WO2020233270A1 - 票据分析方法、分析装置、计算机设备和介质 - Google Patents

票据分析方法、分析装置、计算机设备和介质 Download PDF

Info

Publication number
WO2020233270A1
WO2020233270A1 PCT/CN2020/084094 CN2020084094W WO2020233270A1 WO 2020233270 A1 WO2020233270 A1 WO 2020233270A1 CN 2020084094 W CN2020084094 W CN 2020084094W WO 2020233270 A1 WO2020233270 A1 WO 2020233270A1
Authority
WO
WIPO (PCT)
Prior art keywords
bill
image
information
text
type
Prior art date
Application number
PCT/CN2020/084094
Other languages
English (en)
French (fr)
Inventor
黄光伟
李月
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Publication of WO2020233270A1 publication Critical patent/WO2020233270A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds

Definitions

  • the present invention relates to the technical field of image processing, in particular to a bill analysis method, analysis device, computer readable storage medium and computer equipment.
  • OCR Optical Character Recognition
  • Baidu and Tencent Some companies currently recognize and archive bill images based on OCR (Optical Character Recognition) technology, such as Baidu and Tencent, but their systems still have some problems. For example: (1) A system can only identify one type of bill; (2) The recognition accuracy is low; (3) When the bill is defaced or tampered with, the bill cannot be verified for authenticity; (4) Existing The use of the bill recognition system is more complicated, and it cannot form an effective document for the financial staff to check, etc.
  • the first aspect of the present invention provides a bill analysis method, including:
  • Prompt information is presented based on the inspection result.
  • the step of classifying the received bill image through the dense convolutional network to obtain the type of the bill includes:
  • the first convolution layer is used to convolve the received picture
  • the first dense block is connected to the first convolutional layer
  • the second dense block is connected to the first pooling layer
  • the third dense block is connected to the second pooling layer
  • the bill image is input to the first convolutional layer, and the Softmax classifier outputs the type of the bill.
  • the recognizing the bill image according to the type to obtain the filing information of the bill includes:
  • the performing distortion correction on the bill image to obtain a bill correction image includes:
  • the bill correction image is segmented from the bill image through perspective transformation.
  • the performing text direction detection on the bill correction image to obtain a bill detection image includes:
  • the performing text detection and text recognition on the bill detection image, and performing text filing according to the type to obtain the filing information of the bill includes:
  • the location information use a preset text recognition network model to recognize each of the text box images to obtain the text content of the bill;
  • checking whether the bill meets the judgment standard through the archived information includes:
  • Second result the inspection result is obtained according to the first result and the second result;
  • the method further includes:
  • the judging whether the bill image meets the preset requirements for picture quality includes:
  • a non-reference image quality evaluation algorithm is used to determine whether the bill image meets the preset requirements for image quality.
  • a reference-free image quality evaluation algorithm to determine whether the bill image meets preset requirements for image quality includes:
  • the non-reference image quality evaluation algorithm adaptively adjusts the evaluation threshold in the algorithm according to the resolution of the bill image.
  • the method further includes:
  • the content extraction of the bill to obtain archive information includes:
  • the checking whether the bill meets a preset authenticity judgment standard through the archived information includes:
  • the inspection result is obtained according to the first result, the second result, and the third result.
  • the second aspect of the present invention provides a bill analysis device, including:
  • the first bill processing module is configured to classify the received bill image through a dense convolutional network to obtain the type of the bill, identify the bill according to the type to obtain the filing information of the bill, The judgment standard of the bill, and check whether the bill meets the judgment standard through the archived information;
  • the prompt module is used to present prompt information based on the inspection result.
  • the first bill processing module further includes a quality judgment module for judging whether the bill image meets the preset requirements for picture quality, and if not, prompting to re-upload the bill.
  • the classification module is used to classify the received bill according to the file type to determine whether the bill is the first bill or the second bill;
  • the second bill processing module is configured to extract content from the bill to obtain archived information, and verify whether the bill meets the preset authenticity judgment standard through the archived information;
  • the database module is used to import the archived information and inspection results into the database.
  • a third aspect of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the card identification method as described in the first aspect is implemented.
  • a fourth aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the method described in the first aspect when the program is executed .
  • Fig. 1 shows a flowchart of a bill analysis method according to an embodiment of the present invention
  • FIGS. 2a-2b show schematic diagrams of bill images according to an embodiment of the present invention
  • Figure 3 shows a flow chart of bill classification according to an embodiment of the present invention
  • Fig. 4 shows a block diagram of a dense convolutional network according to an embodiment of the present invention
  • FIG. 5 shows a flow chart of the preprocessing of the bill image of the bill according to the embodiment of the present invention
  • 6a-6b show schematic diagrams of different stages of distortion correction of a bill image according to an embodiment of the present invention
  • Figures 7a-7b show schematic diagrams of different stages of text direction detection of a bill image according to an embodiment of the present invention
  • FIG. 8 shows a flowchart of text detection, text recognition, and text filing of bill images of bills according to an embodiment of the present invention
  • Figures 9a-9b show schematic diagrams of stages of text filing of a bill according to an embodiment of the present invention.
  • FIG. 10 shows a structural block diagram of bill analysis according to an embodiment of the present invention.
  • Fig. 11 shows a schematic structural diagram of a computer device according to an embodiment of the present invention.
  • the invention relates to a bill analysis method, an analysis device, a computer equipment and a medium.
  • the received bill image is classified through a dense convolution network to be able to identify different types of bills, and at the same time, the bills are identified, archived and inspected for presentation
  • the prompt information can make up for the problems in bill processing in the prior art, effectively improve the processing efficiency of various types of bills, and has a wide range of application prospects.
  • an embodiment of the present invention provides a bill analysis method, which includes: classifying received bill images through a dense convolutional network to obtain the type of the bill; and identifying the bill according to the type Images are used to obtain the filing information of the bill; based on the judgment standard for the type of bill, whether the bill meets the judgment standard is checked through the filing information; and prompt information is presented based on the inspection result.
  • the analysis method includes:
  • the received bill image is classified by the dense convolutional network to obtain the type of the bill.
  • the ticket image can be a scanned image of a scanned taxi ticket, or a photo of a taxi ticket taken, such as .jpg, .png, .bmp and other non-editable pictures format.
  • the method further includes: determining whether the bill image meets the preset image quality requirements, and if not, prompting to upload the bill again.
  • the image quality of the bill image of the bill is detected, and if the image quality of the bill image meets the preset requirements, the bill classification is performed. As shown in Figure 2a, the bill image is blurred. If the preset requirements are not met, image recognition is not performed, the analysis process of the bill is ended, and a prompt message is displayed, prompting the user to upload the bill image for resubmitting the bill. In this way, the quality control of the uploaded bill image can be realized, the additional workload due to unrecognition can be avoided, and the processing speed of the bill can be effectively improved.
  • the judging whether the bill image meets the preset requirements for picture quality includes: using no reference
  • the image quality evaluation algorithm determines whether the bill image meets the preset requirements for image quality.
  • the character gradient-based non-reference image quality assessment (CG-DIQA) algorithm is used to determine whether the image quality of the bill image meets the picture preset requirements.
  • the non-reference image quality assessment algorithm is based on the The resolution of the bill image is adaptively adjusted to the evaluation threshold in the algorithm.
  • the non-reference image quality evaluation algorithm can adaptively adjust the evaluation threshold according to the resolution of the bill image to evaluate the quality of the bill image, that is, set the corresponding value according to the resolution of the specific bill image.
  • the evaluation threshold is used to evaluate the quality of the bill image to be evaluated.
  • the bill image After ensuring that the bill image meets the preset requirements, the bill image is processed.
  • the dense convolutional network effectively strengthens the transmission of image features by increasing the number of network model layers, improves the learning ability of image features, reduces the problem of gradient disappearance in the network, and reduces the number of parameters of the network .
  • the bill classification of bill images through the dense convolutional network can improve the accuracy of bill classification and enhance the generalization ability.
  • the classifying the received bill image through the dense convolution network to obtain the type of the bill includes: establishing a dense convolutional network model, wherein the dense convolutional network
  • the convolutional network model includes a first convolutional layer, which is used to convolve the received picture; a first dense block, which is connected to the first convolutional layer; the second convolutional layer of the first dense block, the first Batch normalization layer, first activation function layer, first pooling layer; second dense block, connected to the first pooling layer; sequentially connected to the third convolution layer of the second dense block, and second batch normalization Layer, second activation function layer, second pooling layer; third dense block, connected to the second pooling layer; sequentially connected to the third pooling layer of the third dense block; connected to the third pooling layer
  • the Softmax classifier train the dense convolutional network model; input the bill image into the first convolution layer, and the Softmax classifier outputs the type of the bill.
  • the dense convolutional network model includes three convolutional layers (Convolution), three dense blocks (dense block), and two batch-normalization layers (batch-normalization). , Two activation function layers, three pooling layers (Pooling) batch normalization layer connection, and adaptively changing normalization exponential function Softmax (DorpMax) classifier.
  • the first convolutional layer is used to convolve the received picture; the first dense block is connected to the first convolutional layer; the second convolutional layer of the first dense block is sequentially connected, and the first batch of A first layer, a first activation function layer, a first pooling layer; a second dense block connected to the first pooling layer; a third convolution layer and a second batch normalization layer connected to the second dense block in sequence , The second activation function layer, the second pooling layer; the third dense block, connecting the second pooling layer; sequentially connecting the third pooling layer of the third dense block; connecting the Softmax of the third pooling layer Classifier.
  • the dense convolutional network model requires a large amount of ticket data for training, there are very few existing public data sets.
  • a large number of bill images of various types of bills are collected as a sample library, for example, 6000 value-added tax invoices, 6000 taxi invoices, 6000 train tickets, and 6000 text images of other types of bills are collected.
  • 6000 value-added tax invoices 6000 taxi invoices
  • 6000 train tickets 6000 text images of other types of bills are collected.
  • 1000 are used as the test set to realize the training of the dense convolutional network model.
  • the bill image is input to the first convolutional layer, and the Softmax classifier outputs the type of the bill.
  • the received bill image is input into the dense convolutional network model to obtain the type of the bill.
  • the receipt image of the above taxi ticket is input into the model, and the type result obtained is "taxi ticket".
  • image recognition based on the OCR technology is performed on the bill image to obtain the filing information of the bill.
  • the recognizing the bill image according to the type to obtain the filing information of the bill includes: performing distortion correction on the bill image to obtain a bill correction image;
  • the bill correction image performs text direction detection to obtain a bill detection image; text detection and text recognition are performed on the bill detection image, and text filing is performed according to the type to obtain the filing information of the bill.
  • processing steps are as follows:
  • the first step is to perform distortion correction on the bill image to obtain a bill correction image.
  • the bill in the bill image may be in an abnormal mode, for example, the bill in the bill image is deformed or has background interference.
  • the ticket image of the taxi ticket includes a large number of background areas, and at this time, the ticket image needs to be cropped and corrected to remove background interference.
  • the performing distortion correction on the bill image to obtain the bill correction image includes: performing image binarization on the bill image; and obtaining the horizontal line set and vertical line set of the bill image through straight line detection.
  • the set of straight lines are grouped and approximately parallel lines are combined to determine the optimal boundary and fixed point of the bill; the bill correction image is segmented from the bill image through perspective transformation.
  • the horizontal line set and the vertical line set of the bill image are grouped by line detection, and the approximate parallel lines are merged to determine the optimal boundary and fixed point of the bill.
  • LSD Line Segment Detector, line segment detector
  • line detection is used to obtain the horizontal line set and the vertical line set grouping in the bill image, and the horizontal line set and the vertical line set are combined to approximate parallel lines. So as to determine the optimal boundary and vertex in the image.
  • the bill correction image is segmented from the bill image through perspective transformation.
  • the taxi ticket image is segmented from the ticket image to obtain the ticket correction image.
  • the second step is to perform text direction detection on the bill correction image to obtain a bill detection image.
  • text direction detection is used to correct the bill correction image to improve the accuracy of text detection and recognition.
  • the performing text direction detection on the bill correction image to obtain the bill detection image includes: using a preset full-angle text detection classification model to perform text direction detection on the bill correction image to obtain The bill detection image.
  • a classification model for full-angle text detection is established and trained based on the VGG16 model.
  • the full-angle includes 0-360 degrees, and the image is discretized in steps of 10 degrees to detect the text in the image.
  • Model correction is fast. That is, the bill correction image is input into the classification model, and the bill correction image is adjusted according to the angle output by the classification model, as shown in FIG. 7b is the bill detection image after correction.
  • the third step is to perform text detection and text recognition on the bill detection image, and perform text filing according to the type to obtain filing information of the bill.
  • the performing text detection and text recognition on the bill detection image, and performing text filing according to the type to obtain the filing information of the bill includes: using preset text
  • the detection model detects the bill detection image and obtains multiple text frame images of the bill and the location information corresponding to the text frame image; according to the location information, a preset text recognition network model is used to recognize each text frame Image to obtain the text content of the bill; and use keywords to file the text content according to the type to obtain the filing information of the bill.
  • the text information is a sequence
  • the sequence is composed of characters, part of characters, or multiple characters, instead of including only one independent target in traditional target detection.
  • a detection model is established, and the detection model is trained through a large number of samples.
  • the YoloV3 text detection network is used to detect the bill detection image to obtain multiple text box images.
  • the text area is detected first, and then the text line, that is, the position information of the text of the bill in the text box image is detected.
  • the YoloV3 text detection network has the characteristics of high detection accuracy and fast detection speed, and can effectively improve the detection accuracy and speed of bill analysis.
  • a preset text recognition network model is used to recognize each of the text box images to obtain the text content of the bill.
  • CRNN Convolutional Recurrent Neural Networks, Convolutional Recurrent Neural Networks
  • the text recognition network model is a CRNN text recognition network model, which combines CNN (Convolutional Neural Networks, Convolutional Neural Network) network and RNN (Recurrent Neural Network, Recurrent Neural Network) network characteristics, which can be The text frame image is scaled to a fixed length in the vertical direction. Compared with the traditional text recognition model, the character segmentation and horizontal scaling process are no longer performed, which can improve the recognition speed. At the same time, the model can also recognize character sequences of any length, and is no longer affected by characters. Length limitation. Moreover, compared with the traditional non-end-to-end text recognition network model, this model can realize end-to-end training and improve the accuracy of text recognition. It is worth noting that the training of the model is flexible, and it can train a dictionary-based text recognition network model or a non-dictionary-based text recognition network model, and the model has the characteristics of small size and fast recognition speed.
  • CNN Convolutional Neural Networks, Convolutional Neural Network
  • RNN Recurrent Neural Network
  • the text box image is recognized as a character string, that is, the text content corresponding to each text box image is obtained to obtain the text content of the bill.
  • the CRNN character recognition network model has high accuracy in the recognition results of Chinese and English, and can recognize continuous characters of any length.
  • the documents are filed for different types.
  • the description is still taking taxi tickets as an example.
  • the taxi tickets have the same layout and Layout, each taxi ticket contains the same text items. Therefore, according to the layout and layout of the taxi ticket and the text items, a text archiving template for the taxi ticket is preset, and the text archiving template is based on the taxi ticket.
  • the text items are used as keywords to extract the information of each taxi ticket, such as invoice code, ticket number, license plate number, date, time, mileage, waiting time and actual amount of text items, as shown in Figure 9b, through the text file
  • the template obtains the filing information of the taxi ticket, thereby completing the text filing of the taxi ticket.
  • verifying whether the bill meets the judgment standard through the archived information includes: if the type is a value-added tax invoice: extracting all the bills The QR code information of the value-added tax invoice is compared with the archived information to obtain the first result, and the third-party API interface for the value-added tax invoice is called to verify the authenticity of the value-added tax invoice to obtain the second result , Obtain the inspection result according to the first result and the second result; or if the type is a taxi ticket: make a judgment based on the pre-set reasonableness judgment standard through the archived information of the taxi ticket to obtain the inspection result; or if The type is a train ticket: call the third-party API interface for the train ticket to verify the authenticity of the train ticket to obtain the inspection result; or if it is classified as a ticket other than the above three types of tickets, the type does not belong to value-added Any one of tax invoice, taxi ticket or train ticket: prompt to upload the ticket again.
  • the value-added tax invoice includes two-dimensional code information
  • first extract the two-dimensional code information from the value-added tax invoice compare it with the archived information obtained based on the OCR technology, and mark the comparison result as the first result.
  • the value-added tax invoice can be checked online through the API interface provided by the National Taxation Bureau to check the authenticity of the invoice, and the online check result is recorded as the second result.
  • the judgment criterion includes two-dimensional code information comparison and online detection, and the inspection result of the value-added tax invoice is obtained according to the first result and the second result.
  • the reasonableness judgment standard includes: the boarding and boarding time should not exceed 2 hours, based on the urban area The distance within is used as the judgment threshold, and satisfaction indicates that the taxi ticket is reasonable. That is, the judgment standard is a reasonableness judgment standard, and the judgment result is used as the inspection result. It is worth noting that those skilled in the art should set the rationality judgment standard according to actual application requirements, which will not be repeated here.
  • the authenticity of the train ticket is determined by combining the filed information of the train ticket and the online verification interface. That is, the judgment standard is archived information and online detection, so as to judge the authenticity of the bill, and use the judgment result as the inspection result. For example, compare the archived information of the train ticket with the train number and other information to obtain the comparison result; and check the authenticity of the train ticket online through the API interface provided by the 12306 website, and compare the result and authenticity information As the inspection result of the train ticket.
  • the judgment standard is archived information and online detection, so as to judge the authenticity of the bill, and use the judgment result as the inspection result. For example, compare the archived information of the train ticket with the train number and other information to obtain the comparison result; and check the authenticity of the train ticket online through the API interface provided by the 12306 website, and compare the result and authenticity information As the inspection result of the train ticket.
  • the result of the bill analysis is embodied in the filing information and inspection results of the bill, and at the same time, to assist the financial staff to perform operations, the bill inspection result is presented to remind the financial staff. For example, when the inspection result of the value-added tax invoice or the train ticket is false, the prompt information is presented, and for example, when the inspection result of the taxi ticket is unreasonable, the prompt information is presented.
  • the bill analysis method further includes: importing the archived information and inspection results into a database.
  • the database makes a judgment based on the obtained inspection result. If the bill meets the judgment standard, the bill is considered to be a normal bill, otherwise the bill is considered to be a problem bill, and the inspection result is abnormal.
  • the inspection result of is displayed in a highlighted form as a reminder, for example, to remind the financial staff to pay special attention to the bill, which can effectively improve the work efficiency of the financial staff.
  • the method before the receipt classification is performed on the received receipt image through the dense convolutional network to obtain the type of the receipt, the method further includes: determining the receipt The document type of the bill is a file format or a picture format; if it is a file format, the content of the bill is extracted to obtain archive information, and whether the bill meets the preset authenticity judgment standard is checked through the archive information.
  • the classification is performed according to the file type of the received bill, and it is determined whether the file type of the bill is a file format or a picture format. If the document type of the bill is a file format, such as pdf, it is judged to be the second bill, that is, an electronic invoice; if the document type of the bill is an image format, such as jpg, png, bmp, etc., it is judged to be the first Bills are ordinary invoices. Common invoices include, but are not limited to, value-added tax invoices, taxi invoices, train tickets, and other types of invoices.
  • the electronic invoice is processed according to the following steps:
  • extract the text information of the bill for example, use the python program to extract all the text information in the electronic invoice file.
  • text archiving is performed according to the text information, for example, the various text information is archived using keywords and location information, and unnecessary information is excluded according to the location information corresponding to the text information.
  • the two-dimensional code information in the electronic invoice is extracted, compared with the aforementioned archived information, and the comparison result is taken as the first result.
  • the electronic invoice can be used to detect the authenticity of the invoice online through the API interface provided by the National Taxation Bureau, and the online detection result is recorded as the second result.
  • the electronic invoice includes an electronic signature
  • the electronic signature is checked. If it has not been tampered with, the electronic invoice passes the verification, otherwise the verification fails, and the verification result is taken as the third result.
  • the inspection result of the bill is obtained according to the first result, the second result, and the third result.
  • the archived information and inspection results of the electronic invoice are imported into the database.
  • the database makes judgments based on the obtained inspection results. If the electronic invoice is true, it is considered The electronic invoice is a normal bill, otherwise the inspection result of the electronic invoice with abnormal inspection results will be displayed in a highlighted form for reminding.
  • an embodiment of the present application also provides a bill analysis device. Since the bill analysis device provided in this embodiment of the application corresponds to the bill analysis method provided in the foregoing embodiments, Therefore, the foregoing embodiments are also applicable to the bill analysis device provided in this embodiment, and will not be described in detail in this embodiment.
  • an embodiment of the present application further provides a bill analysis device, including: a first bill processing module, configured to classify the received bill image through a dense convolutional network to obtain the type of the bill , Recognizing the bill according to the type to obtain the filing information of the bill, and verifying whether the bill meets the judgment standard through the filing information based on the judgment criteria for the bill of the type; As a result, prompt information is presented.
  • a first bill processing module configured to classify the received bill image through a dense convolutional network to obtain the type of the bill , Recognizing the bill according to the type to obtain the filing information of the bill, and verifying whether the bill meets the judgment standard through the filing information based on the judgment criteria for the bill of the type; As a result, prompt information is presented.
  • the first bill processing module further includes a quality judgment module for judging whether the bill image meets the preset image quality requirements, and if not, prompting to re-upload the bill.
  • the bill analysis device further includes a classification module, configured to classify the received bill according to the file type of the bill to determine whether the bill is the first bill or the second bill;
  • the processing module is used to extract the contents of the bill to obtain archived information, and verify whether the bill meets the preset authenticity judgment standard through the archived information;
  • the database module is used to import the archived information and the inspection result database.
  • Another embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored.
  • the program is executed by a processor, the program is implemented: the received bill image is classified through a dense convolutional network to obtain the The type of the bill; the bill image is identified according to the type to obtain the filing information of the bill; based on the judgment standard for the type of bill, whether the bill meets the judgment standard is checked by the filing information; As a result, prompt information is presented.
  • the computer-readable storage medium may adopt any combination of one or more computer-readable media.
  • the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above.
  • computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.
  • the computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and computer-readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium.
  • the computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device .
  • the program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
  • the computer program code used to perform the operations of the present invention can be written in one or more programming languages or a combination thereof.
  • the programming languages include object-oriented programming languages-such as Java, Smalltalk, C++, and also conventional Procedural programming language-such as "C" language or similar programming language.
  • the program code can be executed entirely on the user's computer, partly on the user's computer, executed as an independent software package, partly on the user's computer and partly executed on a remote computer, or entirely executed on the remote computer or server.
  • the remote computer can be connected to the user’s computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to pass Internet connection).
  • LAN local area network
  • WAN wide area network
  • Internet service provider for example, using an Internet service provider to pass Internet connection.
  • FIG. 11 a schematic structural diagram of a computer device provided by another embodiment of the present invention.
  • the computer device 12 shown in FIG. 11 is only an example, and should not bring any limitation to the function and application scope of the embodiment of the present invention.
  • the computer device 12 is represented in the form of a general-purpose computing device.
  • the components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16).
  • the bus 18 represents one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any bus structure among multiple bus structures.
  • these architectures include but are not limited to industry standard architecture (ISA) bus, microchannel architecture (MAC) bus, enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and peripheral component interconnection ( PCI) bus.
  • ISA industry standard architecture
  • MAC microchannel architecture
  • VESA Video Electronics Standards Association
  • PCI peripheral component interconnection
  • the computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.
  • the system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32.
  • the computer device 12 may further include other removable/non-removable, volatile/nonvolatile computer system storage media.
  • the storage system 34 may be used to read and write non-removable, non-volatile magnetic media (not shown in FIG. 11, usually referred to as a "hard drive").
  • a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk”) and a removable non-volatile disk (such as CD-ROM, DVD-ROM) can be provided. Or other optical media) read and write optical disc drives.
  • each drive can be connected to the bus 18 through one or more data media interfaces.
  • the memory 28 may include at least one program product having a set of (for example, at least one) program modules, which are configured to perform the functions of the embodiments of the present invention.
  • a program/utility tool 40 having a set of (at least one) program module 42 may be stored in, for example, the memory 28.
  • Such program module 42 includes but is not limited to an operating system, one or more application programs, other program modules, and program data Each of these examples or some combination may include the implementation of a network environment.
  • the program module 42 generally executes the functions and/or methods in the described embodiments of the present invention.
  • the computer device 12 can also communicate with one or more external devices 14 (such as keyboards, pointing devices, displays 24, etc.), and can also communicate with one or more devices that enable users to interact with the computer device 12, and/or communicate with Any device (such as a network card, modem, etc.) that enables the computer device 12 to communicate with one or more other computing devices. This communication can be performed through an input/output (I/O) interface 22.
  • the computer device 12 may also communicate with one or more networks (for example, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through the network adapter 20. As shown in FIG. 11, the network adapter 20 communicates with other modules of the computer device 12 through the bus 18.
  • LAN local area network
  • WAN wide area network
  • public network such as the Internet
  • the processor unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing a bill analysis method provided by an embodiment of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种票据分析方法、分析装置、计算机设备和介质，所述分析方法包括：通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型；根据所述类型识别所述票据图像以获取该票据的归档信息；基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；基于检验结果呈现提示信息。

