CN107798327A - Character identifying method and device - Google Patents

Character identifying method and device Download PDF

Info

Publication number
CN107798327A
CN107798327A CN201711046515.5A CN201711046515A CN107798327A CN 107798327 A CN107798327 A CN 107798327A CN 201711046515 A CN201711046515 A CN 201711046515A CN 107798327 A CN107798327 A CN 107798327A
Authority
CN
China
Prior art keywords
character
identified
image
recognition
characteristic sequence
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN201711046515.5A
Other languages
Chinese (zh)
Inventor
杨松
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Xiaomi Mobile Software Co Ltd
Original Assignee
Beijing Xiaomi Mobile Software Co Ltd
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 Beijing Xiaomi Mobile Software Co Ltd filed Critical Beijing Xiaomi Mobile Software Co Ltd
Priority to CN201711046515.5A priority Critical patent/CN107798327A/en
Publication of CN107798327A publication Critical patent/CN107798327A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/14—Image acquisition
    • G06V30/148—Segmentation of character regions
    • G06V30/153—Segmentation of character regions using recognition of characters or words
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/14—Image acquisition
    • G06V30/146—Aligning or centring of the image pick-up or image-field
    • G06V30/1475—Inclination or skew detection or correction of characters or of image to be recognised
    • G06V30/1478—Inclination or skew detection or correction of characters or of image to be recognised of characters or characters lines

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Character Discrimination (AREA)

Abstract

The disclosure is directed to a kind of character identifying method and device, this method includes：Feature extraction is carried out to character zone image to be identified using the first convolutional neural networks, obtains characteristic pattern matrix；The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；A line characteristic sequence is identified using Recognition with Recurrent Neural Network, obtains the character in the character zone image to be identified.The technical scheme can be effectively between removal process cumulative errors, improve the accuracy rate of character recognition.

