CN110490180A - Work correction method, apparatus, storage medium and server based on image recognition - Google Patents
Work correction method, apparatus, storage medium and server based on image recognition Download PDFInfo
- Publication number
- CN110490180A CN110490180A CN201910603800.5A CN201910603800A CN110490180A CN 110490180 A CN110490180 A CN 110490180A CN 201910603800 A CN201910603800 A CN 201910603800A CN 110490180 A CN110490180 A CN 110490180A
- Authority
- CN
- China
- Prior art keywords
- topic
- image
- region
- papers
- network
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to field of computer technology more particularly to a kind of work correction method, apparatus, computer readable storage medium and servers based on image recognition.The work correction that the method receiving terminal apparatus is sent is requested, and includes the pending image corrected students' papers and the pending job identification corrected students' papers in the work correction request;Determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag;Using preset deep learning network respectively to the topic of each topic answer region image carry out image recognition, obtain the characteristics of image of each topic;The characteristics of image of each topic is handled respectively using preset serializing model, obtains the content of answering of each topic;The model answer of each topic is determined according to the pending job identification corrected students' papers, and is compared according to answer content of the model answer of each topic to each topic, and work correction result is obtained.
Description
Technical field
The invention belongs to field of computer technology more particularly to a kind of work correction method, apparatus based on image recognition,
Computer readable storage medium and server.
Background technique
Currently, the operation that student submits generally requires teacher and corrects by hand, sizable work is brought to teacher
Amount, especially when the student in class is more, needs teacher except normal teaching, additionally extracts a large amount of time out
Work correction is carried out, under efficiency is very low.There is through machine-read card the scheme for improving work correction efficiency in the prior art, it is machine-readable
Card is a kind of simple optical character identification (OCR) technology, and optical character reader (OCR) (card reader) is only sensitive to black, is printed on answering card
There is the stick of black, the direction and position for allowing card reader confirmation to block, the black patch and printed black patch of full-filling of the pencil on card are total to
It is same to constitute the image for there was only black and white.It is approximate with binary " 0 ", " 1 ", card reader scanning after with pre-stored letter
The image that breath generates is compared, to obtain result.Although this mode is also able to achieve correcting automatically to operation, but
There is two apparent defects: first, need the dedicated answering card of additional lithography steps, higher cost;Second, it can only on answering card
It records binary message (black patch for whether having pencil full-filling), often may be only available for answering to multiple-choice question, and it is uncomfortable
Scene for more text informations such as gap-filling questions, simple answers.
Summary of the invention
In view of this, the work correction method, apparatus that the embodiment of the invention provides a kind of based on image recognition, computer
Readable storage medium storing program for executing and server, to solve come higher cost existing when carrying out work correction and not being suitable for by machine-read card
The problem of scenes of more text information such as gap-filling questions, simple answer.
The first aspect of the embodiment of the present invention provides a kind of work correction method based on image recognition, may include:
The work correction that receiving terminal apparatus is sent is requested, and includes the pending image corrected students' papers in the work correction request
And the pending job identification corrected students' papers;
Determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag;
Using preset deep learning network respectively to the topic of each topic answer region image carry out image recognition,
Obtain the characteristics of image of each topic;
The characteristics of image of each topic is handled respectively using preset serializing model, obtains the work of each topic
Content is answered, the serializing model is formed by more than two gating cycles are unit cascaded;
The model answer of each topic is determined according to the pending job identification corrected students' papers, and according to the mark of each topic
The content of answering of each topic is compared in quasi- answer, obtains work correction result.
The second aspect of the embodiment of the present invention provides a kind of work correction device, may include:
Work correction request receiving module, for the work correction request that receiving terminal apparatus is sent, the work correction
It include the pending image corrected students' papers and the pending job identification corrected students' papers in request;
Topic is answered area determination module, for according to preset sign flag in the pending image corrected students' papers really
The topic of fixed each topic is answered region;
Image characteristics extraction module, for being answered area to the topic of each topic respectively using preset deep learning network
The image in domain carries out image recognition, obtains the characteristics of image of each topic;
Serialize processing module, for using preset serializing model respectively to the characteristics of image of each topic at
Reason, obtains the content of answering of each topic, the serializing model is formed by more than two gating cycles are unit cascaded;
Content of answering comparison module, for determining that the standard of each topic is answered according to the pending job identification corrected students' papers
Case, and be compared according to answer content of the model answer of each topic to each topic, obtain work correction result.
The third aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer-readable instruction, and the computer-readable instruction realizes following steps when being executed by processor:
The work correction that receiving terminal apparatus is sent is requested, and includes the pending image corrected students' papers in the work correction request
And the pending job identification corrected students' papers;
Determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag;
Using preset deep learning network respectively to the topic of each topic answer region image carry out image recognition,
Obtain the characteristics of image of each topic;
The characteristics of image of each topic is handled respectively using preset serializing model, obtains the work of each topic
Content is answered, the serializing model is formed by more than two gating cycles are unit cascaded;
The model answer of each topic is determined according to the pending job identification corrected students' papers, and according to the mark of each topic
The content of answering of each topic is compared in quasi- answer, obtains work correction result.
The fourth aspect of the embodiment of the present invention provides a kind of server, including memory, processor and is stored in institute
The computer-readable instruction that can be run in memory and on the processor is stated, the processor executes described computer-readable
Following steps are realized when instruction:
The work correction that receiving terminal apparatus is sent is requested, and includes the pending image corrected students' papers in the work correction request
And the pending job identification corrected students' papers;
Determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag;
Using preset deep learning network respectively to the topic of each topic answer region image carry out image recognition,
Obtain the characteristics of image of each topic;
The characteristics of image of each topic is handled respectively using preset serializing model, obtains the work of each topic
Content is answered, the serializing model is formed by more than two gating cycles are unit cascaded;
The model answer of each topic is determined according to the pending job identification corrected students' papers, and according to the mark of each topic
The content of answering of each topic is compared in quasi- answer, obtains work correction result.
Existing beneficial effect is the embodiment of the present invention compared with prior art: without special in the embodiment of the present invention
Answering card, and preset sign flag need to be only printed in operation, in the job batch for receiving terminal device transmission
After changing request, it can determine that the topic of each topic is answered in the pending image corrected students' papers according to these sign flags
Region greatly reduces the cost of work correction.And distinguished in embodiments of the present invention using preset deep learning network
To the topic of each topic answer region image carry out image recognition, obtain the characteristics of image of each topic, reuse default
Serializing model the characteristics of image of each topic is handled respectively, can effectively capture word, word, the symbol etc. in topic
Dependence between equal text elements, obtains the content of answering of each topic, each for multiple-choice question, gap-filling questions, simple answer etc.
Kind scene is applicable, finally only needs for preset model answer content of answering to be compared, final operation can be obtained
Correct result.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description is only of the invention some
Embodiment for those of ordinary skill in the art without any creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 is a kind of one embodiment flow chart of the work correction method based on image recognition in the embodiment of the present invention;
Fig. 2 is to determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag
Schematic flow diagram;
Fig. 3 is the schematic diagram of the operation comprising a variety of special sign flags;
Fig. 4 is that the topic in the region and each topic where determining each topic in the pending image corrected students' papers is answered
The schematic diagram in region;
Fig. 5 is a kind of one embodiment structure chart of work correction device in the embodiment of the present invention;
Fig. 6 is a kind of schematic block diagram of server in the embodiment of the present invention.
Specific embodiment
In order to make the invention's purpose, features and advantages of the invention more obvious and easy to understand, below in conjunction with the present invention
Attached drawing in embodiment, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that disclosed below
Embodiment be only a part of the embodiment of the present invention, and not all embodiment.Based on the embodiments of the present invention, this field
Those of ordinary skill's all other embodiment obtained without making creative work, belongs to protection of the present invention
Range.
Referring to Fig. 1, a kind of one embodiment of the work correction method based on image recognition can in the embodiment of the present invention
To include:
Step S101, the work correction request that receiving terminal apparatus is sent.
In the present embodiment, a large amount of mathematical problem is stored in the server in advance, to constitute one for teacher
The exam pool used when setting a question can distribute unique topic mark for the ease of distinguishing for each topic in exam pool.Teacher
When setting a question, it is all from from extracting topic in the exam pool and carrying out freely forming all topics in operation namely operation
It is as shown in the table in the exam pool:
Topic serial number in operation | Topic source (the topic mark in exam pool) |
Topic 1 | Topic mark: 5628 |
Topic 2 | Topic mark: 13 |
Topic 3 | Topic mark: 236 |
Topic 4 | Topic mark: 3 |
…… | …… |
It should be noted that one the above is only the corresponding relationship between topic in operation topic and exam pool is specifically shown
Example, during actually setting a question, can be arranged different corresponding relationships according to the actual situation.This correspondence is set in teacher to close
After system, server can save the corresponding relationship, use when correcting so as to subsequent progress paper.Optionally, