CN112232347B - Character recognition method, device, equipment and storage medium based on probability matrix - Google Patents
Character recognition method, device, equipment and storage medium based on probability matrix Download PDFInfo
- Publication number
- CN112232347B CN112232347B CN202010929789.4A CN202010929789A CN112232347B CN 112232347 B CN112232347 B CN 112232347B CN 202010929789 A CN202010929789 A CN 202010929789A CN 112232347 B CN112232347 B CN 112232347B
- Authority
- CN
- China
- Prior art keywords
- character
- sequence
- recognition
- probability
- alternative
- 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.)
- Active
Links
Images
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
- 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
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Character Discrimination (AREA)
Abstract
The scheme relates to a blockchain technology and provides a character recognition method, device, equipment and storage medium based on a probability matrix, namely, recognition characters at all positions of a target image and probability matrixes corresponding to all recognition characters are output through a CTC model, recognition characters corresponding to all positions are screened, the screened recognition characters are combined into character sequences, then the character sequences are circularly screened and combined, a relevant undetermined character sequence set is extracted, and the relevant undetermined character sequence with the largest probability value is determined as the target character sequence to obtain the recognition characters of the target image. In addition, the present invention relates to blockchain techniques in which a recognition character set and a character probability matrix may be stored. And the recognition characters are circularly combined and screened according to the character probability to generate a target character sequence, so that the calculated amount of character recognition is reduced, the character recognition efficiency is improved, and the accuracy of character recognition is improved.
Description
Technical Field
The present invention relates to the field of image recognition technology in artificial intelligence, and in particular, to a method, apparatus, device, and computer readable storage medium for character recognition based on probability matrix.
Background
Training a deep learning algorithm through an RCNN network structure and a CTC loss function to perform English recognition is a popular English recognition method at present, which mainly comprises the steps of calculating character probability distribution of all moments (positions) of English to be recognized through a CNN convolution layer and an RNN circulation layer, calculating character probability loss values of all moments (positions) of the English to be recognized through a CTC mode, then counter-propagating the loss values, correcting network parameters, generating a trained neural network, finally correctly outputting the character probability distribution of each position through the trained neural network, outputting English characters with the largest character probability distribution, and decoding to obtain recognized English words, word combinations or English sentences. However, due to inaccurate training data or low definition of pictures, the probability distribution of the obtained English characters to be identified is inaccurate, for example, when handwritten English is identified, similar characters such as v- > r, t- > f and the like easily reduce the accuracy of English identification. Therefore, how to solve the problem that the accuracy of English recognition is low when the definition of the existing picture is low becomes the technical problem to be solved urgently.
Disclosure of Invention
The invention mainly aims to provide a character recognition method, device and equipment based on a probability matrix and a computer readable storage medium, and aims to solve the technical problem that the English recognition accuracy is low when the definition of the existing picture is low.
In order to achieve the above object, the present invention provides a character recognition method based on a probability matrix, the character recognition method based on a probability matrix comprising the steps of:
inputting a target image into a preset CTC model to generate a recognition character set and a character probability matrix corresponding to the target image, wherein the recognition character set comprises recognition characters corresponding to all positions in the target image, one position corresponds to one recognition character subset, one recognition character corresponds to one probability value, and the character probability matrix comprises all probability values corresponding to the recognition character set;
sequentially acquiring a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sequencing each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from big to small, acquiring a preset number of recognition characters arranged in front, and generating a first alternative character set and the second alternative character set, wherein the first position is the leftmost character position of the target image, and the second position is the leftmost next character position;
Traversing each first alternative character in the first alternative character set and each second alternative character in the second alternative character set, combining each first alternative character with each second alternative character to generate an alternative character sequence set, determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence, determining a first to-be-determined character sequence according to the sequence probability of the alternative character sequence, and generating a first to-be-determined character sequence set;
sequentially acquiring the next position corresponding to the second position as a third position, sequencing and fetching the recognition character subset corresponding to the third position to generate a third alternative character set, and combining the first to generate a second to-be-determined character sequence set until the combination of recognition characters of all positions is completed to generate a related to-be-determined character sequence set;
and determining the relevant undetermined character sequence with the largest relevant sequence probability as a target character sequence in the relevant undetermined character sequence set according to the relevant sequence probability of the relevant character sequence contained in each relevant undetermined character sequence, and outputting an analysis result of the target character sequence to finish the recognition of the target image.
Optionally, the step of sorting and fetching the recognized character subset corresponding to the third position to generate a third alternative character set, and combining the first pending character sequence set with the third alternative character set to generate a second pending character sequence set specifically includes:
sequentially acquiring a third recognition character subset corresponding to a third position, sequencing all characters in the third recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as third alternative characters, and generating a third alternative character set;
acquiring a first character sequence to be determined in the first character sequence set as a current first character sequence to be determined, and acquiring a third alternative character in the third alternative character set as the current third alternative character;
judging whether the current third alternative character is an English character or not;
if the current third alternative character is not an English character, judging whether the current first character sequence to be determined is a recognizable word according to a preset vocabulary;
and if the current first waiting character sequence is a recognizable word, compensating the current first waiting character sequence, and storing a word analysis result corresponding to the current first waiting character sequence.
Optionally, after the step of determining whether the current third alternative character is an english character, the method further includes:
if the current third alternative character is an English character, combining the current first undetermined character sequence with the current third alternative character to generate a second undetermined character sequence;
and acquiring the next third alternative character in the third alternative character set as the current third alternative character until each first alternative character sequence in the first alternative character sequence set and each third alternative character in the third alternative character set are combined to generate a second alternative character sequence set.
