TWM578421U - Character image recognition system - Google Patents

Abstract

The character image recognition system of the file of the present invention is configured to identify a character in a file to be recognized, the file to be identified includes a plurality of character images, and the character image recognition system comprises a character segmentation area recognition module, A textual analysis module, a character segmentation area label module and an output module. A character segmentation area recognition module identifies at least one character segmentation region, the character segmentation region including at least one of the above-described character images. a text analysis module, communicably connecting the character segmentation area identification module, wherein the text analysis module converts the character image in the character segmentation area into an editable character, and the editable character Proofreading. A character division area label module, the character division area label module identifies a relative position of the character division area in the to-be-identified file, and labels the character division area. An output module outputs the editable character, a relative position of the character segmentation area in the to-be-identified file, and a label corresponding to the character segmentation area.

Description

Character image recognition system

The present invention refers to an identification system, and more particularly to a character image recognition system.

At present, in order to effectively reduce the errors in the input of the paper diagnosis book or related documents and improve the efficiency of input, the insurance company uses OCR (Optical Character Recognition) in the operation of inputting the diagnosis or the related documents. Technique to automatically identify the internal image of the diagnostic book or the related document into the corresponding input field. However, today's OCR software usually needs to be matched with a specified image scanner to ensure the accuracy of character recognition. In addition, when the word of the diagnosis or the related document is illegible or dirty, the existing OCR technology cannot correctly recognize the character. In this way, it takes a lot of manpower to perform verification and identification again. Therefore, how to improve the accuracy of the OCR technology to recognize character images is worthy of consideration by those who have common knowledge in the field.

A character image recognition system for identifying a character in a file to be identified, the file to be identified includes a plurality of character images, the character image recognition system includes a character segmentation area recognition module, and a text The analysis module, the one-word segmentation area label module and an output module. The one-character segmentation area identifying module identifies at least one character segmentation area, and the character segmentation area includes at least one of the character image. In addition, a text analysis module communicatively connects the character segmentation area identification module, and the textual analysis module converts the character image in the character segmentation area into an editable character, and edits the character Characters are proofread. And a character segmentation area label module, the character division area label module identifies a relative position of the character division area in the to-be-identified file, and labels the character division area, and an output module And outputting the editable character, the relative position of the character segmentation area in the to-be-identified file, and the label corresponding to the character segmentation area. For example, the character image recognition system further includes a server end and a client, the client having a display screen, wherein the character segmentation area identification module and the text analysis module The character segmentation area label module and the output module are disposed on the servo end, and the output module outputs the editable character to the client and displayed on the display screen. The above described features and advantages will be more apparent from the following description.