Description

Character identifying method and device
Technical field
This disclosure relates to technical field of image processing, more particularly to character identifying method and device.
Background technology
Character recognition is a key areas in computer vision, and character recognition process generally comprises two steps：Word Symbol detection and character recognition.Wherein, character machining is exactly the region for detecting to occur in image character, and conventional method is based on The method of sliding window and the method based on connected domain.Character recognition is exactly to detecting that the region of character is identified, obtaining Corresponding character.
The content of the invention
The embodiment of the present disclosure provides a kind of character identifying method and device.The technical scheme is as follows：
According to the first aspect of the embodiment of the present disclosure, there is provided a kind of character identifying method, including：
Feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtains characteristic pattern matrix；
The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtains the character zone image to be identified In character.
In one embodiment, methods described also includes：
The character zone image to be identified is corrected, obtains the neat correction chart picture of character；
It is described that feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtain characteristic pattern square Battle array, including：
Feature extraction is carried out to the correction chart picture by first convolutional neural networks, obtains characteristic pattern matrix.
In one embodiment, it is described that the character zone image to be identified is corrected, obtain the neat school of character Positive image, including：
The length of the character zone image to be identified and height are scaled in proportion, obtain zoomed image, wherein, The height of the zoomed image is preset height；
Enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain key point；
According to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline function TPS；
TPS conversion is carried out to the zoomed image according to the thin plate spline function TPS, obtains the correction chart picture.
In one embodiment, the Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and the 2nd LSTM nets Network, it is described that a line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtain the character zone image to be identified In character, including：
The characteristic sequence is encoded using the first LSTM networks, the characteristic sequence after being encoded；
The characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains the character zone to be identified Character in image.
In one embodiment, methods described also includes：
The sample of training is obtained, the sample includes the image of character region；
Carry out end-to-end training to overall network using the sample, the overall network trained, it is described train it is total Network includes first convolutional neural networks and the Recognition with Recurrent Neural Network.
According to the second aspect of the embodiment of the present disclosure, there is provided a kind of character recognition device, including：
Extraction module, for carrying out feature extraction to character zone image to be identified using the first convolutional neural networks, obtain To characteristic pattern matrix；
Cutting module, for the characteristic pattern matrix to be carried out into cutting, obtain a line characteristic sequence；
Identification module, for a line characteristic sequence to be identified by Recognition with Recurrent Neural Network, obtain described waiting to know Character in malapropism symbol area image.
In one embodiment, described device also includes：
Correction module, for being corrected to the character zone image to be identified, obtain the neat correction chart picture of character；
The extraction module includes：
Extracting sub-module, for carrying out feature extraction to the correction chart picture by first convolutional neural networks, obtain To characteristic pattern matrix.
In one embodiment, the correction module includes：
Submodule is scaled, for the length of the character zone image to be identified and height to be scaled in proportion, is obtained To zoomed image, wherein, the height of the zoomed image is preset height；
Submodule is positioned, for entering the positioning of line character key point to the zoomed image by the second convolutional neural networks, Obtain key point；
Determination sub-module, for the corresponding relation according to the key point and predeterminated position point, it is determined that corresponding thin plate sample Bar function TPS；
Transformation submodule, for carrying out TPS conversion to the zoomed image according to the thin plate spline function TPS, obtain The correction chart picture.
In one embodiment, the Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and the 2nd LSTM nets Network, the identification module include：
Encoding submodule, for being encoded using the first LSTM networks to the characteristic sequence, the spy after being encoded Levy sequence；
Decoding sub-module, for being decoded using the 2nd LSTM networks to the characteristic sequence after the coding, obtain institute State the character in character zone image to be identified.
In one embodiment, described device also includes：
Acquisition module, for obtaining the sample of training, the sample includes the image of character region；
Training module, for carrying out end-to-end training, the overall network trained, institute to overall network using the sample Stating the overall network trained includes first convolutional neural networks and the Recognition with Recurrent Neural Network.
According to the third aspect of the embodiment of the present disclosure, there is provided a kind of character recognition device, including：
Processor；
For storing the memory of processor-executable instruction；
Wherein, the processor is configured as：
Feature extraction is carried out to character zone image to be identified using the first convolutional neural networks, obtains characteristic pattern matrix；
The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtains the character zone image to be identified In character.
According to the fourth aspect of the embodiment of the present disclosure, there is provided a kind of computer-readable recording medium, be stored with computer and refer to Order, the computer instruction realize the step in the above method when being executed by processor.
The present embodiment can use the first convolutional neural networks to carry out feature extraction to character zone image to be identified, obtain Characteristic pattern matrix；The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；Using Recognition with Recurrent Neural Network to described one Row characteristic sequence is identified, and obtains the character in the character zone image to be identified, in this way, the first convolutional Neural used Network and Recognition with Recurrent Neural Network can all be led, including the overall network of the first convolutional neural networks and Recognition with Recurrent Neural Network can enter Row is trained end to end, there is can when so carrying out character recognition using first convolutional neural networks and Recognition with Recurrent Neural Network Effect eliminates the cumulative errors between step, and by carrying out characteristic sequence knowledge after image is converted into a characteristic sequence Not, it can be avoided and single word cut, the identification mistake for preventing miscut from occurring, further improve character recognition Accuracy rate.
It should be appreciated that the general description and following detailed description of the above are only exemplary and explanatory, not The disclosure can be limited.
Brief description of the drawings
Accompanying drawing herein is merged in specification and forms the part of this specification, shows the implementation for meeting the disclosure Example, and be used to together with specification to explain the principle of the disclosure.
Fig. 1 is a kind of flow chart of character identifying method according to an exemplary embodiment.
Fig. 2 is a kind of flow chart of character correction process according to an exemplary embodiment.
Fig. 3 is a kind of flow chart of character identifying method according to an exemplary embodiment.
Fig. 4 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 5 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 6 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 7 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 8 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 9 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Embodiment
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Following description is related to During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment Described in embodiment do not represent all embodiments consistent with the disclosure.On the contrary, they be only with it is such as appended The example of the consistent apparatus and method of some aspects be described in detail in claims, the disclosure.
OCR (Optical Character Recognition, optical character identification) can be used to enter line character at present Identification, but need to carry out image preprocessing, individual character segmentation, individual character identification, language model decoding in the character recognition process Etc. link, each link is independent step, it is easy to causes cumulative errors.
Fig. 1 is a kind of flow chart of character identifying method according to an exemplary embodiment, as shown in figure 1, the word Accord with recognition methods to be used in the equipment such as terminal, comprise the following steps 101-103：
In a step 101, feature extraction is carried out to character zone image to be identified using the first convolutional neural networks, obtained Characteristic pattern matrix.
In a step 102, the characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence.
In step 103, a line characteristic sequence is identified using Recognition with Recurrent Neural Network, obtained described to be identified Character in character zone image.
Here, first convolutional neural networks (Convolutional Neural Network, CNN) and circulation nerve Network can be terminal oneself training, can also be sent to this terminal after other-end training.Here, first convolution god It can all lead, therefore the first convolutional neural networks and the Recognition with Recurrent Neural Network can be formed through network and the Recognition with Recurrent Neural Network One overall network, this overall network is trained end to end using great amount of samples, finally trained in obtained overall network With regard to including the first convolutional neural networks for feature extraction and the Recognition with Recurrent Neural Network for carrying out characteristic sequence identification.
Here, terminal first can carry out character machining after image is got to the image, detect occur word in image The region of symbol, character zone image to be identified is obtained, the character zone image to be identified obtained here is independent a line character Corresponding character zone image to be identified, the conventional method of character machining have method based on sliding window and based on connected domains Method, those skilled in the art know, will not be described in detail herein.
Here, after terminal obtains character zone image to be identified, for the character zone to be identified belonging to each line character Image, the first convolutional neural networks trained can be used to carry out feature extraction to the character zone image to be identified, obtained Characteristic pattern F corresponding to the character zone image to be identified, needs exist for explanation, and the character zone image to be identified is stored in Data in terminal are a matrixes, and the data of the characteristic pattern F storages extracted in the terminal are a characteristic pattern matrixes, therefore can So that the characteristic pattern extracted is referred to as into characteristic pattern matrix, this feature figure matrix is wlRow wfThe matrix of row, terminal is by this feature figure square Battle array, which carries out cutting, by each column or row can be divided into a vector, exemplified by according to each column cutting, the i-th row can be designated as to Measure fi, so can be obtained by a length is wfCharacteristic sequence, this feature sequence can be designated as
Need exist for illustrating, the character in the character zone image to be identified is from left to right transversely arranged a line Character, characteristic pattern matrix can be now pressed after arranging progress cutting, the spy that the every column vector segmented is belonged in same character Sign vector, facilitates the identification of subsequent characteristics sequence；Certainly, it is from top to bottom for the character in the character zone image to be identified Longitudinal arrangement, characteristic pattern matrix by rows can be subjected to cutting, the every row vector so segmented is belonged in same character Characteristic vector, or, terminal can also by with longitudinal character character zone image travel direction to be identified adjust, obtain The character zone image to be identified of horizontal character, characteristic pattern matrix is then subjected to cutting by row.
Here, the Recognition with Recurrent Neural Network (RNN, Recurrent Neural Networks), foundation be characteristic sequence and The model of relation between character, example, the Recognition with Recurrent Neural Network can be LSTM (Long Short-Term Memory, length Short-term memory) network, the LSTM networks output sequence and the sequence inputted are one-to-one, but in character recognition, identification The character string length gone out is much smaller than the characteristic sequence inputted, so identification, this implementation directly can not be modeled with LSTM networks Example can use CTC (Connectionist Temporal Classification, connectionism time sorter) to solve This problem, a blank character is added in glossary of symbols, is then identified using LSTM networks, finally blank character and in advance The replicator measured eliminates, and than have identified one " -- a-bb " if possible, just corresponding character " ab ", thus allows this The character that LSTM can be less than list entries to length is identified.Certainly, terminal can also be using other method come to right The characteristic sequence is identified, and obtains the character in the character zone image to be identified, is not limited herein.
From the above mentioned, carrying out the overall network of the character identifying method includes the first convolutional neural networks and Recognition with Recurrent Neural Network Two sub-networks, these sub-networks can all be led, and overall network can be trained end to end, obtain the first convolution nerve net Network and Recognition with Recurrent Neural Network, this just effectively eliminates the cumulative errors between step, and by the way that image is converted into a spy Characteristic sequence identification is carried out after levying sequence, can be avoided and single word is cut, the identification for preventing miscut from occurring Mistake, improve the accuracy rate of character recognition.
The present embodiment can use the first convolutional neural networks to carry out feature extraction to character zone image to be identified, obtain Characteristic pattern matrix；The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；Using Recognition with Recurrent Neural Network to the spy Sign sequence is identified, and the character in the character zone image to be identified is obtained, in this way, the first convolutional neural networks used It can all be led with Recognition with Recurrent Neural Network, including the overall network of the first convolutional neural networks and Recognition with Recurrent Neural Network can be held To the training at end, can effectively disappears when so carrying out character recognition using first convolutional neural networks and Recognition with Recurrent Neural Network Characteristic sequence identification is carried out except the cumulative errors between step, and by the way that image is converted into after a characteristic sequence, can be with Avoid and single word is cut, the identification mistake for preventing miscut from occurring, further improve the accurate of character recognition Rate.
In a kind of possible embodiment, above-mentioned character identifying method can also comprise the following steps A1, and step 101 can To be embodied as following steps A2.
In step A1, the character zone image to be identified is corrected, obtains the neat correction chart picture of character.
In step A2, feature extraction is carried out to the correction chart picture by the first convolutional neural networks, obtains characteristic pattern Matrix.
Here, the character recognition in natural scene image is much more difficult relative to the character recognition in scanned document, because Natural scene has uncontrollable illumination, resolution ratio, character direction, character types etc., can cause word there is some deformation, Perspective transform when such as shooting, character direction tilts during shooting, and irregular typesetting of character etc. can cause the word photographed to be deposited In deformation；For the character of these deformation, terminal can be after character zone image to be identified be detected, to the character to be identified Area image is corrected, and obtains the neat correction chart picture of character, character here neatly refer to character overall alignment it is neat and Each character, which is rectified, not to be tilted, and correction here can be the character to rectify by inclined character correction, will be misaligned Character correction for marshalling character etc..
Here, can uses the first convolutional neural networks to correction chart picture after terminal obtains the neat correction chart picture of character Carry out feature extraction and obtain characteristic pattern matrix, and the characteristic pattern matrix is subjected to cutting, obtain a line characteristic sequence；Reuse Characteristic sequence is identified Recognition with Recurrent Neural Network, obtains the character in character zone image to be identified.
The present embodiment can be corrected to character zone image to be identified, obtain the neat correction chart picture of character, then Character neat in correction chart picture is identified, by character correction, the influence that character deformation band comes is effectively reduced, improves character The accuracy rate of identification.
In a kind of possible embodiment, the step A1 in above-mentioned character identifying method may be embodied as following steps A11 to A14.
In step A11, the length of the character zone image to be identified and height are scaled in proportion, contracted Image is put, wherein, the height of the zoomed image is preset height.
In step A12, enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain Key point.
In step A13, according to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline letter Number TPS.
In step A14, the zoomed image is corrected according to the thin plate spline function TPS, obtains the school Positive image.
Here, for character zone image to be identified, the character zone image to be identified is zoomed in and out first, it is high Degree zooms to preset height such as 32 pixels, and width is zoomed in and out by the scaling same with height, and the width after scaling is w The size of image after (width of different images is different here), that is, scaling is w × 32, and the image after note scaling is figure As I；By character zone image scaling to be identified to sustained height, the setting of follow-up predeterminated position point, Yi Jihou can be facilitated The identification of continuous characteristic sequence.
Here, the second convolutional neural networks can be included in the overall network, in end-to-end training overall network, will be trained The second convolutional neural networks for character key point location are obtained, terminal oneself can be trained or be somebody's turn to do from other-end Second convolutional neural networks.
Here, terminal can use the second convolutional neural networks trained to carry out word key point to the image after scaling Positioning, obtains key point, it is assumed that n is the number of key point.Predeterminated position point can be previously provided with terminal, rule is set For：The predeterminated position point is the two parallel location points of row, and the differences in height of 2 column position points is preset height, adjacent position in each column Spacing between point is preset value, and after being arranged such n predeterminated position point, terminal can pass through key point and its predeterminated position The corresponding relation of point, calculates the coefficient of TPS (Thin Plate Spline, thin plate spline function), obtains corresponding TPS M can be designated as, then the TPS is that M carries out TPS conversion to zoomed image I, will obtain correction chart as J, its correcting image J Size it is identical with zoomed image I size, be also w × 32.
Example, Fig. 2 is a kind of flow chart of character correction process according to an exemplary embodiment, with reference to figure 2, Terminal enters the positioning of line character key point using the second convolutional neural networks to zoomed image 201, obtains the pass shown in A figures in Fig. 2 Key point 202, there are 12 key points, be designated as P={ p1,p2,...,p12, then, obtain as shown in B figures in Fig. 2, the key point Predeterminated position point 203 corresponding to 202, the predeterminated position point 203 also have 13, are designated as P '={ p1′,p2′,...,p12', terminal It can calculate TPS coefficient by key point P and its predeterminated position point P ' corresponding relation, obtain corresponding TPS, so The TPS carries out TPS conversion to zoomed image afterwards, the correction chart in Fig. 2 shown in C figures will be obtained as 204, such as C figures, school in Fig. 2 Character arrangements in positive image 204 are neat, and character, which is rectified, not to be tilted.
The present embodiment can zoom in and out the character zone image to be identified, obtain zoomed image, use volume Two Product neutral net enters the positioning of line character key point to the zoomed image, obtains key point；According to the key point and default position Corresponding relation a little is put, it is determined that corresponding thin plate spline function TPS；The scaling is schemed according to the thin plate spline function TPS As carrying out TPS conversion, the correction chart picture is obtained, so can fast accurately obtain the neat correction chart picture of character.
In a kind of possible embodiment, the Recognition with Recurrent Neural Network includes the first LSTM networks and the 2nd LSTM nets Network, the step 103 in above-mentioned character identifying method may be embodied as following steps B1 and B2.
In step bl is determined, the characteristic sequence is encoded using the first LSTM networks, the feature sequence after being encoded Row.
In step B2, the characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains described treat Identify the character in character zone image.
Here, above-mentioned Recognition with Recurrent Neural Network includes the first LSTM networks and the 2nd LSTM networks, the first LSTM networks For encoder, the 2nd LSTM networks are decoder.Can be by the second convolutional neural networks network, first convolutional neural networks Network, the first LSTM networks, the 2nd LSTM networks form an overall network, and overall network is carried out end to end using same sample Training obtain aforementioned four sub-network after, using four sub-networks carry out respectively character correction, feature extraction, feature coding and Feature decodes, and can effectively eliminate the accumulated error between step.