Description

票据分析方法、分析装置、计算机设备和介质
相关申请的交叉引用
本申请要求于2019年5月20日提交的公开名称为“一种票据分析方法、分析装置、计算机设备和介质”的中国专利申请第201910417242.3号的优先权，该申请的公开通过引用被全部结合于此。
技术领域
本发明涉及图像处理技术领域，特别是涉及一种票据分析方法、分析装置、计算机可读存储介质和计算机设备。
背景技术
随着图像处理技术的发展，目前已有一些公司基于OCR(Optical Character Recognition，光学字符识别)技术对拍摄的票据图像进行识别并归档，如百度、腾讯等，但它们的系统仍存在一些问题。例如：(1)一个系统只能对一种票据进行识别；(2)识别准确率较低；(3)当票据污损或被篡改时，无法对票据进行真伪检验；(4)现有的票据识别系统使用方式较为复杂，无法形成有效的文档，供财务人员查验等等。
发明内容
本发明第一方面提供一种票据分析方法，包括：
通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型；
根据所述类型识别所述票据图像以获取该票据的归档信息；
基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；
基于检验结果呈现提示信息。
进一步的，所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型包括：
建立密集卷积网络模型，其中所述密集卷积网络模型包括
第一卷积层，用于对接收的图片进行卷积；
第一密集块，连接所述第一卷积层；
顺序连接第一密集块的第二卷积层、第一批量归一化层、第一激活函数层、第一池化层；
第二密集块，连接所述第一池化层；
顺序连接第二密集块的第三卷积层、第二批量归一化层、第二激活函数层、第二池化层；
第三密集块，连接所述第二池化层；
顺序连接第三密集块的第三池化层；
连接所述第三池化层的Softmax分类器；
训练所述密集卷积网络模型；
将所述票据图像输入第一卷积层，由所述Softmax分类器输出所述票据的类型。
进一步的，所述根据所述类型识别所述票据图像以获取该票据的归档信息包括：
对所述票据图像进行畸变校正以获取票据校正图像；
对所述票据校正图像进行文本方向检测以获取票据检测图像；
对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息。
进一步的，所述对所述票据图像进行畸变校正以获取票据校正图像包括：
对所述票据图像进行图像二值化；
通过直线检测以获取所述票据图像的水平直线集和竖直直线集分组、合并近似平行线以确定所述票据的最优边界和定点；
通过透视变换从所述票据图像内分割出所述票据校正图像。
进一步的，所述对所述票据校正图像进行文本方向检测以获取票据检测图像包括：
使用预置的全角度文本检测分类模型对所述票据校正图像进行文本方向检测以获取所述票据检测图像。
进一步的，所述对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息包括：
使用预置的文字检测模型检测所述票据检测图像并获取所述票据的多个文字框图像和所述文字框图像对应的位置信息；
根据所述位置信息，使用预置的文字识别网络模型识别各所述文字框图像以获取所述票据的文本内容；
根据所述类型使用关键字对所述文本内容进行文本归档以获取所述票据的归档信息。
进一步的，所述基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准包括：
若所述类型为增值税发票：
提取所述增值税发票的二维码信息并与所述归档信息进行比对以获取第一结果，调用针对所述增值税发票的第三方API接口验证所述增值税发票的真伪以获取第二结果，根据所述第一结果和第二结果获取检验结果；或
若所述类型为出租车票：
基于预设合理性判断标准通过所述出租车票的归档信息进行判断以获取检验结果；
或
若所述类型为火车票：
调用针对所述火车票的第三方API接口验证所述火车票的真伪以获取检验结果；或
若所述类型不属于增值税发票、出租车票和火车票中任一者：
提示重新上传票据。
进一步的，在通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：
判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
进一步的，所述判断所述票据图像是否满足图片质量预设要求包括：
使用无参考图像质量评估算法判断所述票据图像是否满足图片质量预设要求。
进一步的，所述使用无参考图像质量评估算法判断所述票据图像是否满足图片质量预设要求包括：
所述无参考图像质量评估算法根据所述票据图像的分辨率自适应调整该算法中的评估阈值。
进一步的，在所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：
判断接收的所述票据的文件类型为文件格式还是图片格式；
若是文件格式则对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准。
进一步的，所述对所述票据进行内容提取以获取归档信息包括：
提取所述票据的文本信息；
使用关键字和所述文本信息对应的位置信息对所述文本信息进行文本归档以获取所 述票据的归档信息。
进一步的，所述通过所述归档信息检验所述票据是否符合预设的真伪判断标准包括：
提取所述票据的二维码信息并与所述归档信息进行比对以获取第一结果；
调用针对所述票据的第三方API接口验证所述票据的真伪以获取第二结果；
对所述票据的电子签名进行防篡改验证以获取第三结果；
根据所述第一结果、第二结果和第三结果获取检验结果。
进一步的，将所述归档信息和检验结果导入数据库。
本发明第二方面提供一种票据分析装置，包括：
第一票据处理模块，用于通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型、根据所述类型识别所述票据以获取该票据的归档信息、基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；
提示模块，用于基于检验结果呈现提示信息。
进一步的，所述第一票据处理模块还包括质量判断模块，用于判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
进一步的，还包括
分类模块，用于根据接收的所述票据的文件类型进行分类以确定所述票据为第一票据或第二票据；
第二票据处理模块，用于对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准；
数据库模块，用于将所述归档信息和检验结果导入数据库。
本发明第三方面提供一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如第一方面所述的卡证识别方法。
本发明第四方面提供一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如第一方面所述的方法。
附图说明
下面结合附图对本发明的具体实施方式作进一步详细的说明。
图1示出本发明的实施例所述票据分析方法的流程图；
图2a-2b示出本发明的实施例所述票据图像的示意图；
图3示出本发明的实施例所述票据分类的流程图；
图4示出本发明的实施例所述密集卷积网络的框图；
图5示出本发明的实施例所述票据的票据图像预处理的流程图；
图6a-6b示出本发明的实施例所述票据图像的畸变校正的分阶段示意图；
图7a-7b示出本发明的实施例所述票据图像的文本方向检测的分阶段示意图；
图8示出本发明的实施例所述票据的票据图像的文字检测、文字识别和文本归档的流程图；
图9a-9b示出本发明的实施例所述票据的文本归档的分阶段示意图；
图10示出本发明的实施例所述票据分析的结构框图；
图11示出本发明的实施例所述的一种计算机设备的结构示意图。
具体实施方式
为了更清楚地说明本发明，下面结合优选实施例和附图对本发明做进一步的说明。附图中相似的部件以相同的附图标记进行表示。本领域技术人员应当理解，下面所具体描述的内容是说明性的而非限制性的，不应以此限制本发明的保护范围。
本发明涉及一种票据分析方法、分析装置、计算机设备和介质，并通过密集卷积网络对接收的票据图像进行分类能够识别不同类型的票据，同时对所述票据进行识别、归档和检验以呈现提示信息，能够弥补了现有技术中处理票据存在的问题，有效提高各类型票据的处理效率，具有广泛的应用前景。
如图1所示，本发明的实施例提供了一种票据分析方法，包括：通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型；根据所述类型识别所述票据图像以获取该票据的归档信息；基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；基于检验结果呈现提示信息。
在一个具体的示例中，如图1所示，以出租车票为例，所述分析方法包括：
第一、通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型。
将出租车票的票据图像上传，所述票据图像可以为扫描的出租车票的扫描图像，也可以为拍摄的出租车票的照片，例如为.jpg、.png、.bmp等不可编辑的图片格式。
考虑到所述出租车票可能存在污损情况，或者接收的出租车票的票据图像可能存在 模糊无法识别的问题，为解决上述问题，在一个可选的实施例中，在所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
即在进行票据分类之前，先对所述票据的票据图像的图像质量进行检测，若所述票据图像的图像质量满足预设要求则进行票据分类，如图2a所示，所述票据图像模糊，不符合预设要求，则不进行图像识别，结束该票据的分析流程，并显示提示信息，提示上传用户重新提交票据的票据图像。如此能够实现对上传的票据图像的质量控制，避免因无法识别而增加额外的工作量，能够有效提高所述票据的处理速度。
考虑到不同分辨率的票据图像在判断图像质量时，应该采用不同的预设要求，在一个可选的实施例中，所述判断所述票据图像是否满足图片质量预设要求包括：使用无参考图像质量评估算法判断所述票据图像是否满足图片质量预设要求。
在本实施例中，使用基于字符梯度的无参考图像质量评估(CG-DIQA)算法判断所述票据图像的图像质量是否满足图片预设要求，具体的，所述无参考图像质量评估算法根据所述票据图像的分辨率自适应调整该算法中的评估阈值。