The present invention is best understood by reference to the detailed description and the accompanying drawings set forth herein. Various embodiments are discussed below with reference to the drawings. However, those skilled in the art will readily appreciate that the detailed description of the drawings herein is for the purpose of explanation and description For example, the teachings presented and the needs of a particular application may result in a variety of alternative and suitable methods for implementing the functionality of any of the details described herein. Thus, any method may extend beyond the specific implementation selections in the following embodiments described and illustrated. Certain terms are used throughout the description and following claims to refer to particular elements. It should be understood by those of ordinary skill in the art that hardware manufacturers may refer to the same elements by different nouns. The scope of this specification and the subsequent patent application do not use the difference of the names as the means for distinguishing the elements, but the difference in function of the elements as the criterion for distinguishing. The term "including" as used throughout the specification and subsequent claims is an open term and should be interpreted as "including but not limited to". In addition, the term "coupled" is used herein to include any direct and indirect electrical connection. Therefore, if a first device is coupled to a second device, it means that the first device can be directly electrically connected to the second device or indirectly electrically connected to the second device through other devices or connection means. Please refer to FIG. 1. FIG. 1 illustrates an embodiment of the novel character image recognition system. The character image recognition system 100 includes a character segmentation area recognition module 110, a text analysis module 120, a character segmentation area label module 130, and an output module 140. The character image recognition system 100 is also electrically Connected to an image input device 10, such as a scanning device or a digital camera. By means of the image input device 10, a file to be recognized (Fig. 2A) can be imported into the character image recognition system 100. In this embodiment, the character segmentation area identification module 110, the semantic analysis module 120, the character segmentation area label module 130, and the output module 140 are disposed on the server end 102, and the server end 102 is, for example, a One or more servers. In addition, the output module 140 is electrically connected to a client 104. The client 104 can be an electronic device having a display screen 104a, such as a personal computer, a notebook computer, or a smart phone. In addition, please refer to FIG. 2A , which illustrates one embodiment of the file to be identified. In this embodiment, the file to be identified is a diagnostic certificate. As can be seen from FIG. 2A, the file to be identified includes a plurality of characters, and when the image of the file to be recognized is captured by the image input device 10, the characters on the file to be recognized are of course also imaged, that is, The characters that the image input device 10 imports into the file to be recognized of the character image recognition system 100 cannot be edited, and these characters are hereinafter referred to as character images. In addition, please refer to FIG. 3 at the same time, and FIG. 3 illustrates an embodiment of the present character image recognition method. First, the step S210 is implemented to import the file to be identified as shown in FIG. 2A, and the detailed process thereof is as described above, and details are not described herein again. Next, step S220 is implemented to recognize the character segmentation area 81 in the file to be identified. In FIG. 2B, the character division area 81 is an area framed by a broken line, and the character division area 81 is recognized by, for example, the character division area recognition module 110. As is clear from Fig. 2B, the character segmentation area 81 selects the character image on the file to be identified. In a preferred embodiment, after step S220 is implemented, step S222 is further implemented to divide the character segmentation area 81 into a set of demand words and a set of non-demand words. The set of required characters refers to a set of characters that need to be output in subsequent processing, such as the area indicated by the component symbol 81a in Fig. 2B. The non-required character set refers to a set of characters that need not be output in subsequent processing, such as the area indicated by the component symbol 81b in Fig. 2B. In more detail, in FIG. 2B, the stamp "below blank" is classified as a non-required character set because it may be irrelevant for subsequent data processing. Thereafter, in step S230, the character image in the character segmentation area 81 is converted into an editable character by the semantic analysis module 120. That is to say, the image of the image to be recognized originally imported by the image input device 10 is uneditable; however, the semantic analysis module 120 can convert the image of the character into an editable character. For example, it is a technique using OCR (Optical Character Recognition). However, if the technique of OCR is simply adopted, when the character image on the document to be identified is blurred or dirty, a discriminating error may occur. For example, the original "Double and Hospital" was created with the characters shown in the top of Figure 4 because it was contaminated. If you only use OCR technology, you may identify "Double Hospital" as "Double Hospital". (As shown in the center of Figure 4). However, in this embodiment, the semantic analysis module 120 may perform step S240 to correct the converted editable character, for example, correcting "Shuangli Hospital" into "Double and Hospital" (as shown in FIG. 4). Shown). Then, in step S250, the character segmentation area tag module 130 can recognize the relative position of each character segmentation area 81 in the file to be identified, and label the character segmentation area 81. For example, in the file to be identified in FIG. 2B, the character segmentation area label module 130 can recognize the relative position of the "name" field in the file to be identified, and also the field name "name" is called " The name of the label. Then, in step S260, the output module 140 outputs the editable character, the relative position of the character segmentation area 81 in the file to be identified, and the label corresponding to the character segmentation area 81 to the client 104. Thus, the characters on the file to be identified can be displayed on the display screen 104a of the client 104. It should be noted that since the output module 140 also outputs the relative position of the character segmentation area 81 in the file to be identified, each character on the display screen 104a of the client 104 is also reproduced on the file to be identified. position. That is to say, if the file to be identified is a file in the form of a table (as shown in FIG. 2A), after being processed by the character image recognition system 100 of the present invention, not only the editable characters but also the characters can be reproduced. Edit the position of the character in the table, which is not possible with the general OCR software. In addition, since the label of the character segmentation area 81 is also output, the client 104 can further process the received data, for example, into an Excel table or by the database management system to identify the data in the file. Organize and categorize. For example, if the data of the diagnostic certificate is obtained by the general OCR software, even if the characters are accurately distinguished, the user needs to manually set the correspondence between the patient and the disease name. However, with the character image recognition system 100, the correspondence between the patient and the disease name can be automatically established, thereby reducing labor costs. In summary, compared with the general OCR software, the character image recognition system 100 of the present invention has the function of correcting the characters that are recognized incorrectly, reproducing the relative positions of the characters in the file to be identified, and the segmentation area 81 of the pair of characters. Labeling to facilitate subsequent data processing and other advantages. In the above embodiment, the character segmentation area recognition module 110 recognizes, for example, the character segmentation area 81 in the file to be identified by a first type of neural network model 112, for example, the first type of neural network model 112. The training is performed by means of deep learning in order to improve the accuracy of the first type of neural network model 112. The training method of the first type of neural network model 112 will be described below. Please also refer to FIG. 5, which is illustrated in FIG. Flow chart of the first type of neural network model in the training phase. First, in step S310, for a certain category of files to be identified (for example, the