Here, the characteristic sequence that the first LSTM networks are used to obtain step 102 is encoded, the feature after being encoded Sequence Fencoder, here after the first LSTM coding, obtained not from the character zone image to be identified of different length Characteristic sequence can with length is encoded into the characteristic sequence with same length, conveniently second subsequently as decoder The decoding of LSTM networks.Here, the characteristic sequence F after the codingencoderLength can be defaulted as 256 or 128 etc..
Here, the first LSTM networks are the one of RNN (Recurrent Neural Networks, Recognition with Recurrent Neural Network) Kind, its cataloged procedure is as follows：By characteristic sequenceAs the first LSTM list entries, and close this The output of one LSTM networks, after characteristic sequence inputs, the hidden state vector of the first LSTM networks is exactly after encoding Characteristic sequence Fencoder。
Here, the 2nd LSTM networks are used to be decoded the characteristic sequence after the first LSTM network codes, the knot of decoding Fruit is exactly the character in character zone image to be identified, and the 2nd LSTM networks are RNN one kind, and its decoding process is as follows：Will Characteristic sequence F after codingencoderInitial hidden as LSTM is vectorial, and then input opening flag symbol starts the cycle over defeated Go out the result of decoding, one cycle exports a character, and using the output of previous cycle as circulate next time input (when So, can not also be using the output of previous cycle as the input circulated next time), then stop following when its end of output flag bit Ring.
The present embodiment will can be encoded using the first LSTM networks to the characteristic sequence, the feature after being encoded Sequence；Then the characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains the character area to be identified Character in area image, the identification of characteristic sequence is so carried out, it is convenient accurate.
In a kind of possible embodiment, the above-mentioned further comprising the steps of C1 and C2 of character identifying method.
In step C1, the sample of training is obtained, the sample includes the image of character region.
In step C2, end-to-end training is carried out to overall network using the sample, the overall network trained is described The overall network trained includes the first convolutional neural networks network and the Recognition with Recurrent Neural Network.
Here, the first convolutional neural networks network and Recognition with Recurrent Neural Network are included for carrying out the overall network of character recognition, Terminal can obtain the sample of a large amount of training, and these samples are all the images of neat character region line by line, and terminal can So that these samples are inputted into overall network, after the processing of the first convolutional neural networks network and LSTM networks, output identifies Character, terminal may determine that whether are the character of the output and the character in the image, if the same identification correctly, if not The same then identify mistake, terminal can constantly adjust the parameter in overall network in each network, until identifying the accuracy of the sample More than predetermined threshold value such as 95% etc., then the overall network trained, then, terminal can is by character zone image to be identified Input the overall network trained, carry out step 101 to after 103, it is possible to export the character in character zone image to be identified.
Explanation is needed exist for, if character zone image to be identified has the character of deformation, can also be wrapped in overall network The second convolutional neural networks network for correcting character is included, the sample for a large amount of training that terminal obtains includes deformation character The image of region, these samples can be inputted overall network by terminal, by the first convolutional neural networks network, the first convolution After the processing of neutral net network and Recognition with Recurrent Neural Network, the character identified is exported, terminal may determine that the character of the output With the character in the image whether, identified if the same correct, mistake identified if different, terminal can be adjusted constantly Parameter in overall network in each network, until identifying that the accuracy of the sample exceedes predetermined threshold value such as 95% etc., then trained Good overall network, then, character zone image to be identified is inputted the overall network trained by terminal can, by character school Just, can exports the character in character zone image to be identified after feature extraction and characteristic sequence identify.
The present embodiment can use sample to carry out end-to-end training, the overall network trained, the instruction to overall network The overall network perfected includes the first convolutional neural networks network and Recognition with Recurrent Neural Network, and such whole network can carry out end-to-end Training, effectively eliminated the accumulated error between step.
Implementation process is discussed in detail below by several embodiments.
Fig. 3 is a kind of flow chart of character identifying method according to an exemplary embodiment, as shown in figure 3, the party Method can be realized by equipment such as terminals, including step 301-310.
In step 301, the sample of training is obtained, the sample includes the image of character region.
In step 302, end-to-end training, the overall network trained, institute are carried out to overall network using the sample State the overall network that trains include the first convolutional neural networks network, the second convolutional neural networks network, the first LSTM networks and 2nd LSTM networks.
In step 303, the length of the character zone image to be identified and height are scaled in proportion, contracted Image is put, wherein, the height of the zoomed image is preset height.
In step 304, enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain Key point.
In step 305, according to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline letter Number TPS.
Within step 306, TPS conversion is carried out to the zoomed image according to the thin plate spline function TPS, obtained described Correction chart picture.
In step 307, feature extraction is carried out to the correction chart picture by the first convolutional neural networks, obtains characteristic pattern Matrix.
In step 308, the characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence.
In a step 309, a line characteristic sequence is encoded using the first LSTM networks, the spy after being encoded Levy sequence.
In the step 310, the characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains described treat Identify the character in character zone image.
Following is embodiment of the present disclosure, can be used for performing embodiments of the present disclosure.
Fig. 4 is a kind of block diagram of character recognition device according to an exemplary embodiment, and the device can be by soft Part, hardware or both are implemented in combination with as some or all of of electronic equipment.As shown in figure 4, the character recognition device Including：Extraction module 401, cutting module 402 and identification module 403；Wherein：
Extraction module 401, for carrying out feature extraction to character zone image to be identified by the first convolutional neural networks, Obtain characteristic pattern matrix；
Cutting module 402, for the characteristic pattern matrix to be carried out into cutting, obtain a line characteristic sequence；
Identification module 403, for a line characteristic sequence to be identified by Recognition with Recurrent Neural Network, obtain described treat Identify the character in character zone image.
As a kind of possible embodiment, Fig. 5 is a kind of character recognition device according to an exemplary embodiment Block diagram, as shown in figure 5, character recognition device disclosed above can be configured to include correction module 404, the extraction Module 401 is configured to include extracting sub-module 4011, wherein：
Correction module 404, for being corrected to the character zone image to be identified, obtain the neat correction chart of character Picture；
Extracting sub-module 4011, for carrying out feature extraction to the correction chart picture by the first convolutional neural networks, obtain To characteristic pattern matrix.
As a kind of possible embodiment, Fig. 6 is a kind of character recognition device according to an exemplary embodiment Block diagram, as shown in fig. 6, character recognition device disclosed above can also be configured to the correction module 404 to include scaling Module 4041, positioning submodule 4042, determination sub-module 4043 and transformation submodule 4044, wherein：
Submodule 4041 is scaled, for the length of the character zone image to be identified and height to be contracted in proportion Put, obtain zoomed image, wherein, the height of the zoomed image is preset height；
Submodule 4042 is positioned, is determined for entering line character key point to the zoomed image using the second convolutional neural networks Position, obtains key point；
Determination sub-module 4043, for the corresponding relation according to the key point and predeterminated position point, it is determined that corresponding thin Plate spline function TPS；
Transformation submodule 4044, for carrying out TPS conversion to the zoomed image according to the thin plate spline function TPS, Obtain the correction chart picture.
As a kind of possible embodiment, the Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and second LSTM networks, Fig. 7 is a kind of block diagram of character recognition device according to an exemplary embodiment, as shown in fig. 7, above-mentioned public affairs The character recognition device opened can also be configured to the identification module 403 to include encoding submodule 4031 and decoding sub-module 4032, wherein：
Encoding submodule 4031, for being encoded using the first LSTM networks to a line characteristic sequence, compiled Characteristic sequence after code；
Decoding sub-module 4032, for being decoded using the 2nd LSTM networks to the characteristic sequence after the coding, obtain To the character in the character zone image to be identified.
As a kind of possible embodiment, Fig. 8 is a kind of character recognition device according to an exemplary embodiment Block diagram, as shown in figure 8, character recognition device disclosed above can be configured to include acquisition module 405 and training module 406, wherein：
Acquisition module 405, for obtaining the sample of training, the sample includes the image of character region；
Training module 406, for carrying out end-to-end training, the total net trained to overall network using the sample Network, the overall network trained include first convolutional neural networks and the Recognition with Recurrent Neural Network.
On the device in above-described embodiment, wherein modules perform the concrete mode of operation in relevant this method Embodiment in be described in detail, explanation will be not set forth in detail herein.
Fig. 9 is a kind of block diagram of character recognition device according to an exemplary embodiment, and the device is applied to terminal Equipment.For example, device 900 can be mobile phone, and game console, computer, tablet device, personal digital assistant etc..
Device 900 can include following one or more assemblies：Processing component 901, memory 902, power supply module 903, Multimedia groupware 904, audio-frequency assembly 905, input/output (I/O) interface 906, sensor cluster 907, and communication component 908。
