and j = 1 ~ m , j is the total feature number of each data.

Next, the preferred embodiment of the present invention further subtracts the features in each data and the average value of each row of features and then multiplies them to generate a covariance matrix [Covariance matrix] representing the differences between the overall data, as follows formula:

Where C is the covariance matrix.

Next, the preferred embodiment of the present invention finds the eigenvalues and eigenvectors of the covariation matrix. The size of the eigenvalues represents the degree of variation in the covariation matrix. Therefore, the eigenvalues can be sorted in descending order. Ranked in front, and then select the corresponding feature vector by the selected features, you can get the conversion matrix, as follows: w = u _k

Where u _k is the feature vector, and k is the number of feature roots selected for comparison.

Finally, according to the preferred embodiment of the present invention, according to P _ALL = w ^T X _ALL , the feature space database P _ALL is converted from the input data set, and the establishment of the training database has been completed.

For example, in determining the answer A, B, C, or D, when the preferred embodiment of the present invention obtains w and P _ALL , it can enable the detection system to have the effect of transforming the data to be tested into features and provide comparable For the data data, assuming that the currently input test image has been processed, then adjust it to a single line of data to extract the feature space data of the currently detected image, and calculate the distance L between the features as a basis for classification, as follows Formula: P _V = w ^T X _V

X _V is the test image, and P _V is the feature space data of the test image.

Next, the preferred embodiment of the present invention calculates the distance L between the test image and each sample data in the training database, as follows:

Where L is the distance between the test image and the sample data, and n is the number of image data.

In the preferred embodiment of the present invention, the distance L obtained by the above formula is calculated. The smaller the value, the higher the similarity, and vice versa, the lower the similarity, thereby determining whether the test image is qualified or unqualified.

FIG. 3 shows a block schematic diagram of an automatic examination paper correction system of the second preferred embodiment of the present invention, which corresponds to the automatic examination paper correction system of FIG. 1. Please refer to FIG. 3, for example, compared with the first embodiment, the automatic test paper correction system 100 of the second preferred embodiment of the present invention chooses to use a data wireless transmission module [data wireless transmission module] 40, and The data wireless transmission module 40 can optionally connect and transmit to a cloud server 41 [cloud server] 41, so as to transmit the data to be compared, the standard data, the results of the correction or related data.

Please refer to FIG. 3 again. For example, the image detection module 1 of the automatic test paper correcting system 100 of the second preferred embodiment of the present invention may be selected from a digital camera 10 and an image scan. The device [scanner device] 11 or a device with a similar function, or is connected to the digital camera 10 or the image scanning device 11 to automatically input the test paper image.

FIG. 4 shows a schematic diagram of the automatic test paper correcting system of the third preferred embodiment of the present invention combined with the first automatic test paper conveying device. Please refer to FIG. 4, for example, compared with the first embodiment, the present invention is third The automatic test paper correcting system 100 of the preferred embodiment selects and adopts a first automatic test paper conveying device 5, and a first test paper output port 6 is provided above the automatic test paper correcting system 100.

FIG. 5 shows a schematic diagram of the automatic test paper correcting system of the fourth preferred embodiment of the present invention combined with the second automatic test paper conveying device, which corresponds to the automatic test paper correcting system of FIG. 4. Please refer to FIG. 5, for example, compared to the third embodiment, the automatic test paper correcting system 100 of the fourth preferred embodiment of the present invention chooses to use a second automatic test paper conveying device 51, and the second automatic test paper The conveying device 51 can automatically feed a stack of the test papers into the automatic test paper correcting system 100, and a second test paper output port 61 and a test paper collecting slot [not shown] are provided above the automatic test paper correcting system 100.

The foregoing preferred embodiment merely exemplifies the present invention and its technical features, and the technology of this embodiment can still be appropriately implemented with various substantially equivalent modifications and / or replacements; therefore, the scope of rights of the present invention shall be subject to the attached patent application The scope defined by the scope shall prevail. The copyright in this case is restricted to the use of patent applications in the Republic of China.

Claims (10)

An automatic examination paper correction system includes: an image detection module for detecting at least one examination paper image or inputting at least one examination paper image on an examination paper, and the examination paper image includes several pixels to be recognized; an edge A detection module, which uses the edge detection module to capture several neighboring pixels in each of the pixels to be recognized in the test paper image; and a data processing module, which is used to calculate each of the neighboring regions A local intensity average value of one pixel and a local signal energy variation to obtain a grayscale image; wherein the test image is converted from the grayscale image into a binary image, and then the binary image is used to identify the test paper Several data to be compared in the image, and then using a standard data to compare and characterize the data to be compared to produce at least one batch of correction results. According to the test paper automatic review system described in item 1 of the patent application scope, the pixel to be recognized is set as a central pixel. According to the test paper automatic revision system described in item 1 of the patent scope, the edge detection module is used to detect a strong edge structure or a weak edge structure. According to the test paper automatic revision system described in item 1 of the patent application scope, wherein the pixel to be recognized is located in a predetermined recognition area. The automatic examination paper correcting system according to item 4 of the patent application scope, wherein the predetermined identification area includes at least one examination paper marking area or at least one answering area. An automatic examination paper correcting method, comprising: detecting at least one examination paper image or inputting at least one examination paper image on an examination paper by using an image detection module, and the examination paper image includes several pixels to be identified; An edge detection module is used in each of the pixels to be identified to capture several neighboring pixels; a local intensity average and a local signal energy variation of each of the neighboring pixels are calculated to obtain a grayscale image; Convert the test paper image from the grayscale image to a binary image; and use the binary image to identify several data to be compared for the test paper image, and then use a standard data to compare with the data to be compared Recognize and identify words to produce at least one batch of correction results. The method for automatically revising test papers according to item 6 of the patent application scope, which also includes detecting the front of one test paper or the back of a test paper, and the test paper is selected from a paper test paper, a recycled paper test paper or a writable Class papers. According to the method for automatically correcting test papers described in item 6 of the patent application scope, the test papers are transported by an automatic transport device. According to the test paper automatic correction method described in item 6 of the patent scope, wherein the image detection module is selected from an image pickup module or an image scanning module, so as to take or scan the test paper. According to the method for automatically correcting test papers described in item 6 of the patent application scope, the binary image is generated from a threshold image.