diagnostic certificate shown in FIG. 2A), a certain number of N1 image samples of the same category are obtained (for convenience of distinction, N1 may be referred to elsewhere in this document). The first number). Then, step S320 is implemented to divide the collected image samples into a training set, a test set, and a verification set. Next, step S330 is implemented to use the training set as a training sample of a neural network model, and the neural network model is trained to generate a prototype of the first type of neural network model 112. Then, step S340 is performed to verify the correctness of the prototype of the first type of neural network model 112 by using the test set. If the verification pass rate is less than a first preset threshold (for example, 90%), step S350 is performed, if otherwise, the steps are implemented. S360. Here, the so-called pass means that the first type of neural network model 112 can correctly recognize the character segmentation area 81. In a preferred embodiment, the character segmentation area 81 can be further classified into a set of required character sets. Demand character set. In step S350, a certain number of image samples of the same category of N2 need to be acquired again (for convenience of distinction, N2 may be referred to as a second number elsewhere in the text), and steps S320 to S340 are repeated. In step S360, the correctness of the first type of neural network model 112 is verified by using the verification set. If the verification pass rate is less than a second predetermined threshold (for example, 95%), step S370 is performed, otherwise step S380 is implemented. In step S370, a certain number of N3 image samples of the same category are acquired again (for convenience of distinction, N3 may be referred to as a third number elsewhere in the text), and step S360 is repeated. In step S380, the training of the first type of neural network model 112 is completed, that is, the first type of neural network model 112 can be put into practical use. In the above embodiment, the semantic analysis module 120 corrects the editable characters by, for example, a second type of neural network model 122. For example, the "Shuangli Hospital" is corrected to "double" as shown in FIG. And hospital." The second type of neural network model 122 is trained, for example, by means of deep learning, in order to improve the accuracy of the second type of neural network model 122. The training method of the second type of neural network model 122 will be described below, also refer to FIG. 6 and FIG. 6 are flowcharts showing a second type of neural network model in a training phase. First, in step S410, for a certain character segmentation area 81 in the file to be identified, a certain number of editable character samples of the character segmentation area of N4 are obtained (for convenience of distinction, N4 may be referred to elsewhere in this document) For the fourth quantity). For example, the area indicated by the symbol 81a in Fig. 2B refers to the hospital name. In step S410, the hospital name of the fourth number N4 is input. Next, step S420 is implemented to divide the collected editable character samples into a training set, a test set, and a verification set. Next, step S430 is implemented to use the training set as a training sample of a neural network model, and the neural network model is trained to generate a prototype of the second type neural network model 122. Then, step S440 is performed to verify the correctness of the prototype of the second type of neural network model 122 by using the test set. If the verification pass rate is less than a first preset threshold (for example, 90%), step S450 is performed, if otherwise, the steps are implemented. S460. Here, the so-called pass means that the second type of neural network model 122 can correctly correct the error of the editable character. In step S450, a certain number N5 of editable character samples of the same category are acquired again (for convenience of distinction, N5 may be referred to as a fifth number elsewhere in the text), and steps S420 to S440 are repeated. In step S460, the correctness of the second type of neural network model 122 is verified by using the verification set. If the verification pass rate is less than a second predetermined threshold (for example, 95%), step S470 is performed, otherwise step S480 is implemented. In step S470, a certain number of N6 editable character samples of the same category are acquired again (for convenience of distinction, N6 may be referred to as a sixth number elsewhere in the text), and step S460 is repeated. In step S480, the training of the second type of neural network model 122 is completed, that is, the second type of neural network model 122 can be put into practical use. In the above embodiment, the character segmentation area label module 130 recognizes the relative position of each character segmentation area 81 in the file to be identified, for example, by a third type of neural network model 132, and segmentes the characters. Area 81 is tagged. The third type of neural network model 132 is trained, for example, by means of deep learning, in order to improve the accuracy of the third type of neural network model 132. The training method of the third type of neural network model 132 will be described below, and also refers to FIG. 7 and FIG. 7 are flowcharts showing a third type of neural network model in a training phase. First, in step S510, for a character segmentation area 81 in the file to be identified, a label of the character segmentation area 81 of a seventh number N7 and a relative position in the file to be identified are acquired. The tag of the character segmentation area 81 is, for example, manually labeled first and then input into a neural network model. For example, a plurality of data taggers may be hired to indicate which side of the file to be identified, and the patient's name is in the file to be identified. Where? For how to mark, you can refer to another patent application (application number: 107140893) of the applicant. Next, step S520 is implemented to divide the collected samples of the labels of the character segmentation areas 81 into a training set, a test set, and a verification set. Next, step S530 is implemented to use the training set as a training sample of a neural network model, and the neural network model completes the training to generate a prototype of the third type neural network model 132. Then, step S540 is implemented to verify the correctness of the prototype of the third type neural network model 132 by using the test set. If the verification pass rate is less than a first preset threshold (for example, 90%), step S550 is performed, if otherwise, the steps are implemented. S560. Here, the so-called pass means that the third type of neural network model 132 can correctly recognize the relative position of each character segmentation area 81 in the file to be identified, and label the character division area 81. In step S550, a certain number of N8 tag samples of the same category are acquired again (for convenience of distinction, N8 may be referred to as an eighth number elsewhere in the text), and steps S520 to S540 are repeated. In step S560, the correctness of the third type of neural network model 132 is verified by using the verification set. If the verification pass rate is less than a second predetermined threshold (for example, 95%), step S570 is performed, otherwise step S580 is implemented. In step S570, a certain number of N9 tag samples of the same category are acquired again (for convenience of distinction, N9 may be referred to as a ninth number elsewhere in the text), and step S560 is repeated. In step S580, the training of the third type of neural network model 132 is completed, that is, the third type of neural network model 132 can be put into practical use. In the above embodiment, whether it is the first type of neural network model 112, the second type of neural network model 122, or the third type of neural network model 132, the samples are divided into training sets and test sets during training. And the verification set, first trained by the training set, and then tested by the test set, and if passed, verified by the verification set. Compared with the conventional neural network model, only the samples are divided into a training set and a test set during training. The first type of neural network model 112, the second type of neural network model 122, and the third type of neural network are the case. The road model 132 can achieve a higher accuracy rate. In addition, in the foregoing embodiment, the first type of neural network model 112, the second type of neural network model 122, or the third type of neural network model 132 may be a Recurrent Neural Network, Long Short-Term Memory or Convolutional Neural Network, please note that this is only an embodiment of the present invention and is not a limitation of the present invention. Although the present invention has been disclosed in the above preferred embodiments, it is not intended to limit the present invention, and any person skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of this new type is subject to the definition of the scope of the patent application.