The integrated operation of the usual control device 900 of processing component 901, such as communicated with display, call, data, phase The operation that machine operates and record operation is associated.Processing component 901 can refer to including one or more processors 920 to perform Order, to complete all or part of step of above-mentioned method.In addition, processing component 901 can include one or more modules, just Interaction between processing component 901 and other assemblies.For example, processing component 901 can include multi-media module, it is more to facilitate Interaction between media component 904 and processing component 901.
Memory 902 is configured as storing various types of data to support the operation in device 900.These data are shown Example includes the instruction of any application program or method for being operated on device 900, contact data, telephone book data, disappears Breath, picture, video etc..Memory 902 can be by any kind of volatibility or non-volatile memory device or their group Close and realize, as static RAM (SRAM), Electrically Erasable Read Only Memory (EEPROM) are erasable to compile Journey read-only storage (EPROM), programmable read only memory (PROM), read-only storage (ROM), magnetic memory, flash Device, disk or CD.
Power supply module 903 provides electric power for the various assemblies of device 900.Power supply module 903 can include power management system System, one or more power supplys, and other components associated with generating, managing and distributing electric power for device 900.
Multimedia groupware 904 is included in the screen of one output interface of offer between described device 900 and user.One In a little embodiments, screen can include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel, screen Curtain may be implemented as touch-screen, to receive the input signal from user.Touch panel includes one or more touch sensings Device is with the gesture on sensing touch, slip and touch panel.The touch sensor can not only sensing touch or sliding action Border, but also detect and touched or the related duration and pressure of slide with described.In certain embodiments, more matchmakers Body component 904 includes a front camera and/or rear camera.When device 900 is in operator scheme, such as screening-mode or During video mode, front camera and/or rear camera can receive outside multi-medium data.Each front camera and Rear camera can be a fixed optical lens system or have focusing and optical zoom capabilities.
Audio-frequency assembly 905 is configured as output and/or input audio signal.For example, audio-frequency assembly 905 includes a Mike Wind (MIC), when device 900 is in operator scheme, during such as call model, logging mode and speech recognition mode, microphone by with It is set to reception external audio signal.The audio signal received can be further stored in memory 902 or via communication set Part 908 is sent.In certain embodiments, audio-frequency assembly 905 also includes a loudspeaker, for exports audio signal.
I/O interface 906 provides interface, above-mentioned peripheral interface module between processing component 901 and peripheral interface module Can be keyboard, click wheel, button etc..These buttons may include but be not limited to：Home button, volume button, start button and Locking press button.
Sensor cluster 907 includes one or more sensors, and the state for providing various aspects for device 900 is commented Estimate.For example, sensor cluster 907 can detect opening/closed mode of device 900, and the relative positioning of component, for example, it is described Component is the display and keypad of device 900, and sensor cluster 907 can be with 900 1 components of detection means 900 or device Position change, the existence or non-existence that user contacts with device 900, the orientation of device 900 or acceleration/deceleration and device 900 Temperature change.Sensor cluster 907 can include proximity transducer, be configured to detect in no any physical contact The presence of neighbouring object.Sensor cluster 907 can also include optical sensor, such as CMOS or ccd image sensor, for into As being used in application.In certain embodiments, the sensor cluster 907 can also include acceleration transducer, gyro sensors Device, Magnetic Sensor, pressure sensor or temperature sensor.
Communication component 908 is configured to facilitate the communication of wired or wireless way between device 900 and other equipment.Device 900 can access the wireless network based on communication standard, such as WiFi, 2G or 3G, or combinations thereof.In an exemplary implementation In example, communication component 908 receives broadcast singal or broadcast related information from external broadcasting management system via broadcast channel. In one exemplary embodiment, the communication component 908 also includes near-field communication (NFC) module, to promote junction service.Example Such as, in NFC module radio frequency identification (RFID) technology can be based on, Infrared Data Association (IrDA) technology, ultra wide band (UWB) technology, Bluetooth (BT) technology and other technologies are realized.
In the exemplary embodiment, device 900 can be believed by one or more application specific integrated circuits (ASIC), numeral Number processor (DSP), digital signal processing appts (DSPD), PLD (PLD), field programmable gate array (FPGA), controller, microcontroller, microprocessor or other electronic components are realized, for performing the above method.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium including instructing, example are additionally provided Such as include the memory 902 of instruction, above-mentioned instruction can be performed to complete the above method by the processor 920 of device 900.For example, The non-transitorycomputer readable storage medium can be ROM, random access memory (RAM), CD-ROM, tape, floppy disk With optical data storage devices etc..
A kind of computer-readable recording medium is present embodiments provided, when the instruction in the storage medium is by device 900 Computing device when realize following steps：
Feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtains characteristic pattern matrix；
The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtains the character zone image to be identified In character.
Instruction in the storage medium can also realize following steps when being executed by processor：
Methods described also includes：
The character zone image to be identified is corrected, obtains the neat correction chart picture of character；
It is described that feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtain characteristic pattern square Battle array, including：
Feature extraction is carried out to the correction chart picture by first convolutional neural networks, obtains characteristic pattern matrix.
Instruction in the storage medium can also realize following steps when being executed by processor：
It is described that the character zone image to be identified is corrected, the neat correction chart picture of character is obtained, including：
The length of the character zone image to be identified and height are scaled in proportion, obtain zoomed image, wherein, The height of the zoomed image is preset height；
Enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain key point；
According to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline function TPS；
TPS conversion is carried out to the zoomed image according to the thin plate spline function TPS, obtains the correction chart picture.
Instruction in the storage medium can also realize following steps when being executed by processor：
The Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and the 2nd LSTM networks, described to pass through circulation A line characteristic sequence is identified neutral net, obtains the character in the character zone image to be identified, including：
A line characteristic sequence is encoded using the first LSTM networks, the characteristic sequence after being encoded；
The characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains the character zone to be identified Character in image.
Instruction in the storage medium can also realize following steps when being executed by processor：
Methods described also includes：
The sample of training is obtained, the sample includes the image of character region；
Carry out end-to-end training to overall network using the sample, the overall network trained, it is described train it is total Network includes first convolutional neural networks and the Recognition with Recurrent Neural Network.
The present embodiment additionally provides a kind of character recognition device, including：
Processor；
For storing the memory of processor-executable instruction；
Wherein, the processor is configured as：
Feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtains characteristic pattern matrix；
The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtains the character zone image to be identified In character.
The processor can be additionally configured to：
Methods described also includes：
The character zone image to be identified is corrected, obtains the neat correction chart picture of character；
It is described that feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtain characteristic pattern square Battle array, including：
Feature extraction is carried out to the correction chart picture by first convolutional neural networks, obtains characteristic pattern matrix.
The processor can be additionally configured to：
It is described that the character zone image to be identified is corrected, the neat correction chart picture of character is obtained, including：
The length of the character zone image to be identified and height are scaled in proportion, obtain zoomed image, wherein, The height of the zoomed image is preset height；
Enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain key point；
According to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline function TPS；
TPS conversion is carried out to the zoomed image according to the thin plate spline function TPS, obtains the correction chart picture.
The processor can be additionally configured to：The Recognition with Recurrent Neural Network include the first long short-term memory LSTM networks and 2nd LSTM networks, it is described that a line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtain the word to be identified The character in area image is accorded with, including：
A line characteristic sequence is encoded using the first LSTM networks, the characteristic sequence after being encoded；
The characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains the character zone to be identified Character in image.
The processor can be additionally configured to：
Methods described also includes：
The sample of training is obtained, the sample includes the image of character region；
Carry out end-to-end training to overall network using the sample, the overall network trained, it is described train it is total Network includes first convolutional neural networks and the Recognition with Recurrent Neural Network.
Those skilled in the art will readily occur to the disclosure its after considering specification and putting into practice disclosure disclosed herein Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or Person's adaptations follow the general principle of the disclosure and including the undocumented common knowledges in the art of the disclosure Or conventional techniques.Description and embodiments are considered only as exemplary, and the true scope of the disclosure and spirit are by following Claim is pointed out.
It should be appreciated that the precision architecture that the disclosure is not limited to be described above and is shown in the drawings, and And various modifications and changes can be being carried out without departing from the scope.The scope of the present disclosure is only limited by appended claim.