在本实施例中，如图2b所示，所述无参考图像质量评估算法能够根据票据图像的分辨率自适应调整评估阈值以评估票据图像的质量，即根据具体的票据图像的分辨率设置对应的评估阈值，使用该评估阈值对待评估的票据图像进行质量评估。
确保所述票据图像符合预设要求后，对票据图像进行处理。相比较传统票据分类模型，密集卷积网络通过增加网络模型层数，有效加强图像特征的传递，提高对图像特征的学习能力，减轻该网络中存在的梯度消失问题，并减少该网络的参数量。基于上述特点，本实施例通过密集卷积网络对票据图像进行票据分类能够提高票据分类的准确性，增强泛化能力。
在一个可选的实施例中，如图3所示，所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型包括：建立密集卷积网络模型，其中所述密集卷积网络模型包括第一卷积层，用于对接收的图片进行卷积；第一密集块，连接所述第一卷积层；顺序连接第一密集块的第二卷积层、第一批量归一化层、第一激活函数层、第一池化层；第二密集块，连接所述第一池化层；顺序连接第二密集块的第三卷积层、第二批量归一化层、第二激活函数层、第二池化层；第三密集块，连接所述第二池化层；顺序连接第三密集块的第三池化层；连接所述第三池化层的Softmax分类器；训练所述密集卷积网络模型；将所述票据图像输入第一卷积层，由所述Softmax分类器输出所述票据 的类型。
首先，建立密集卷积网络模型。
在本实施例中，如图4所示，所述密集卷积网络模型包括三个卷积层(Convolution)、三个密集块(dense block)，两个批量归一化层(batch-normalization)、两个激活函数层、三个池化层(Pooling)批量归一化层连接，以及自适应变化的归一化指数函数Softmax(DorpMax)分类器。
具体的，第一卷积层，用于对接收的图片进行卷积；第一密集块，连接所述第一卷积层；顺序连接第一密集块的第二卷积层、第一批量归一化层、第一激活函数层、第一池化层；第二密集块，连接所述第一池化层；顺序连接第二密集块的第三卷积层、第二批量归一化层、第二激活函数层、第二池化层；第三密集块，连接所述第二池化层；顺序连接第三密集块的第三池化层；连接所述第三池化层的Softmax分类器。
然后，训练所述密集卷积网络模型。
由于所述密集卷积网络模型需要大量票据数据进行训练，而现有的公开数据集极少。在本实施例中，采集了大量各种类型的票据的票据图像作为样本库，例如采集了6000张增值税发票、6000张出租车发票，6000张火车票，以及其它类型票据的6000张文本图像，其中将每种类型的5000张票据图像作为训练集，1000张作为测试集，以实现对所述密集卷积网络模型的训练。
最后，将所述票据图像输入第一卷积层，由所述Softmax分类器输出所述票据的类型。
在本实施例中，将接收的票据图像输入所述密集卷积网络模型以获取所述票据的类型。例如将上述出租车票的票据图像输入该模型，获得的类型结果为“出租车票”。
第二、根据所述类型识别所述票据图像以获取该票据的归档信息。
当根据所述票据图像获取所述票据的类型后，对所述票据图像进行基于OCR技术的图像识别以获取所述票据的归档信息。
在一个可选的实施例中，如图5所示，所述根据所述类型识别所述票据图像以获取该票据的归档信息包括：对所述票据图像进行畸变校正以获取票据校正图像；对所述票据校正图像进行文本方向检测以获取票据检测图像；对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息。
具体的，所述处理步骤如下：
第一步，对所述票据图像进行畸变校正以获取票据校正图像。
所述票据图像中的票据可能处于非正常模式，例如票据图像中的票据存在变形情况、或者存在背景干扰等。如图6a所示，所述出租车票的票据图像中包括大量背景区域，则此时需要对票据图像进行裁剪、校正以去除背景干扰。
在本实施例中，所述对所述票据图像进行畸变校正以获取票据校正图像包括：对所述票据图像进行图像二值化；通过直线检测以获取所述票据图像的水平直线集和竖直直线集分组、合并近似平行线以确定所述票据的最优边界和定点；通过透视变换从所述票据图像内分割出所述票据校正图像。
首先，对所述票据图像进行图像二值化，将图像二值化处理。
然后，通过直线检测以获取所述票据图像的水平直线集和竖直直线集分组、合并近似平行线以确定所述票据的最优边界和定点。在本实施例中采用LSD(Line Segment Detector，线段检测器)直线检测，获取票据图像中的水平直线集和竖直直线集分组，将水平直线集和竖直直线集中近似的平行线进行合并，从而确定该图像中的最优边界和顶点。
最后，通过透视变换从所述票据图像内分割出所述票据校正图像。在本实施例中通过四点透视变化，如图6b所示，将出租车票图像从票据图像中分割出来以获得票据校正图像。
第二步，对所述票据校正图像进行文本方向检测以获取票据检测图像。
如图7a所示，考虑到所述票据校正图像中的票据有可能存在倾斜、倒置等情况，采用文本方向检测校正所述票据校正图像以提高文本检测和识别的准确性。
在一个可选的实施例中，所述对所述票据校正图像进行文本方向检测以获取票据检测图像包括：使用预置的全角度文本检测分类模型对所述票据校正图像进行文本方向检测以获取所述票据检测图像。
在本实施例中，基于VGG16模型建立并训练全角度文本检测的分类模型，所述全角度包括0-360度，按照10度为步长对图像进行离散化以检测图像中的文本，该分类模型校正速度快。即将所述票据校正图像输入该分类模型，按照该分类模型输出的角度调整所述票据校正图像，如图7b所示为校正后的票据检测图像。
第三步，对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息。
在本实施例中，如图8所示，所述对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息包括：使用预置的文字检测模 型检测所述票据检测图像并获取所述票据的多个文字框图像和所述文字框图像对应的位置信息；根据所述位置信息，使用预置的文字识别网络模型识别各所述文字框图像以获取所述票据的文本内容；根据所述类型使用关键字对所述文本内容进行文本归档以获取所述票据的归档信息。
首先，使用预置的文字检测模型检测所述票据检测图像并获取所述票据的多个文字框图像和所述文字框图像对应的位置信息。
考虑到文字信息为一个序列，所述序列由字符、字符的一部分或多字符组成，而不是传统目标检测中只包括一个独立目标。针对文字信息的特点，建立检测模型，通过采集的大量样本对所述检测模型进行训练。
在本实施例中，采用YoloV3文字检测网络对所述票据检测图像进行检测以获取多个文字框图像。先检测出文本区域，再检测文本线，即所述票据的文本在所述文字框图像中的位置信息。所述YoloV3文字检测网络相较于传统文字检测网络具有检测精度高、检测速度快等特点，能够有效提高票据分析的检测精度和速度。
然后，根据所述位置信息，使用预置的文字识别网络模型识别各所述文字框图像以获取所述票据的文本内容。
考虑到文字框图像大小不一，需要进一步通过文字识别网络对各文字框图像中的文字进行识别。为提高识别结果的准确性，在本实施例中采用CRNN(Convolutional Recurrent Neural Networks，卷积递归神经网络)文字识别网络模型，具体步骤如下：
在本实施例中，所述文字识别网络模型为CRNN文字识别网络模型，该模型结合CNN(Convolutional Neural Networks，卷积神经网络)网络和RNN(Recurrent Neural Network，递归神经网络)网络特点，能够按照垂直方向将文字框图像缩放到固定长度，相较于传统文字识别模型，不再进行字符分割和水平缩放处理，能够提高识别速度；同时该模型还能够识别任意长度的字符序列，不再受字符长度的限制。并且，相比较传统的非端到端文字识别网络模型，该模型能够实现端到端训练，能够提高文字识别的准确性。值得说明的是，该模型的训练灵活，能够训练基于词典的文字识别网络模型或不基于词典的文字识别网络模型，并且该模型具有体积小、识别速度快的特点。
将通过文字检测获取的多个文字框图像输入到所述CRNN文字识别网络模型中，例如按照CNN(卷积神经网络)-LSTM(长短期记忆网络)-CTC(联结机制时间分类)的流程获取识别结果，将文字框图像识别为字符串，即获取各文字框图像对应的文本内容以获取所述票据的文本内容。所述CRNN文字识别网络模型对中英文的识别结果准确 率高，能够识别任意长度的连续文字。
最后，根据所述类型使用关键字对所述文本内容进行文本归档以获取所述票据的归档信息。
在已经获得的所述类型的基础上，针对不同的类型对所述票据进行文本归档处理，仍以出租车票为例进行说明，如图9a所示，所述出租车票具有相同的排版和布局，每张出租车票包含的文本项目相同，因此根据所述出租车票的排版和布局，以及文本项目预先设置用于出租车票的文字归档模板，所述文字归档模板以出租车票的文本项目作为关键字提取各出租车票的信息，例如发票代码、车票号码、车牌号、日期、时间、里程、等待时间和实收金额等文本项目，如图9b所示，通过所述文字归档模板获取出租车票的归档信息，从而完成针对该出租车票的文字归档。
第三、基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准。
结合实际应用中通常使用票据进行报账的问题，检验所述票据是否符合判断标准成为票据分析所必须考虑的问题。不同类型的票据可以有不同的判断标准，因此票据的检验可以基于票据所属类型的判断标准进行。
在一个可选的实施例中，所述基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准包括：若所述类型为增值税发票：提取所述增值税发票的二维码信息并与所述归档信息进行比对以获取第一结果，调用针对所述增值税发票的第三方API接口验证所述增值税发票的真伪以获取第二结果，根据所述第一结果和第二结果获取检验结果；或若所述类型为出租车票：基于预设合理性判断标准通过所述出租车票的归档信息进行判断以获取检验结果；或若所述类型为火车票：调用针对所述火车票的第三方API接口验证所述火车票的真伪以获取检验结果；或若分类为上述三类票据之外的票据即所述类型不属于增值税发票、出租车票和火车票中任一者：提示重新上传票据。
在本实施例中，当所述类型为增值税发票：
考虑到所述增值税发票包括二维码信息，先从增值税发票中提取二维码信息，与前述基于OCR技术获得的归档信息进行比对，将比对结果标记为第一结果。
考虑到所述增值税发票可以通过国税局提供的API接口在线检测该发票的真伪，并将在线检测结果记为第二结果。
即所述判断标准包括二维码信息比对和在线检测，则根据所述第一结果和第二结果 获得所述增值税发票的检验结果。
在本实施例中，当所述类型为出租车票：
根据预设置的合理性判断标准判断所述出租车票的是否合理，从而辨别该出租车票的合理性，例如所述合理性判断标准包括：所述上下车时间不得超过2个小时，以城区内的距离作为判断阈值，满足则表明所述出租车票合理。即所述判断标准为合理性判断标准，将该判断结果作为检验结果。值得说明的是，本领域技术人员应当根据实际应用需求设置合理性判断标准，在此不再赘述。
在本实施例中，当所述类型为火车票，结合火车票的归档信息和在线验证接口判断所述火车票的真伪。即所述判断标准为归档信息和在线检测，从而判断所述票据的真伪，并将判断结果作为所述检验结果。例如通过火车票的归档信息与该火车票的车次等信息进行比对以获得比对结果；以及通过12306网站提供的API接口在线检测该火车票的真伪，并将比对结果和真伪信息作为所述火车票的检验结果。
第四、基于检验结果呈现提示信息。
在本实施例中，票据分析的结果体现在票据的归档信息和检验结果，同时，为辅助财务人员进行操作，将所述票据检验结果呈现出来以提示财务人员。例如当增值税发票或火车票的检验结果为假时呈现提示信息，又例如当出租车票的检验结果为不合理时呈现提示信息。
为了方便随时调用各票据的分析结果，在一个可选的实施例中，所述票据分析方法还包括：将所述归档信息和检验结果导入数据库。
例如，将前述获得的所述票据的归档信息和检验结果导入Excel，并存储在数据库中。在本实施例中，所述数据库根据获得的检验结果进行判断，若所述票据符合判断标准则认为所述票据为正常票据，否则认为所述票据为问题票据，将检验结果存在异常的问题票据的检验结果以高亮的形式显示出来用以提示，例如提示财务人员需要特别关注该票据，能够有效提高财务人员的工作效率。
考虑到电子发票的广泛应用，在一个可选的实施例中，在所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：判断接收的所述票据的文件类型为文件格式还是图片格式；若是文件格式则对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准。
具体的，根据接收的票据的文件类型进行划分，判断所述票据的文件类型为文件格式还是图片格式。若所述票据的文件类型为文件格式，如pdf等，则判断为第二票据， 即电子发票；若所述票据的文件类型为图像格式，如jpg,png,bmp等，则判断为第一票据，即普通发票。普通发票例如包括但不限于增值税发票、出租车发票、火车票以及其他类票据。
在本实施例中，对电子发票按照如下步骤进行处理：
首先，提取所述票据的文本信息，例如使用python程序提取电子发票文件中的所有文本信息。
其次，使用关键字和所述文本信息对应的位置信息对所述文本信息进行文本归档以获取所述票据的归档信息。即根据所述文本信息进行文本归档，例如使用关键字和位置信息对所述各文本信息进行归档，同时根据所述文本信息对应的位置信息排除不必要的信息。
最后，通过所述归档信息检验所述票据是否符合预设的真伪判断标准。
具体的，提取所述票据的二维码信息并与所述归档信息进行比对并获取第一结果；调用针对所述票据的第三方API接口验证所述票据的真伪以获取第二结果；对所述票据的电子签名进行防篡改验证以获取第三结果；根据所述第一结果、第二结果和第三结果获取检验结果。
考虑到所述电子发票中具有二维码信息，则提取电子发票中的二维码信息，与前述归档信息进行比对，并将比对结果作为第一结果。
考虑到所述电子发票可以通过国税局提供的API接口在线检测该发票的真伪，并将在线检测结果记为第二结果。
考虑到所述电子发票中包括电子签名，查验所述电子签名，如未被篡改则所述电子发票通过验证，否则检验不通过，将检验结果作为第三结果。
根据所述第一结果、第二结果和第三结果获取所述票据的检验结果。
值得说明的是，为了方便随时调用各票据的分析结果，将所述电子发票的归档信息和检验结果导入数据库，所述数据库根据获得的检验结果进行判断，若所述电子发票为真则认为所述电子发票为正常票据，否则将检验结果存在异常的电子发票的检验结果以高亮的形式显示出来用以提示。
与上述实施例提供的票据分析方法相对应，本申请的一个实施例还提供一种票据分析装置，由于本申请实施例提供的票据分析装置与上述几种实施例提供的票据分析方法相对应，因此在前述实施方式也适用于本实施例提供的票据分析装置，在本实施例中不再详细描述。
如图10所示，本申请的一个实施例还提供一种票据分析装置，包括：第一票据处理模块，用于通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型、根据所述类型识别所述票据以获取该票据的归档信息、基于针对所述类型的票据的判断标准通过所述归档信息检验所述票据是否符合所述判断标准；提示模块，用于基于检验结果呈现提示信息。
在一个可选的实施例中，所述第一票据处理模块还包括质量判断模块，用于判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
在另一个可选的实施例中，所述票据分析装置还包括分类模块，用于根据接收的所述票据的文件类型进行分类以确定所述票据为第一票据或第二票据；第二票据处理模块，用于对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准；数据库模块，用于将所述归档信息和检验结果导入数据库。
本发明的另一个实施例提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现：通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型；根据所述类型识别所述票据图像以获取该票据的归档信息；基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；基于检验结果呈现提示信息。
在实际应用中，所述计算机可读存储介质可以采用一个或多个计算机可读的介质的任意组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本实时例中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于无线、电线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言或其组合来编写用于执行本发明操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。
如图11所示，本发明的另一个实施例提供的一种计算机设备的结构示意图。图11显示的计算机设备12仅仅是一个示例，不应对本发明实施例的功能和使用范围带来任何限制。
如图11所示，计算机设备12以通用计算设备的形式表现。计算机设备12的组件可以包括但不限于：一个或者多个处理器或者处理单元16，系统存储器28，连接不同系统组件(包括系统存储器28和处理单元16)的总线18。
总线18表示几类总线结构中的一种或多种，包括存储器总线或者存储器控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构包括但不限于工业标准体系结构(ISA)总线，微通道体系结构(MAC)总线，增强型ISA总线、视频电子标准协会(VESA)局域总线以及外围组件互连(PCI)总线。
计算机设备12典型地包括多种计算机系统可读介质。这些介质可以是任何能够被计算机设备12访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。
系统存储器28可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(RAM)30和/或高速缓存存储器32。计算机设备12可以进一步包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储系统34可以用于读写不可移动的、非易失性磁介质(图11未显示，通常称为“硬盘驱动器”)。尽管图11中未示出，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘(例如CD-ROM，DVD-ROM或者其它光介质)读写的光盘驱动器。 在这些情况下，每个驱动器可以通过一个或者多个数据介质接口与总线18相连。存储器28可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程序模块被配置以执行本发明各实施例的功能。
具有一组(至少一个)程序模块42的程序/实用工具40，可以存储在例如存储器28中，这样的程序模块42包括但不限于操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块42通常执行本发明所描述的实施例中的功能和/或方法。
计算机设备12也可以与一个或多个外部设备14(例如键盘、指向设备、显示器24等)通信，还可与一个或者多个使得用户能与该计算机设备12交互的设备通信，和/或与使得该计算机设备12能与一个或多个其它计算设备进行通信的任何设备(例如网卡，调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口22进行。并且，计算机设备12还可以通过网络适配器20与一个或者多个网络(例如局域网(LAN)，广域网(WAN)和/或公共网络，例如因特网)通信。如图11所示，网络适配器20通过总线18与计算机设备12的其它模块通信。应当明白，尽管图11中未示出，可以结合计算机设备12使用其它硬件和/或软件模块，包括但不限于：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储系统等。
处理器单元16通过运行存储在系统存储器28中的程序，从而执行各种功能应用以及数据处理，例如实现本发明实施例所提供的一种票据分析方法。
需要说明的是，本发明实施例提供的视线追踪方法步骤的先后顺序可以进行适当谓整，步骤也可以根据情况进行相应增减，任何熟悉本技术领域的技术人员在本发费揭露的技术范围内，可轻易程到变化的方法，都应涵盖在本发明的保护范围之内，因此不再赘述。
显然，本发明的上述实施例仅仅是为清楚地说明本发明所作的举例，而并非是对本发明的实施方式的限定，对于所属领域的普通技术人员来说，在上述说明的基础上还可以做出其它不同形式的变化或变动，这里无法对所有的实施方式予以穷举，凡是属于本发明的技术方案所引伸出的显而易见的变化或变动仍处于本发明的保护范围之列。