Optionally, after the step of determining whether the current first predetermined character sequence is a recognizable word according to a preset vocabulary if the current third candidate character is not an english character, the method further includes:
if the current first character sequence is not the recognizable word, storing the current first character sequence as a character string to be recognized into a data table;
determining the target number according to the word processing number corresponding to a preset language model, sequentially acquiring character strings to be recognized of the target number in the data table, and inputting the character strings to be recognized of the target number into the language model so as to calculate continuous word relation probability of the character strings to be recognized of the target number through the language model;
And when the continuous word relation probability is larger than a preset threshold value, storing continuous word analysis results corresponding to the character strings to be identified of the target number.
Optionally, the step of sequentially obtaining a first recognition character subset corresponding to the first position and a second recognition character subset corresponding to the second position, sorting each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from big to small, obtaining the recognition characters of which the number is preset before, and generating a first candidate character set and the second candidate character set specifically includes:
sequentially acquiring a first recognition character subset corresponding to a first position, sequencing each character in the first recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as first alternative characters, and generating a first alternative character set;
and acquiring the next position of the first position as a second position, acquiring a second recognition character subset corresponding to the second position, sequencing all characters in the second recognition character subset according to the sequence from the big probability to the small probability, acquiring the recognition characters arranged in the preset number as second candidate characters, and generating a second candidate character set.
Optionally, the recognition character set and the character probability matrix corresponding to the target image are stored in a blockchain.
Optionally, the step of determining the sequence probability of the alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence, determining the first to-be-determined character sequence according to the sequence probability of the alternative character sequence, and generating the first to-be-determined character sequence set specifically includes:
determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence;
sequencing each character sequence in the alternative character sequence set according to the sequence probability from high to low;
and acquiring character sequences arranged in the preset number as a first character sequence to be determined, and generating a first character sequence set to be determined.
In addition, in order to achieve the above object, the present invention also provides a character recognition device based on a probability matrix, the character recognition device based on a probability matrix comprising:
the character probability acquisition module is used for inputting a target image into a preset CTC model and generating a recognition character set and a character probability matrix corresponding to the target image, wherein the recognition character set comprises recognition characters corresponding to all positions in the target image, one position corresponds to one recognition character subset, one recognition character corresponds to one probability value, and the character probability matrix comprises all probability values corresponding to the recognition character set;
The candidate character set generation module is used for sequentially acquiring a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sequencing each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from high to low, acquiring a preset number of recognition characters arranged in front, and generating a first candidate character set and the second candidate character set, wherein the first position is the leftmost character position of the target image, and the second position is the leftmost next character position;
the device comprises a to-be-determined sequence set generation module, a first set of first candidate characters and a second set of second candidate characters, wherein the to-be-determined sequence set generation module is used for traversing each first candidate character in the first candidate character set and each second candidate character in the second candidate character set, combining each first candidate character with each second candidate character to generate a candidate character sequence set, determining the sequence probability of each candidate character sequence according to the character probability of the candidate character contained in each candidate character sequence, determining a first to-be-determined character sequence according to the sequence probability of the candidate character sequence, and generating a first to-be-determined character sequence set;
The related sequence set generating module is used for sequentially acquiring the next position corresponding to the second position, serving as a third position, sequencing and fetching the recognition character subset corresponding to the third position to generate a third alternative character set, combining the first to-be-determined character sequence set with the third alternative character set to generate a second to-be-determined character sequence set until the combination of recognition characters of all positions is completed, and generating the related to-be-determined character sequence set;
and the identification character output module is used for determining the relevant undetermined character sequence with the largest relevant sequence probability as a target character sequence in the relevant undetermined character sequence set according to the relevant sequence probability of the relevant character sequence contained in each relevant undetermined character sequence, and outputting an analysis result of the target character sequence to finish the identification of the target image.
In addition, in order to achieve the above object, the present invention also provides a probability matrix based character recognition apparatus including a processor, a memory, and a probability matrix based character recognition program stored on the memory and executable by the processor, wherein the probability matrix based character recognition program, when executed by the processor, implements the steps of the probability matrix based character recognition method as described above.
In addition, in order to achieve the above object, the present invention also provides a computer-readable storage medium having stored thereon a probability matrix-based character recognition program, wherein the probability matrix-based character recognition program, when executed by a processor, implements the steps of the probability matrix-based character recognition method as described above.
The invention provides a character recognition method based on a probability matrix, which outputs recognition characters at all positions of a target image and probability matrixes corresponding to all recognition characters through a CTC model, screens the recognition characters corresponding to all positions, combines the screened recognition characters into character sequences, circularly screens and combines the character sequences to extract relevant undetermined character sequence sets, and determines the relevant undetermined character sequence with the largest probability value in the relevant undetermined character sequence sets as a target character sequence to obtain the recognition characters of the target image. The character probability matrix of the target image is output through the CTC model, and then the recognition characters are circularly combined and screened according to the character probability to generate a target character sequence, so that the calculated amount of character recognition is reduced, the character recognition efficiency is improved, the accuracy of character recognition is improved, and the technical problem that the accuracy of English recognition is low when the definition of the existing picture is low is solved.
Drawings
FIG. 1 is a schematic diagram of a hardware architecture of a probability matrix-based character recognition device according to an embodiment of the present invention;
FIG. 2 is a flow chart of a first embodiment of a character recognition method based on a probability matrix according to the present invention;
fig. 3 is a schematic functional block diagram of a first embodiment of a character recognition device based on a probability matrix according to the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The character recognition method based on the probability matrix is mainly applied to the character recognition equipment based on the probability matrix, and the character recognition equipment based on the probability matrix can be equipment with display and processing functions such as a PC, a portable computer and a mobile terminal.
Referring to fig. 1, fig. 1 is a schematic hardware structure of a character recognition apparatus based on a probability matrix according to an embodiment of the present invention. In an embodiment of the present invention, the probability matrix based character recognition apparatus may include a processor 1001 (e.g., CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein the communication bus 1002 is used to enable connected communications between these components; the user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard); the network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface); the memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory, and the memory 1005 may alternatively be a storage device independent of the processor 1001.
Those skilled in the art will appreciate that the hardware architecture shown in fig. 1 does not constitute a limitation of the probability matrix based character recognition device, and may include more or fewer components than illustrated, or may combine certain components, or a different arrangement of components.
With continued reference to fig. 1, the memory 1005 in fig. 1, which is a computer-readable storage medium, may include an operating system, a network communication module, and a probability matrix-based character recognition program.
In fig. 1, the network communication module is mainly used for connecting with a server and performing data communication with the server; and the processor 1001 may call the character recognition program based on the probability matrix stored in the memory 1005 and execute the character recognition method based on the probability matrix provided in the embodiment of the present invention.
The embodiment of the invention provides a character recognition method based on a probability matrix.
Referring to fig. 2, fig. 2 is a flowchart of a first embodiment of a character recognition method based on a probability matrix according to the present invention.
In this embodiment, the character recognition method based on the probability matrix includes the following steps:
step S10, inputting a target image into a preset CTC model, and generating a recognition character set and a character probability matrix corresponding to the target image, wherein the recognition character set comprises recognition characters corresponding to all positions in the target image, one position corresponds to one recognition character subset, one recognition character corresponds to one probability value, and the character probability matrix comprises all probability values corresponding to the recognition character set;
At present, the probability of the obtained character is not necessarily the correct probability distribution due to the reasons of training data, picture definition and the like. For example v- > r, t- > f, etc., are often similar characters when handwriting. Some a priori knowledge is added to assist in decoding to help get the correct character. The invention is mainly used for analyzing English characters, and priori knowledge used for analyzing English generally has a vocabulary and a word level language model. The vocabulary helps to parse out correct words rather than non-existent words, and the language model helps to choose more inter-word combinations on inter-word relationships, e.g., look forward go- > look forward to. The decoding method for the probability distribution of the sequence characters mainly comprises the following steps:
greeny decoding method: the letters with the highest probability of each position are directly selected, so that the method is rough and rapid. But does not take any a priori information into account, the probability of error is large.
Beam search: when the sequence is decoded, a certain number of search beams which are not repeated are generated, the search beams are decoded backwards step by step, when the search beams extend, the search beams with a certain number of search beams with high probability are selected to continue to extend backwards, a character-level language model can be added in the decoding process, the probability is corrected, and finally the optimal search beam, namely the search beam with the highest probability, is found out from the search beams
TokenPasing, referring to the beamlsearch decoding mode, generating words according to the vocabulary, but generating only word output, if the words are non-words, the words can not be output. This is highly inapplicable in some situations, such as the common non-alphabetic string- "1234", and also in examination situations, the wrongly written words should not be corrected and should not be entirely vocabulary words.
Wordbeamsearch: the current widely used and accepted scheme is more skilled when the BeamSearch adds a vocabulary and word level language model, it is lexically enough to a character tree, the tree can inquire the possibility of the following letters according to the prefix words, for example, the words with the prefix of abdom are abdomen, abdominal, the following letters with the prefix of abdom are [ e, i ], and the word level language model probability can be calculated simultaneously when the words with non-letters are encountered. This approach solves the problem of inability to generate non-vocabulary words in TokenPassing, but does not solve the problem thoroughly enough because no prefix unpaired results can be generated, otherwise decoding behind the prefix becomes abnormal, and the tail letters have no character tree constraints.
In order to solve the above problems, the embodiment outputs the character probability matrix of the target image through the CTC model, and then performs cyclic combination screening on the recognition characters according to the character probability to generate the target character sequence, thereby reducing the calculation amount of character recognition, improving the character recognition efficiency and improving the accuracy of character recognition. Specifically, the image features of the target image to be identified may be extracted in advance, and the image features may be input to a CTC model after training, that is, the identified characters corresponding to each position of the target image and the probability distribution of each identified character are output through a CTC neural network, that is, a 200×96 matrix is output, which means the probability distribution of 96 characters on each of 200 positions. The 96 characters are divided into case word characters and non-word characters, wherein the word characters are characters used for forming words, such as abcd, and the non-word characters are symbol characters, such as commas, periods, quotation marks, and the like. The method comprises the steps of training a deep learning training English character recognition model according to an RCNN network structure and a CTC loss function, namely calculating character probability distribution of all moments (positions) mainly through a CNN convolution layer and an RNN circulation layer, calculating character probability loss values of all moments (positions) through a CTC mode, then back-propagating the loss values, correcting network parameters, and finally outputting characters and character probability distribution of each position correctly through a neural network.
It should be emphasized that, to further ensure the privacy and security of the identified character set and character probability matrix information, the identified character set and character probability matrix information may also be stored in nodes of a blockchain.