Claims (12)

  1. A kind of 1. character identifying method, it is characterised in that including：
    Feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, obtains characteristic pattern matrix；
    The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
    A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtained in the character zone image to be identified Character.
  2. 2. according to the method for claim 1, it is characterised in that methods described also includes：
    The character zone image to be identified is corrected, obtains the neat correction chart picture of character；
    It is described that feature extraction is carried out to character zone image to be identified by the first convolutional neural networks, characteristic pattern matrix is obtained, Including：
    Feature extraction is carried out to the correction chart picture by first convolutional neural networks, obtains characteristic pattern matrix.
  3. 3. according to the method for claim 2, it is characterised in that described that school is carried out to the character zone image to be identified Just, the neat correction chart picture of character is obtained, including：
    The length of the character zone image to be identified and height are scaled in proportion, obtain zoomed image, wherein, it is described The height of zoomed image is preset height；
    Enter the positioning of line character key point to the zoomed image by the second convolutional neural networks, obtain key point；
    According to the key point and the corresponding relation of predeterminated position point, it is determined that corresponding thin plate spline function TPS；
    TPS conversion is carried out to the zoomed image according to the thin plate spline function TPS, obtains the correction chart picture.
  4. 4. according to the method for claim 1, it is characterised in that the Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and the 2nd LSTM networks, it is described that a line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtain institute The character in character zone image to be identified is stated, including：
    A line characteristic sequence is encoded using the first LSTM networks, the characteristic sequence after being encoded；
    The characteristic sequence after the coding is decoded using the 2nd LSTM networks, obtains the character zone image to be identified In character.
  5. 5. according to the method for claim 1, it is characterised in that methods described also includes：
    The sample of training is obtained, the sample includes the image of character region；
    End-to-end training, the overall network trained, the overall network trained are carried out to overall network using the sample Including first convolutional neural networks and the Recognition with Recurrent Neural Network.
  6. A kind of 6. character recognition device, it is characterised in that including：
    Extraction module, for carrying out feature extraction to character zone image to be identified by the first convolutional neural networks, obtain spy Levy figure matrix；
    Cutting module, for the characteristic pattern matrix to be carried out into cutting, obtain a line characteristic sequence；
    Identification module, for a line characteristic sequence to be identified by Recognition with Recurrent Neural Network, obtain the word to be identified Accord with the character in area image.
  7. 7. device according to claim 6, it is characterised in that described device also includes：
    Correction module, for being corrected to the character zone image to be identified, obtain the neat correction chart picture of character；
    The extraction module includes：
    Extracting sub-module, for carrying out feature extraction to the correction chart picture by first convolutional neural networks, obtain spy Levy figure matrix.
  8. 8. device according to claim 7, it is characterised in that the correction module includes：
    Submodule is scaled, for the length of the character zone image to be identified and height to be scaled in proportion, is contracted Image is put, wherein, the height of the zoomed image is preset height；
    Submodule is positioned, for entering the positioning of line character key point to the zoomed image by the second convolutional neural networks, is obtained Key point；
    Determination sub-module, for the corresponding relation according to the key point and predeterminated position point, it is determined that corresponding thin plate spline letter Number TPS；
    Transformation submodule, for carrying out TPS conversion to the zoomed image according to the thin plate spline function TPS, obtain described Correction chart picture.
  9. 9. device according to claim 6, it is characterised in that the Recognition with Recurrent Neural Network includes the first long short-term memory LSTM networks and the 2nd LSTM networks, the identification module include：
    Encoding submodule, for being encoded using the first LSTM networks to a line characteristic sequence, the spy after being encoded Levy sequence；
    Decoding sub-module, for being decoded using the 2nd LSTM networks to the characteristic sequence after the coding, obtain described treat Identify the character in character zone image.
  10. 10. device according to claim 6, it is characterised in that described device also includes：
    Acquisition module, for obtaining the sample of training, the sample includes the image of character region；
    Training module, for carrying out end-to-end training, the overall network trained, the instruction to overall network using the sample The overall network perfected includes first convolutional neural networks and the Recognition with Recurrent Neural Network.
  11. A kind of 11. character recognition device, it is characterised in that including：
    Processor；
    For storing the memory of processor-executable instruction；
    Wherein, the processor is configured as：
    Feature extraction is carried out to character zone image to be identified using the first convolutional neural networks, obtains characteristic pattern matrix；
    The characteristic pattern matrix is subjected to cutting, obtains a line characteristic sequence；
    A line characteristic sequence is identified by Recognition with Recurrent Neural Network, obtained in the character zone image to be identified Character.
  12. 12. a kind of computer-readable recording medium, is stored with computer instruction, it is characterised in that the computer instruction is located Reason device realizes the step in claim 1 to 5 methods described when performing.
CN201711046515.5A 2017-10-31 2017-10-31 Character identifying method and device Pending CN107798327A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711046515.5A CN107798327A (en) 2017-10-31 2017-10-31 Character identifying method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711046515.5A CN107798327A (en) 2017-10-31 2017-10-31 Character identifying method and device

Publications (1)

Publication Number Publication Date
CN107798327A true CN107798327A (en) 2018-03-13

Family

ID=61548299

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711046515.5A Pending CN107798327A (en) 2017-10-31 2017-10-31 Character identifying method and device

Country Status (1)

Country Link
CN (1) CN107798327A (en)

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596168A (en) * 2018-04-20 2018-09-28 北京京东金融科技控股有限公司 For identification in image character method, apparatus and medium
CN109255356A (en) * 2018-07-24 2019-01-22 阿里巴巴集团控股有限公司 A kind of character recognition method, device and computer readable storage medium
CN109271910A (en) * 2018-09-04 2019-01-25 阿里巴巴集团控股有限公司 A kind of Text region, character translation method and apparatus
CN109376658A (en) * 2018-10-26 2019-02-22 信雅达系统工程股份有限公司 A kind of OCR method based on deep learning
CN109583439A (en) * 2018-12-04 2019-04-05 龙马智芯（珠海横琴）科技有限公司 The method and device of text correction, storage medium, processor
CN109919076A (en) * 2019-03-04 2019-06-21 厦门商集网络科技有限责任公司 The method and medium of confirmation OCR recognition result reliability based on deep learning
CN109977969A (en) * 2019-03-27 2019-07-05 北京经纬恒润科技有限公司 A kind of image-recognizing method and device
CN110059705A (en) * 2019-04-22 2019-07-26 厦门商集网络科技有限责任公司 A kind of OCR recognition result decision method and equipment based on modeling
CN110070085A (en) * 2019-04-30 2019-07-30 北京百度网讯科技有限公司 Licence plate recognition method and device
CN110163204A (en) * 2019-04-15 2019-08-23 平安国际智慧城市科技股份有限公司 Businessman's monitoring and managing method, device and storage medium based on image recognition
CN110490199A (en) * 2019-08-26 2019-11-22 北京香侬慧语科技有限责任公司 A kind of method, apparatus of text identification, storage medium and electronic equipment
CN110490031A (en) * 2018-05-15 2019-11-22 武汉斗鱼网络科技有限公司 A kind of general digital knows method for distinguishing, storage medium, electronic equipment and system
CN110555433A (en) * 2018-05-30 2019-12-10 北京三星通信技术研究有限公司 Image processing method, image processing device, electronic equipment and computer readable storage medium
WO2019232853A1 (en) * 2018-06-04 2019-12-12 平安科技（深圳）有限公司 Chinese model training method, chinese image recognition method, device, apparatus and medium
CN110659643A (en) * 2018-06-29 2020-01-07 杭州海康威视数字技术股份有限公司 Container number identification method and device
CN110796138A (en) * 2019-10-15 2020-02-14 湖北工业大学 Irregular scene character recognition method based on dominant correction mechanism
CN110942057A (en) * 2018-09-25 2020-03-31 杭州海康威视数字技术股份有限公司 Container number identification method and device and computer equipment
CN110942073A (en) * 2018-09-25 2020-03-31 杭州海康威视数字技术股份有限公司 Container trailer number identification method and device and computer equipment
CN111046859A (en) * 2018-10-11 2020-04-21 杭州海康威视数字技术股份有限公司 Character recognition method and device
CN111147891A (en) * 2019-12-31 2020-05-12 杭州威佩网络科技有限公司 Method, device and equipment for acquiring information of object in video picture
CN111382810A (en) * 2018-12-29 2020-07-07 阿里巴巴集团控股有限公司 Character string recognition method and device and storage medium
CN111860682A (en) * 2020-07-30 2020-10-30 上海高德威智能交通系统有限公司 Sequence identification method, sequence identification device, image processing equipment and storage medium
CN112132139A (en) * 2020-09-22 2020-12-25 深兰科技（上海）有限公司 Character recognition method and device

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779275A (en) * 2012-07-04 2012-11-14 广州广电运通金融电子股份有限公司 Paper characteristic identification method and relative device
CN103530600A (en) * 2013-06-06 2014-01-22 东软集团股份有限公司 License plate recognition method and system under complicated illumination
CN103984930A (en) * 2014-05-21 2014-08-13 南京航空航天大学 Digital meter recognition system and method based on vision
CN104751427A (en) * 2015-04-16 2015-07-01 山东师范大学 Improved TPS (thin plate spline) model based gel image correction algorithm
CN105654130A (en) * 2015-12-30 2016-06-08 成都数联铭品科技有限公司 Recurrent neural network-based complex image character sequence recognition system
CN106407971A (en) * 2016-09-14 2017-02-15 北京小米移动软件有限公司 Text recognition method and device
CN106548462A (en) * 2016-11-02 2017-03-29 西安电子科技大学 Non-linear SAR image geometric correction method based on thin-plate spline interpolation
CN106557768A (en) * 2016-11-25 2017-04-05 北京小米移动软件有限公司 The method and device is identified by word in picture