10‧‧‧Image input device

100‧‧‧ character image recognition system

110‧‧‧ character segmentation area recognition module

112‧‧‧First type of neural network model

120‧‧‧literal analysis module

122‧‧‧Second type neural network model

130‧‧‧Character segmentation area label module

132‧‧‧The third type of neural network model

140‧‧‧Output module

102‧‧‧Server

104‧‧‧Client

104a‧‧‧Electronic device

81‧‧‧ character segmentation area

81a‧‧‧Component symbol

81b‧‧‧Component symbol

S210~S260‧‧‧ character image recognition method implementation steps

S310~S380‧‧‧ Implementation steps of the first type of neural network model in the training phase

S410~S480‧‧‧ implementation steps of the second type of neural network model in the training phase

S510~S580‧‧‧ implementation steps of the third type of neural network model in the training phase

The various embodiments are described below with reference to the accompanying drawings, in which An embodiment of a character image recognition system. 2A and 2B illustrate one embodiment of a file to be identified. FIG. 3 illustrates an embodiment of a novel character image recognition method. FIG. 4 illustrates an embodiment of a calibration process performed by the semantic analysis module. FIG. 5 is a flow chart of the first type of neural network model in the training phase. FIG. 6 is a flow chart of the second type of neural network model in the training phase. Figure 7 is a flow chart showing the third type of neural network model in the training phase.

Claims (8)

A character image recognition system for identifying a character in a file to be identified, the file to be identified includes a plurality of character images, the character image recognition system comprising: a server, comprising: a character segmentation area recognition The module identifies at least one character segmentation region, the character segmentation region includes at least one of the character image images; and a textual analysis module, communicatively connecting the character segmentation region identification module, and the textual analysis module Converting the character image in the character segmentation area into an editable character element and correcting the editable character element; a character segmentation area label module, the character segmentation area label module identifying the character element Dividing a relative position of the area in the to-be-identified file, and labeling the character division area; and an output module, outputting the editable character, a relative position of the character division area in the to-be-identified file, a label corresponding to the character segmentation area; and a client having a display screen; wherein the output module outputs the editable character to the client and displayed on the Displayed on the screen. The character image recognition system of claim 1, wherein the character segmentation area recognition module classifies the character segmentation region into a set of demand characters and a set of non-demand characters. The character image recognition system according to claim 1 or 2, wherein the character segmentation region recognition module identifies the segmentation region by a first type of neural network model. For example, the character image recognition system described in claim 3, wherein the first type of neural network model is used in the training phase, the following steps are used for training: (b1) for the file to be identified, obtain a first a quantity of image samples of the same category; (b2) dividing the image samples into a training set, a test set, and a verification set; (b3) using the training set as a training sample to generate the first type of nerve Prototype of the network model; (b4) verifying the correctness of the prototype of the first type of neural network model by using the test set, and if the verification pass rate of the test set is less than a first preset threshold, proceeding to step (b5) If the verification pass rate of the test set is not less than the first preset threshold, proceed to step (b6); (b5) obtain a second quantity of image samples of the same category, and repeat (b2)~(b4) Step (b6) verifying the correctness of the first type of neural network model by using the verification set, and if the verification pass rate of the verification set is less than a second preset threshold, proceeding to step (b7), if the test set The verification pass rate is not less than the second preset threshold, then Step (B7) acquiring image sample and then a third number of the same category, and repeating (b6) a;; the step (b8) (b8) to complete the training of the neural network model of the first type. The character image recognition system of claim 1, wherein the text analysis module corrects the editable character by a second type of neural network model. The character image recognition system according to claim 5, wherein the second type of neural network model is trained in the training stage by the following steps: (d1) for the word in the file to be identified a meta-segment region, obtaining a fourth number of editable character samples of the character segmentation region; (d2) dividing the editable character samples into a training set, a test set, and a verification set; (d3 Using the training set as a training sample to generate a prototype of the second type of neural network model; (d4) verifying the correctness of the second type of neural network model using the test set, if the verification pass rate of the test set If it is less than a fourth preset threshold, proceed to step (d5). If the verification pass rate of the test set is not less than the fourth preset threshold, proceed to step (d6); (d5) obtain a fifth number of the same The editable character sample of the category, and repeat the steps of (d2)~(d4); (d6) verify the correctness of the character extraction model by using the verification set, if the verification pass rate of the verification set is less than a fifth pre- Set the threshold, then go to step (d7), if the test set is tested If the pass rate is not less than the fifth preset threshold, proceed to step (d8); (d7) obtain a sixth number of editable character samples of the same category, and repeat the steps of (d2)~(d6); D8) Complete the training of the second type of neural network model. The character image recognition system of claim 1, wherein the character segmentation area tag module identifies the character segmentation area in the to-be-identified file by a third type of neural network model pair The relative position of the character and the label segmentation area. The character image recognition system according to claim 7, wherein the third type of neural network model is trained in the training stage by the following steps: (e1) for the word in the file to be identified a meta-segment area, obtaining a seventh number of labels of the character segmentation area and a relative position in the to-be-identified file; (e2) dividing the label samples into a training set, a test set, and a verification set (e3) using the training set as a training sample to generate a prototype of the third type of neural network model; (e4) using the test set to verify the correctness of the third type of neural network model, if the test set If the verification pass rate is less than a seventh preset threshold, proceed to step (e5). If the verification pass rate of the test set is not less than the seventh preset threshold, proceed to step (e6); (e5) obtain an eighth a quantity of label samples of the same category, and repeating the steps of (e2)~(e4); (e6) verifying the correctness of the third type of neural network model by using the verification set, if the verification pass rate of the verification set is less than one The eighth preset threshold is entered in step (e7), if If the verification pass rate of the test set is not less than the eighth preset threshold, proceed to step (e8); (e7) acquire a ninth number of image samples of the same category, and repeat the steps of (e2) to (e6); (e8) Complete the training of the third type of neural network model.