Step S20, sequentially acquiring a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sequencing each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from big to small, acquiring recognition characters arranged in a preset number, and generating a first alternative character set and the second alternative character set, wherein the first position is the leftmost character position of the target image, and the second position is the leftmost next character position;
in this embodiment, in order to facilitate subsequent character stitching, after english character recognition is performed on the target image, the recognition characters are stored in a left-to-right order according to the order, where the first position is the leftmost character position of the target image, the second position is the leftmost next character position, a first recognition character subset corresponding to the first position is sequentially obtained, each character in the first recognition character subset is ordered according to the order of the character probability from large to small, a preset number of recognition characters arranged in front are obtained as first candidate characters, and a first candidate character set is generated; and acquiring the next position of the first position as a second position, acquiring a second recognition character subset corresponding to the second position, sequencing all characters in the second recognition character subset according to the sequence from the big probability to the small probability, acquiring the recognition characters arranged in the preset number as second candidate characters, and generating a second candidate character set. Specifically, a search bundle is initialized as Beams, the search bundle being used to hold a plurality of recognition characters that are currently resolved. A set of identification characters corresponding to a location is stored in a search beam. A location is stored in a Beam of the search bundle corresponding to an identification character, which Beam can be used to store information such as the identification character, character string, word, probability, etc. Since the search space cannot be excessively large, it is assumed that the number of Beams in Beams is maximum bwoidth, and the larger bwoidth, the more comprehensive the search, but the larger the calculation amount. Traversing the character probability distribution in each search bundle, taking the probability distribution of 96 characters each time, taking the previous bwoidth characters and their probabilities, and initializing a temporary blank beam as a newbeam for storing the combined character string. For example, a target image with hello is identified, the character at the first position may be any one of 96 characters, the probability corresponding to the 96 characters is obtained through CTC, and according to the searched bwoidth, for example, 5, the character with the probability of 5 is obtained from the 96 characters, and is stored in each Beam of the search Beam.
Step S30, traversing each first alternative character in the first alternative character set and each second alternative character in the second alternative character set, combining each first alternative character with each second alternative character to generate an alternative character sequence set, determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence, determining a first to-be-determined character sequence according to the sequence probability of the alternative character sequence, and generating a first to-be-determined character sequence set;
in this embodiment, each first alternative character in the first alternative character set is traversed, each second alternative character in the second alternative character set is traversed, and then each first alternative character and each second alternative character are combined in sequence to generate a plurality of alternative character sequences. For example, if there are 5 first candidate characters and 5 second candidate characters, then the combined results in 25 are generated, where each combined result is that the first candidate character in the first position is before and the second candidate character in the second position is after. I.e. the recognition character of the second position follows the recognition character of the first position, the recognition character of the third position follows the recognition character of the second position, and so on. Determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence; sequencing each character sequence in the alternative character sequence set according to the sequence probability from high to low; and acquiring character sequences arranged in the preset number as a first character sequence to be determined, and generating a first character sequence set to be determined.
Step S40, sequentially obtaining the next position corresponding to the second position, as a third position, sequencing and fetching the recognition character subset corresponding to the third position to generate a third alternative character set, and combining the first to generate a second to-be-determined character sequence set until the combination of the recognition characters of all positions is completed to generate a related to-be-determined character sequence set;
in this embodiment, according to the above process, the recognition character subsets at each position of the target image are screened to generate each candidate character set with a larger probability, then the candidate character set corresponding to the third position is combined with the undetermined character sequences generated by the first position and the second position, and the undetermined character sequence sets after being combined are screened according to the probability and are further combined with the candidate character set at the next position. And the combination screening is circulated until the combination of the identification characters at all positions is completed, and a relevant undetermined character sequence set is generated.
And S50, determining the relevant undetermined character sequence with the largest relevant sequence probability as a target character sequence in the relevant undetermined character sequence set according to the relevant sequence probability of the relevant character sequence contained in each relevant undetermined character sequence, and outputting an analysis result of the target character sequence to finish the recognition of the target image.
In this embodiment, according to the correlation sequence probability of the correlation character sequence included in each correlation pending character sequence, the correlation pending character sequence with the largest correlation sequence probability is determined in the correlation pending character sequence set to be the target character sequence, and the analysis result of the target character sequence is output.
The invention provides a character recognition method based on a probability matrix, which outputs recognition characters at all positions of a target image and probability matrixes corresponding to all recognition characters through a CTC model, screens the recognition characters corresponding to all positions, combines the screened recognition characters into character sequences, circularly screens and combines the character sequences to extract relevant undetermined character sequence sets, and determines the relevant undetermined character sequence with the largest probability value in the relevant undetermined character sequence sets as a target character sequence to obtain the recognition characters of the target image. The character probability matrix of the target image is output through the CTC model, and then the recognition characters are circularly combined and screened according to the character probability to generate a target character sequence, so that the calculated amount of character recognition is reduced, the character recognition efficiency is improved, the accuracy of character recognition is improved, and the technical problem that the accuracy of English recognition is low when the definition of the existing picture is low is solved.
Based on the embodiment shown in fig. 2, a second embodiment of a character recognition method based on a probability matrix is provided in this embodiment, and after step S40, the method further includes:
sequentially acquiring a third recognition character subset corresponding to a third position, sequencing all characters in the third recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as third alternative characters, and generating a third alternative character set;
acquiring a first character sequence to be determined in the first character sequence set as a current first character sequence to be determined, and acquiring a third alternative character in the third alternative character set as the current third alternative character;
judging whether the current third alternative character is an English character or not;
if the current third alternative character is not an English character, judging whether the current first character sequence to be determined is a recognizable word according to a preset vocabulary;
and if the current first waiting character sequence is a recognizable word, compensating the current first waiting character sequence, and storing a word analysis result corresponding to the current first waiting character sequence.