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779275A (en) * 2012-07-04 2012-11-14 广州广电运通金融电子股份有限公司 Paper characteristic identification method and relative device
CN103530600A (en) * 2013-06-06 2014-01-22 东软集团股份有限公司 License plate recognition method and system under complicated illumination
CN103984930A (en) * 2014-05-21 2014-08-13 南京航空航天大学 Digital meter recognition system and method based on vision
CN104751427A (en) * 2015-04-16 2015-07-01 山东师范大学 Improved TPS (thin plate spline) model based gel image correction algorithm
CN105654130A (en) * 2015-12-30 2016-06-08 成都数联铭品科技有限公司 Recurrent neural network-based complex image character sequence recognition system
CN106407971A (en) * 2016-09-14 2017-02-15 北京小米移动软件有限公司 Text recognition method and device
CN106548462A (en) * 2016-11-02 2017-03-29 西安电子科技大学 Non-linear SAR image geometric correction method based on thin-plate spline interpolation
CN106557768A (en) * 2016-11-25 2017-04-05 北京小米移动软件有限公司 The method and device is identified by word in picture

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
BAOGUANG SHI ET AL: "An end-to-end trainable neural network for image-based sequence recogniton and its application to scene text recognition", 《ARXIV》 *

Cited By (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596168A (en) * 2018-04-20 2018-09-28 北京京东金融科技控股有限公司 For identification in image character method, apparatus and medium
CN110490031A (en) * 2018-05-15 2019-11-22 武汉斗鱼网络科技有限公司 A kind of general digital knows method for distinguishing, storage medium, electronic equipment and system
CN110555433A (en) * 2018-05-30 2019-12-10 北京三星通信技术研究有限公司 Image processing method, image processing device, electronic equipment and computer readable storage medium
CN110555433B (en) * 2018-05-30 2024-04-26 北京三星通信技术研究有限公司 Image processing method, device, electronic equipment and computer readable storage medium
WO2019232853A1 (en) * 2018-06-04 2019-12-12 平安科技（深圳）有限公司 Chinese model training method, chinese image recognition method, device, apparatus and medium
CN110659643B (en) * 2018-06-29 2022-08-26 杭州海康威视数字技术股份有限公司 Container number identification method and device
CN110659643A (en) * 2018-06-29 2020-01-07 杭州海康威视数字技术股份有限公司 Container number identification method and device
CN109255356A (en) * 2018-07-24 2019-01-22 阿里巴巴集团控股有限公司 A kind of character recognition method, device and computer readable storage medium
CN109255356B (en) * 2018-07-24 2022-02-01 创新先进技术有限公司 Character recognition method and device and computer readable storage medium
CN109271910A (en) * 2018-09-04 2019-01-25 阿里巴巴集团控股有限公司 A kind of Text region, character translation method and apparatus
CN110942073A (en) * 2018-09-25 2020-03-31 杭州海康威视数字技术股份有限公司 Container trailer number identification method and device and computer equipment
CN110942057A (en) * 2018-09-25 2020-03-31 杭州海康威视数字技术股份有限公司 Container number identification method and device and computer equipment
CN111046859B (en) * 2018-10-11 2023-09-29 杭州海康威视数字技术股份有限公司 Character recognition method and device
CN111046859A (en) * 2018-10-11 2020-04-21 杭州海康威视数字技术股份有限公司 Character recognition method and device
CN109376658A (en) * 2018-10-26 2019-02-22 信雅达系统工程股份有限公司 A kind of OCR method based on deep learning
CN109583439A (en) * 2018-12-04 2019-04-05 龙马智芯（珠海横琴）科技有限公司 The method and device of text correction, storage medium, processor
CN111382810A (en) * 2018-12-29 2020-07-07 阿里巴巴集团控股有限公司 Character string recognition method and device and storage medium
CN109919076B (en) * 2019-03-04 2022-01-04 厦门商集网络科技有限责任公司 Method and medium for confirming reliability of OCR recognition result based on deep learning
CN109919076A (en) * 2019-03-04 2019-06-21 厦门商集网络科技有限责任公司 The method and medium of confirmation OCR recognition result reliability based on deep learning
CN109977969A (en) * 2019-03-27 2019-07-05 北京经纬恒润科技有限公司 A kind of image-recognizing method and device
CN110163204A (en) * 2019-04-15 2019-08-23 平安国际智慧城市科技股份有限公司 Businessman's monitoring and managing method, device and storage medium based on image recognition
CN110059705A (en) * 2019-04-22 2019-07-26 厦门商集网络科技有限责任公司 A kind of OCR recognition result decision method and equipment based on modeling
CN110070085A (en) * 2019-04-30 2019-07-30 北京百度网讯科技有限公司 Licence plate recognition method and device
CN110490199A (en) * 2019-08-26 2019-11-22 北京香侬慧语科技有限责任公司 A kind of method, apparatus of text identification, storage medium and electronic equipment
CN110796138A (en) * 2019-10-15 2020-02-14 湖北工业大学 Irregular scene character recognition method based on dominant correction mechanism
CN111147891A (en) * 2019-12-31 2020-05-12 杭州威佩网络科技有限公司 Method, device and equipment for acquiring information of object in video picture
CN111860682A (en) * 2020-07-30 2020-10-30 上海高德威智能交通系统有限公司 Sequence identification method, sequence identification device, image processing equipment and storage medium
CN112132139A (en) * 2020-09-22 2020-12-25 深兰科技（上海）有限公司 Character recognition method and device

Similar Documents

Publication Publication Date Title
CN107798327A (en) Character identifying method and device
JP6401873B2 (en) Region recognition method and apparatus
CN107527059A (en) Character recognition method, device and terminal
CN105550633B (en) Area recognizing method and device
CN105528607A (en) Region extraction method and model training method and device
CN106530227A (en) Image restoration method and device
CN106682736A (en) Image identification method and apparatus
CN106651955A (en) Method and device for positioning object in picture
CN106295638A (en) Certificate image sloped correcting method and device
CN110602527A (en) Video processing method, device and storage medium
CN110516672A (en) Card card information identifying method, device and terminal
CN109934275B (en) Image processing method and device, electronic equipment and storage medium
CN105335754A (en) Character recognition method and device
CN107679483A (en) Number plate recognition methods and device
CN105426818A (en) Area extraction method and device
CN105335713A (en) Fingerprint identification method and device
CN107609560A (en) Character recognition method and device
CN105354560A (en) Fingerprint identification method and device
CN106228556A (en) Image quality analysis method and device
CN106778773A (en) The localization method and device of object in picture
CN107145859A (en) E-book conversion process method, device and computer-readable recording medium
CN107679533A (en) Character recognition method and device
CN107911641A (en) Video watermark generation method, device and terminal
CN104077597B (en) Image classification method and device
CN107563994A (en) The conspicuousness detection method and device of image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20180313