Claims (19)

  1. 一种票据分析方法，包括：
    通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型；
    根据所述类型识别所述票据图像以获取该票据的归档信息；
    基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；
    基于检验结果呈现提示信息。
  2. 根据权利要求1所述的票据分析方法，其中，所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型包括：
    建立密集卷积网络模型，其中所述密集卷积网络模型包括
    第一卷积层，用于对接收的图片进行卷积；
    第一密集块，连接所述第一卷积层；
    顺序连接第一密集块的第二卷积层、第一批量归一化层、第一激活函数层、第一池化层；
    第二密集块，连接所述第一池化层；
    顺序连接第二密集块的第三卷积层、第二批量归一化层、第二激活函数层、第二池化层；
    第三密集块，连接所述第二池化层；
    顺序连接第三密集块的第三池化层；
    连接所述第三池化层的Softmax分类器；
    训练所述密集卷积网络模型；
    将所述票据图像输入第一卷积层，由所述Softmax分类器输出所述票据的类型。
  3. 根据权利要求1所述的票据分析方法，其中，所述根据所述类型识别所述票据图像以获取该票据的归档信息包括：
    对所述票据图像进行畸变校正以获取票据校正图像；
    对所述票据校正图像进行文本方向检测以获取票据检测图像；
    对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获 取所述票据的归档信息。
  4. 根据权利要求3所述的票据分析方法，其中，所述对所述票据图像进行畸变校正以获取票据校正图像包括：
    对所述票据图像进行图像二值化；
    通过直线检测以获取所述票据图像的水平直线集和竖直直线集分组、合并近似平行线以确定所述票据的最优边界和定点；
    通过透视变换从所述票据图像内分割出所述票据校正图像。
  5. 根据权利要求3所述的票据分析方法，其中，所述对所述票据校正图像进行文本方向检测以获取票据检测图像包括：
    使用预置的全角度文本检测分类模型对所述票据校正图像进行文本方向检测以获取所述票据检测图像。
  6. 根据权利要求3所述的票据分析方法，其中，所述对所述票据检测图像进行文字检测和文字识别，并根据所述类型进行文本归档以获取所述票据的归档信息包括：
    使用预置的文字检测模型检测所述票据检测图像并获取所述票据的多个文字框图像和所述文字框图像对应的位置信息；
    根据所述位置信息，使用预置的文字识别网络模型识别各所述文字框图像以获取所述票据的文本内容；
    根据所述类型使用关键字对所述文本内容进行文本归档以获取所述票据的归档信息。
  7. 根据权利要求1所述的票据分析方法，其中，所述基于针对所述类型的票据的判断标准，根据所述类型通过所述归档信息检验所述票据是否符合所述判断标准包括：
    若所述类型为增值税发票：
    提取所述增值税发票的二维码信息并与所述归档信息进行比对以获取第一结果，
    调用针对所述增值税发票的第三方API接口验证所述增值税发票的真伪以获取第二结果，
    根据所述第一结果和第二结果获取检验结果；或
    若所述类型为出租车票：
    基于预设合理性判断标准通过所述出租车票的归档信息进行判断以获取检验结果；或
    若所述类型为火车票：
    调用针对所述火车票的第三方API接口验证所述火车票的真伪以获取检验结果；或
    若所述类型不属于增值税发票、出租车票和火车票中任一者：
    提示重新上传票据。
  8. 根据权利要求1所述的票据分析方法，其中，在通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：
    判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
  9. 根据权利要求8所述的票据分析方法，其中，所述判断所述票据图像是否满足图片质量预设要求包括：
    使用无参考图像质量评估算法判断所述票据图像是否满足图片质量预设要求。
  10. 根据权利要求9所述的票据分析方法，其中，所述使用无参考图像质量评估算法判断所述票据图像是否满足图片质量预设要求包括：
    所述无参考图像质量评估算法根据所述票据图像的分辨率自适应调整该算法中的评估阈值。
  11. 根据权利要求1所述的票据分析方法，其中，在所述通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型之前，所述方法还包括：
    判断接收的所述票据的文件类型为文件格式还是图片格式；
    若是文件格式则对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准。
  12. 根据权利要求11所述的票据分析方法，其中，所述对所述票据进行内容提取以获取归档信息包括：
    提取所述票据的文本信息；
    使用关键字和所述文本信息对应的位置信息对所述文本信息进行文本归档以获取所述票据的归档信息。
  13. 根据权利要求11所述的票据分析方法，其中，所述通过所述归档信息检验所述票据是否符合预设的真伪判断标准包括：
    提取所述票据的二维码信息并与所述归档信息进行比对以获取第一结果；
    调用针对所述票据的第三方API接口验证所述票据的真伪以获取第二结果；
    对所述票据的电子签名进行防篡改验证以获取第三结果；
    根据所述第一结果、第二结果和第三结果获取检验结果。
  14. 根据权利要求1-13中任一项所述的票据分析方法，其中，将所述归档信息和检验结果导入数据库。
  15. 一种票据分析装置，，包括：
    第一票据处理模块，用于通过密集卷积网络对接收的票据图像进行票据分类以获取所述票据的类型、根据所述类型识别所述票据以获取该票据的归档信息、基于针对所述类型的票据的判断标准，通过所述归档信息检验所述票据是否符合所述判断标准；
    提示模块，用于基于检验结果呈现提示信息。
  16. 根据权利要求15所述的票据分析装置，其中，所述第一票据处理模块还包括质量判断模块，用于判断所述票据图像是否满足图片质量预设要求，若不满足则提示重新上传票据。
  17. 根据权利要求15所述的票据分析装置，其中，还包括
    分类模块，用于根据接收的所述票据的文件类型进行分类以确定所述票据为第一票据或第二票据；
    第二票据处理模块，用于对所述票据进行内容提取以获取归档信息，通过所述归档信息检验所述票据是否符合预设的真伪判断标准；
    数据库模块，用于将所述归档信息和检验结果导入数据库。
  18. 一种计算机可读存储介质，其上存储有计算机程序，其中，该程序被处理器执行时实现如权利要求1-14中任一项所述的票据分析方法。
  19. 一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现如权利要求1-14中任一所述的票据分析方法。
PCT/CN2020/084094 2019-05-20 2020-04-10 票据分析方法、分析装置、计算机设备和介质 WO2020233270A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910417242.3 2019-05-20
CN201910417242.3A CN111178345A (zh) 2019-05-20 2019-05-20 一种票据分析方法、分析装置、计算机设备和介质

Publications (1)

Publication Number Publication Date
WO2020233270A1 true WO2020233270A1 (zh) 2020-11-26

Family

ID=70651855

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/084094 WO2020233270A1 (zh) 2019-05-20 2020-04-10 票据分析方法、分析装置、计算机设备和介质

Country Status (2)

Country Link
CN (1) CN111178345A (zh)
WO (1) WO2020233270A1 (zh)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112464892A (zh) * 2020-12-14 2021-03-09 招商局金融科技有限公司 票据区域识别方法、装置、电子设备及可读存储介质
CN112560754A (zh) * 2020-12-23 2021-03-26 北京百度网讯科技有限公司 票据信息的获取方法、装置、设备及存储介质
CN112597773A (zh) * 2020-12-08 2021-04-02 上海深杳智能科技有限公司 文档结构化方法、系统、终端及介质
CN112613367A (zh) * 2020-12-14 2021-04-06 盈科票据服务(深圳)有限公司 票据信息文本框获取方法、系统、设备及存储介质
CN113344889A (zh) * 2021-06-18 2021-09-03 广东电网有限责任公司 一种财务票据图像质量评价方法及装置
CN113409278A (zh) * 2021-06-22 2021-09-17 平安健康保险股份有限公司 图像质量检测方法、装置、设备及介质
CN113554009A (zh) * 2021-09-22 2021-10-26 北京奇虎科技有限公司 票据核验方法、装置、设备及存储介质
CN113627395A (zh) * 2021-09-17 2021-11-09 平安银行股份有限公司 文本识别方法、装置、介质及电子设备
CN113657334A (zh) * 2021-08-25 2021-11-16 展讯通信(天津)有限公司 图片识别方法和装置
CN114240407A (zh) * 2021-11-17 2022-03-25 广东电网有限责任公司 一种基于区块链的票据风险传导量化评估系统及方法
CN114419651A (zh) * 2022-03-03 2022-04-29 深圳依时货拉拉科技有限公司 智能票据识别方法、计算机可读存储介质及计算机设备
CN117727059A (zh) * 2024-02-18 2024-03-19 蓝色火焰科技成都有限公司 汽车金融发票信息核验方法、装置、电子设备及存储介质

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112053343A (zh) * 2020-09-02 2020-12-08 平安科技（深圳）有限公司 用户图片数据的处理方法、装置、计算机设备及存储介质
CN112183594A (zh) * 2020-09-17 2021-01-05 微民保险代理有限公司 票据图像的处理方法和装置、存储介质及电子设备
CN112465525A (zh) * 2020-12-02 2021-03-09 无锡朝阳供应链科技股份有限公司 一种农贸市场商品进货溯源方法、装置及系统
CN112364837A (zh) * 2020-12-09 2021-02-12 四川长虹电器股份有限公司 一种基于目标检测和文本识别的票据信息识别方法
CN112686238B (zh) * 2020-12-21 2023-07-21 福建新大陆软件工程有限公司 一种基于深度学习的航运单识别方法
CN112818823B (zh) * 2021-01-28 2024-04-12 金科览智科技(北京)有限公司 一种基于票据内容和位置信息的文本抽取方法
CN113158895B (zh) * 2021-04-20 2023-11-14 北京中科江南信息技术股份有限公司 票据识别方法、装置、电子设备及存储介质
CN113326895A (zh) * 2021-06-25 2021-08-31 湖南星汉数智科技有限公司 客票行程单识别方法、装置、计算机设备和存储介质
CN113657162A (zh) * 2021-07-15 2021-11-16 福建新大陆软件工程有限公司 一种基于深度学习的票据ocr识别方法
CN117392419A (zh) * 2023-12-12 2024-01-12 武汉大学 一种基于深度学习的药品图片相似度比对方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012252657A (ja) * 2011-06-06 2012-12-20 Toshiba Corp 帳票識別システム、帳票識別方法、および帳票処理プログラム
US20180268448A1 (en) * 2015-10-07 2018-09-20 Way2Vat Ltd. System and methods of an expense management system based upon business document analysis
CN108717545A (zh) * 2018-05-18 2018-10-30 北京大账房网络科技股份有限公司 一种基于手机拍照的票据识别方法及系统
CN108734850A (zh) * 2018-04-27 2018-11-02 深圳怡化电脑股份有限公司 纸币识别方法、纸币识别装置及终端设备

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107292823A (zh) * 2017-08-20 2017-10-24 平安科技（深圳）有限公司 电子装置、发票分类的方法及计算机可读存储介质
CN109034206A (zh) * 2018-06-29 2018-12-18 泰康保险集团股份有限公司 图像分类识别方法、装置、电子设备及计算机可读介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012252657A (ja) * 2011-06-06 2012-12-20 Toshiba Corp 帳票識別システム、帳票識別方法、および帳票処理プログラム
US20180268448A1 (en) * 2015-10-07 2018-09-20 Way2Vat Ltd. System and methods of an expense management system based upon business document analysis
CN108734850A (zh) * 2018-04-27 2018-11-02 深圳怡化电脑股份有限公司 纸币识别方法、纸币识别装置及终端设备
CN108717545A (zh) * 2018-05-18 2018-10-30 北京大账房网络科技股份有限公司 一种基于手机拍照的票据识别方法及系统

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112597773B (zh) * 2020-12-08 2022-12-13 上海深杳智能科技有限公司 文档结构化方法、系统、终端及介质
CN112597773A (zh) * 2020-12-08 2021-04-02 上海深杳智能科技有限公司 文档结构化方法、系统、终端及介质
CN112613367A (zh) * 2020-12-14 2021-04-06 盈科票据服务(深圳)有限公司 票据信息文本框获取方法、系统、设备及存储介质
CN112464892A (zh) * 2020-12-14 2021-03-09 招商局金融科技有限公司 票据区域识别方法、装置、电子设备及可读存储介质
CN112464892B (zh) * 2020-12-14 2024-02-13 招商局金融科技有限公司 票据区域识别方法、装置、电子设备及可读存储介质
CN112560754A (zh) * 2020-12-23 2021-03-26 北京百度网讯科技有限公司 票据信息的获取方法、装置、设备及存储介质
CN113344889A (zh) * 2021-06-18 2021-09-03 广东电网有限责任公司 一种财务票据图像质量评价方法及装置
CN113409278A (zh) * 2021-06-22 2021-09-17 平安健康保险股份有限公司 图像质量检测方法、装置、设备及介质
CN113409278B (zh) * 2021-06-22 2024-04-26 平安健康保险股份有限公司 图像质量检测方法、装置、设备及介质
CN113657334A (zh) * 2021-08-25 2021-11-16 展讯通信(天津)有限公司 图片识别方法和装置
CN113627395A (zh) * 2021-09-17 2021-11-09 平安银行股份有限公司 文本识别方法、装置、介质及电子设备
CN113627395B (zh) * 2021-09-17 2023-11-17 平安银行股份有限公司 文本识别方法、装置、介质及电子设备
CN113554009B (zh) * 2021-09-22 2022-03-22 北京奇虎科技有限公司 票据核验方法、装置、设备及存储介质
CN113554009A (zh) * 2021-09-22 2021-10-26 北京奇虎科技有限公司 票据核验方法、装置、设备及存储介质
CN114240407A (zh) * 2021-11-17 2022-03-25 广东电网有限责任公司 一种基于区块链的票据风险传导量化评估系统及方法
CN114419651A (zh) * 2022-03-03 2022-04-29 深圳依时货拉拉科技有限公司 智能票据识别方法、计算机可读存储介质及计算机设备
CN117727059A (zh) * 2024-02-18 2024-03-19 蓝色火焰科技成都有限公司 汽车金融发票信息核验方法、装置、电子设备及存储介质
CN117727059B (zh) * 2024-02-18 2024-05-03 蓝色火焰科技成都有限公司 汽车金融发票信息核验方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN111178345A (zh) 2020-05-19

Similar Documents

Publication Publication Date Title
WO2020233270A1 (zh) 票据分析方法、分析装置、计算机设备和介质
US11126892B2 (en) Dual stage neural network pipeline systems and methods
CN109902622B (zh) 一种用于登机牌信息验证的文字检测识别方法
US8867828B2 (en) Text region detection system and method
US11893765B2 (en) Method and apparatus for recognizing imaged information-bearing medium, computer device and medium
US11816883B2 (en) Region proposal networks for automated bounding box detection and text segmentation
CN112149663A (zh) 结合rpa和ai的图像文字的提取方法、装置及电子设备
CN114140649A (zh) 票据分类方法、票据分类装置、电子设备和存储介质
Arslan End to end invoice processing application based on key fields extraction
CN114663871A (zh) 图像识别方法、训练方法、装置、系统及存储介质
CN111178254A (zh) 一种签名识别方法及设备
WO2022103564A1 (en) Fraud detection via automated handwriting clustering
CN111914706B (zh) 一种文字检测输出结果质量检测和控制的方法与装置
CN111598099A (zh) 图像文本识别性能的测试方法、装置、测试设备及介质
US20230048495A1 (en) Method and platform of generating document, electronic device and storage medium
CN111008635A (zh) 一种基于ocr的多票据自动识别方法及识别系统
CN114663899A (zh) 金融票据的处理方法、装置、设备及介质
WO2019071476A1 (zh) 一种基于智能终端的快递信息录入方法及录入系统
CN114120305A (zh) 文本分类模型的训练方法、文本内容的识别方法及装置
Girinath et al. Automatic Number Plate Detection using Deep Learning
WO2020172767A1 (zh) 电子签购单识别方法、装置及终端设备
US11869260B1 (en) Extracting structured data from an image
CN116109914B (zh) 银行流水图像的鉴别真伪方法、装置、电子设备及介质
CN112464892B (zh) 票据区域识别方法、装置、电子设备及可读存储介质
CN111881778B (zh) 文本检测的方法、装置、设备和计算机可读介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20810163

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20810163

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20810163

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 25.07.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20810163

Country of ref document: EP

Kind code of ref document: A1