If the current third alternative character is an English character, combining the current first undetermined character sequence with the current third alternative character to generate a second undetermined character sequence;
and acquiring the next third alternative character in the third alternative character set as the current third alternative character until each first alternative character sequence in the first alternative character sequence set and each third alternative character in the third alternative character set are combined to generate a second alternative character sequence set.
Further, after the step of determining whether the current first predetermined character sequence is a recognizable word according to a preset vocabulary if the current third candidate character is not an english character, the method further includes:
if the current first character sequence is not the recognizable word, storing the current first character sequence as a character string to be recognized into a data table;
determining the target number according to the word processing number corresponding to a preset language model, sequentially acquiring character strings to be recognized of the target number in the data table, and inputting the character strings to be recognized of the target number into the language model so as to calculate continuous word relation probability of the character strings to be recognized of the target number through the language model;
And when the continuous word relation probability is larger than a preset threshold value, storing continuous word analysis results corresponding to the character strings to be identified of the target number.
In this embodiment, a search bundle is initialized as Beams, and the search bundle is used to save a plurality of recognition characters that are currently resolved. A set of identification characters corresponding to a location is stored in a search beam. A location is stored in a Beam of the search bundle corresponding to an identification character, which Beam can be used to store information such as the identification character, character string, word, probability, etc. Since the search space cannot be excessively large, it is assumed that the number of Beams in Beams is maximum bwoidth, and the larger bwoidth, the more comprehensive the search, but the larger the calculation amount. Traversing the character probability distribution in each search bundle, taking the probability distribution of 96 characters each time, taking the previous bwoidth characters and their probabilities, and initializing a temporary blank beam as a newbeam for storing the combined character string. I.e. traversing the corresponding alternative character set of each position and combining each recognized character in the alternative character set of the subsequent position with the recognized character in the alternative character set of the previous position or with the pending character sequence of the previous position. While traversing the candidate character set of the next position, determining whether each candidate character of the next position is a word character, if so, adding this character to all Beams in the Beams, and adding these processed Beams to the NewBeams by multiplying the probability of the Beam by the probability of the current character. If it is not a word character, it is represented as a symbol, representing that the preceding character in the Beam may constitute a word, and the following is performed for all Beams:
a) Checking whether the last character of the analyzed character of the Beam is also a non-word character, if so, directly jumping to the step e, because the correction of the vocabulary and the language model is already carried out, otherwise, sequentially carrying out the subsequent steps;
b) The character string analyzed by the Beam is divided according to the non-word characters, and one or more words are generated;
c) Taking the last word, checking if the word is in the vocabulary, multiplying the Beam probability by a compensation, e.g., 10, if in the vocabulary, and not if not in the vocabulary;
d) Taking N words, where N represents the number of words that the language model can handle, for example we use a simple 2-gram language model, then n=2. Putting N words into the language model calculates its continuous word relationship probability P. Converting the probability into a compensation, e.g. multiplying the probability of Beam by (1+p×10);
e) Finally, adding characters into the Beam, multiplying the probability after compensating the Beam by the probability of the current character, and adding the processed Beam into the Newbeams.
When character combination of a position is completed, namely, a set of pending sequences is generated, the set of pending sequences is screened and filtered according to the probability size, so that the calculation amount is saved, and as many possibilities as possible are saved in a limited space. The filtering method is to combine the tail N words of each Beam as keys, and the beams with the same keys only keep the one with the highest probability, so as to greatly reduce the number of the beams in decoding.
5. The resulting NewBeams were taken the previous bwodth Beams and replaced with the previous Beams.
6. After the loop combination screening is completed, the remaining Beams need word number compensation. The reason is that each time a word is resolved during resolving, probability compensation is performed, so that more words are resolved, the more the compensation is, so that in the re-resolving process, a combined word, such as a home work, tends to resolve both home and work probabilistically, because there are two probability compensations. And finally counting the maximum number of Beam words MaxWordsNum in all Beams, taking the maximum number of the Beams as a reference, wherein the number of the Beams is 1 less, the probability of the Beam is multiplied by the probability compensation of the words, and finally taking the Beam with the maximum probability as the best analysis result.
The present embodiment does not force the vocabulary, but provides some probability trends when generating a word, because the search bundles that parse the word at parsing make additional probability compensation, but does not force the word generation, and all can generate any non-vocabulary words, thus avoiding the disadvantages of the token pass method and the lack of constraint on the tail characters like the word tree of WordBeamSearch. In addition, when non-word characters are encountered, namely the front character string is possibly a word, the vocabulary retrieval is performed, and each character is not queried in a character tree like WordBeamSearch, so that the retrieval times are greatly reduced, and the execution efficiency is improved. The embodiment further considers the influence scope of the language model, because the language model can only influence the information of a certain word quantity and length during analysis, for example, the 2-gram language model can only influence the first two words of the current position, the probability of the previous characters is fixed, the search bundles with the same influence scope are combined in time, only the search bundles with the highest probability are reserved, the extension of useless search bundles is reduced, the calculated amount is reduced, space is saved under the condition that the quantity of the reserved search bundles is limited, and the search domain is enriched. In addition, probability compensation is performed when words are generated during analysis, and the probability is higher as more words are generated during analysis, so that more words are prone to decoding during decoding, probability correction is performed through the number of words in the method, and the accuracy of the probability is guaranteed.
In addition, the embodiment of the invention also provides a character recognition device based on the probability matrix.
Referring to fig. 3, fig. 3 is a schematic functional block diagram of a first embodiment of a character recognition device based on a probability matrix according to the present invention.
In this embodiment, the character recognition device based on the probability matrix includes:
the character probability acquisition module 10 is configured to input a target image into a preset CTC model, and generate a recognition character set and a character probability matrix corresponding to the target image, where the recognition character set includes recognition characters corresponding to each position in the target image, one position corresponds to a recognition character subset, one recognition character corresponds to a probability value, and the character probability matrix includes all probability values corresponding to the recognition character set;
the candidate character set generating module 20 is configured to sequentially obtain a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sort each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from high to low, obtain a preset number of recognition characters arranged in front, and generate a first candidate character set and a second candidate character set;
A to-be-determined sequence set generating module 30, configured to traverse each first candidate character in the first candidate character set and each second candidate character in the second candidate character set, combine each first candidate character with each second candidate character to generate a candidate character sequence set, determine a sequence probability of each candidate character sequence according to a character probability of each candidate character contained in the candidate character sequence, determine a first to-be-determined character sequence according to the sequence probability of the candidate character sequence, and generate a first to-be-determined character sequence set;
a related sequence set generating module 40, configured to sequentially obtain a next position corresponding to the second position, as a third position, sort and fetch the recognized character subsets corresponding to the third position to generate a third alternative character set, and combine the first to-be-determined character sequence set with the third alternative character set to generate a second to-be-determined character sequence set until the combination of the recognized characters in all positions is completed, so as to generate a related to-be-determined character sequence set;
the recognition character output module 50 is configured to determine, in the set of related pending character sequences, a related pending character sequence with a maximum probability of a related sequence as a target character sequence according to the related sequence probabilities of the related character sequences included in the related pending character sequences, and output an analysis result of the target character sequence, thereby completing recognition of the target image.
Further, the related sequence set generating module 40 is further configured to:
sequentially acquiring a third recognition character subset corresponding to a third position, sequencing all characters in the third recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as third alternative characters, and generating a third alternative character set;
acquiring a first character sequence to be determined in the first character sequence set as a current first character sequence to be determined, and acquiring a third alternative character in the third alternative character set as the current third alternative character;
judging whether the current third alternative character is an English character or not;
if the current third alternative character is not an English character, judging whether the current first character sequence to be determined is a recognizable word according to a preset vocabulary;
and if the current first waiting character sequence is a recognizable word, compensating the current first waiting character sequence, and storing a word analysis result corresponding to the current first waiting character sequence.
Further, the related sequence set generating module 40 is further configured to:
if the current third alternative character is an English character, combining the current first undetermined character sequence with the current third alternative character to generate a second undetermined character sequence;
And acquiring the next third alternative character in the third alternative character set as the current third alternative character until each first alternative character sequence in the first alternative character sequence set and each third alternative character in the third alternative character set are combined to generate a second alternative character sequence set.
Further, the related sequence set generating module 40 is further configured to:
if the current first character sequence is not the recognizable word, storing the current first character sequence as a character string to be recognized into a data table;
determining the target number according to the word processing number corresponding to a preset language model, sequentially acquiring character strings to be recognized of the target number in the data table, and inputting the character strings to be recognized of the target number into the language model so as to calculate continuous word relation probability of the character strings to be recognized of the target number through the language model;
and when the continuous word relation probability is larger than a preset threshold value, storing continuous word analysis results corresponding to the character strings to be identified of the target number.
Further, the candidate character set generating module 20 is further configured to:
sequentially acquiring a first recognition character subset corresponding to a first position, sequencing each character in the first recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as first alternative characters, and generating a first alternative character set;
And acquiring the next position of the first position as a second position, acquiring a second recognition character subset corresponding to the second position, sequencing all characters in the second recognition character subset according to the sequence from the big probability to the small probability, acquiring the recognition characters arranged in the preset number as second candidate characters, and generating a second candidate character set.
Further, the pending sequence set generation module 30 is further configured to:
determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence;
sequencing each character sequence in the alternative character sequence set according to the sequence probability from high to low;
and acquiring character sequences arranged in the preset number as a first character sequence to be determined, and generating a first character sequence set to be determined.
Wherein, each module in the character recognition device based on the probability matrix corresponds to each step in the character recognition method embodiment based on the probability matrix, and the functions and implementation processes thereof are not repeated here.
The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.
In addition, the embodiment of the invention also provides a computer readable storage medium.
The computer readable storage medium of the present invention stores a character recognition program based on a probability matrix, wherein the character recognition program based on the probability matrix realizes the steps of the character recognition method based on the probability matrix when being executed by a processor.
The method implemented when the character recognition program based on the probability matrix is executed may refer to various embodiments of the character recognition method based on the probability matrix according to the present invention, which are not described herein.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (10)
1. The character recognition method based on the probability matrix is characterized by comprising the following steps of:
inputting a target image into a preset CTC model to generate a recognition character set and a character probability matrix corresponding to the target image, wherein the recognition character set comprises recognition characters corresponding to all positions in the target image, one position corresponds to one recognition character subset, one recognition character corresponds to one probability value, and the character probability matrix comprises all probability values corresponding to the recognition character set;
sequentially acquiring a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sequencing each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from big to small, acquiring a preset number of recognition characters arranged in front, and generating a first alternative character set and a second alternative character set, wherein the first position is the leftmost character position of the target image, and the second position is the leftmost next character position;
traversing each first alternative character in the first alternative character set and each second alternative character in the second alternative character set, combining each first alternative character with each second alternative character to generate an alternative character sequence set, determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence, determining a first to-be-determined character sequence according to the sequence probability of the alternative character sequence, and generating a first to-be-determined character sequence set;
Sequentially acquiring the next position corresponding to the second position as a third position, sequencing and fetching the recognition character subset corresponding to the third position to generate a third alternative character set, and combining the first to generate a second to-be-determined character sequence set until the combination of recognition characters of all positions is completed to generate a related to-be-determined character sequence set;
and determining the relevant undetermined character sequence with the largest relevant sequence probability as a target character sequence in the relevant undetermined character sequence set according to the relevant sequence probability of the relevant character sequence contained in each relevant undetermined character sequence, and outputting an analysis result of the target character sequence to finish the recognition of the target image.
2. The method for recognizing characters based on a probability matrix according to claim 1, wherein the step of sorting and fetching the recognized character subset corresponding to the third position to generate a third alternative character set, and combining the first predetermined character sequence set with the third alternative character set to generate a second predetermined character sequence set specifically comprises:
Sequentially acquiring a third recognition character subset corresponding to a third position, sequencing all characters in the third recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as third alternative characters, and generating a third alternative character set;
acquiring a first character sequence to be determined in the first character sequence set as a current first character sequence to be determined, and acquiring a third alternative character in the third alternative character set as the current third alternative character;
judging whether the current third alternative character is an English character or not;
if the current third alternative character is not an English character, judging whether the current first character sequence to be determined is a recognizable word according to a preset vocabulary;
and if the current first waiting character sequence is a recognizable word, compensating the current first waiting character sequence, and storing a word analysis result corresponding to the current first waiting character sequence.
3. The method for character recognition based on probability matrix as claimed in claim 2, wherein after the step of determining whether the current third alternative character is an english character, further comprising:
If the current third alternative character is an English character, combining the current first undetermined character sequence with the current third alternative character to generate a second undetermined character sequence;
and acquiring the next third alternative character in the third alternative character set as the current third alternative character until each first alternative character sequence in the first alternative character sequence set and each third alternative character in the third alternative character set are combined to generate a second alternative character sequence set.
4. The method for recognizing characters based on probability matrix as claimed in claim 2, wherein if the current third candidate character is not an english character, the step of judging whether the current first predetermined character sequence is a recognizable word according to a preset vocabulary further comprises:
if the current first character sequence is not the recognizable word, storing the current first character sequence as a character string to be recognized into a data table;
determining the target number according to the word processing number corresponding to a preset language model, sequentially acquiring character strings to be recognized of the target number in the data table, and inputting the character strings to be recognized of the target number into the language model so as to calculate continuous word relation probability of the character strings to be recognized of the target number through the language model;
And when the continuous word relation probability is larger than a preset threshold value, storing continuous word analysis results corresponding to the character strings to be identified of the target number.
5. The method for recognizing characters based on probability matrix as claimed in claim 1, wherein the steps of sequentially obtaining a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sorting each character in the first recognition character subset and the second recognition character subset according to the order of the character probability from big to small, obtaining recognition characters arranged in a preset number, and generating a first candidate character set and the second candidate character set specifically comprise:
sequentially acquiring a first recognition character subset corresponding to a first position, sequencing each character in the first recognition character subset according to the sequence of the character probability from large to small, acquiring the recognition characters with the preset number arranged in front as first alternative characters, and generating a first alternative character set;
and acquiring the next position of the first position as a second position, acquiring a second recognition character subset corresponding to the second position, sequencing all characters in the second recognition character subset according to the sequence from the big probability to the small probability, acquiring the recognition characters arranged in the preset number as second candidate characters, and generating a second candidate character set.
6. The character recognition method based on probability matrix according to claim 1, wherein the recognition character set corresponding to the target image and the character probability matrix are stored in a blockchain.
7. The method for recognizing characters based on probability matrix according to any one of claims 1 to 6, wherein the step of determining the sequence probability of each alternative character sequence based on the character probability of the alternative character contained in the alternative character sequence, determining the first predetermined character sequence based on the sequence probability of the alternative character sequence, and generating the first predetermined character sequence set specifically comprises:
determining the sequence probability of each alternative character sequence according to the character probability of the alternative character contained in each alternative character sequence;
sequencing each character sequence in the alternative character sequence set according to the sequence probability from high to low;
and acquiring character sequences arranged in the preset number as a first character sequence to be determined, and generating a first character sequence set to be determined.
8. A probability matrix based character recognition apparatus, comprising:
the character probability acquisition module is used for inputting a target image into a preset CTC model and generating a recognition character set and a character probability matrix corresponding to the target image, wherein the recognition character set comprises recognition characters corresponding to all positions in the target image, one position corresponds to one recognition character subset, one recognition character corresponds to one probability value, and the character probability matrix comprises all probability values corresponding to the recognition character set;
The candidate character set generation module is used for sequentially acquiring a first recognition character subset corresponding to a first position and a second recognition character subset corresponding to a second position, sequencing each character in the first recognition character subset and the second recognition character subset according to the sequence of the character probability from big to small, acquiring recognition characters arranged in a preset number, and generating a first candidate character set and a second candidate character set, wherein the first position is the leftmost character position of the target image, and the second position is the leftmost next character position;
the device comprises a to-be-determined sequence set generation module, a first set of first candidate characters and a second set of second candidate characters, wherein the to-be-determined sequence set generation module is used for traversing each first candidate character in the first candidate character set and each second candidate character in the second candidate character set, combining each first candidate character with each second candidate character to generate a candidate character sequence set, determining the sequence probability of each candidate character sequence according to the character probability of the candidate character contained in each candidate character sequence, determining a first to-be-determined character sequence according to the sequence probability of the candidate character sequence, and generating a first to-be-determined character sequence set;
The related sequence set generating module is used for sequentially acquiring the next position corresponding to the second position, serving as a third position, sequencing and fetching the recognition character subset corresponding to the third position to generate a third alternative character set, combining the first undetermined character sequence set with the third alternative character set to generate a second undetermined character sequence set until the combination of the recognition characters of all the positions is completed, and generating the related undetermined character sequence set;
and the identification character output module is used for determining the relevant undetermined character sequence with the largest relevant sequence probability as a target character sequence in the relevant undetermined character sequence set according to the relevant sequence probability of the relevant character sequence contained in each relevant undetermined character sequence, and outputting an analysis result of the target character sequence to finish the identification of the target image.
9. A probability matrix based character recognition device comprising a processor, a memory, and a probability matrix based character recognition program stored on the memory and executable by the processor, wherein the probability matrix based character recognition program, when executed by the processor, implements the steps of the probability matrix based character recognition method of any one of claims 1 to 7.
10. A computer-readable storage medium, on which a probability matrix based character recognition program is stored, wherein the probability matrix based character recognition program, when executed by a processor, implements the steps of the probability matrix based character recognition method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010929789.4A CN112232347B (en) | 2020-09-07 | 2020-09-07 | Character recognition method, device, equipment and storage medium based on probability matrix |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010929789.4A CN112232347B (en) | 2020-09-07 | 2020-09-07 | Character recognition method, device, equipment and storage medium based on probability matrix |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112232347A CN112232347A (en) | 2021-01-15 |
| CN112232347B true CN112232347B (en) | 2023-06-20 |
Family
ID=74116679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010929789.4A Active CN112232347B (en) | 2020-09-07 | 2020-09-07 | Character recognition method, device, equipment and storage medium based on probability matrix |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112232347B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016155627A1 (en) * | 2015-04-03 | 2016-10-06 | 北京奇虎科技有限公司 | Method and apparatus for recognizing characters in picture |
| CN108288078A (en) * | 2017-12-07 | 2018-07-17 | 腾讯科技（深圳）有限公司 | Character identifying method, device and medium in a kind of image |
| CN110796092A (en) * | 2019-10-30 | 2020-02-14 | 北京一起教育信息咨询有限责任公司 | Character recognition method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599900B (en) * | 2015-10-20 | 2020-04-21 | 华中科技大学 | A method and apparatus for identifying character strings in images |
-
2020
- 2020-09-07 CN CN202010929789.4A patent/CN112232347B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016155627A1 (en) * | 2015-04-03 | 2016-10-06 | 北京奇虎科技有限公司 | Method and apparatus for recognizing characters in picture |
| CN108288078A (en) * | 2017-12-07 | 2018-07-17 | 腾讯科技（深圳）有限公司 | Character identifying method, device and medium in a kind of image |
| CN110796092A (en) * | 2019-10-30 | 2020-02-14 | 北京一起教育信息咨询有限责任公司 | Character recognition method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112232347A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111814466B (en) | Information extraction method based on machine reading understanding and related equipment thereof | |
| CN110489760B (en) | Text automatic correction method and device based on deep neural network | |
| CN111310443B (en) | Text error correction method and system | |
| EP2585962B1 (en) | Password checking | |
| WO2018040899A1 (en) | Error correction method and device for search term | |
| CN111324743A (en) | Text relation extraction method and device, computer equipment and storage medium | |
| CN109766538B (en) | Text error correction method and device, electronic equipment and storage medium | |
| CN110795938B (en) | Text sequence word segmentation method, device and storage medium | |
| CN112329447B (en) | Training method of Chinese error correction model, chinese error correction method and device | |
| CN112613293B (en) | Digest generation method, digest generation device, electronic equipment and storage medium | |
| TWI567569B (en) | Natural language processing systems, natural language processing methods, and natural language processing programs | |
| CN111506709B (en) | Entity linking method and device, electronic equipment and storage medium | |
| CN113961768B (en) | Sensitive word detection method and device, computer equipment and storage medium | |
| CN111178049A (en) | Text correction method and device, readable medium and electronic equipment | |
| CN112861519B (en) | Medical text error correction method, device and storage medium | |
| CN111782892B (en) | Similar character recognition method, device, apparatus and storage medium based on prefix tree | |
| CN117291149A (en) | Rarely used word decoding method and device, electronic equipment and storage medium | |
| CN112508102B (en) | Text recognition method, device, storage medium and computer equipment | |
| CN110929514B (en) | Text collation method, text collation apparatus, computer-readable storage medium, and electronic device | |
| CN112232347B (en) | Character recognition method, device, equipment and storage medium based on probability matrix | |
| CN111310457B (en) | Word mismatching recognition method and device, electronic equipment and storage medium | |
| CN111985486A (en) | Image information identification method and device, storage medium and computer equipment | |
| CN113723096B (en) | Text recognition method and device, computer-readable storage medium and electronic device | |
| CN116484844A (en) | Document OCR recognition result error correction method, system, equipment and medium | |
| CN112733530B (en) | Text punctuation correction method, device, electronic device and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |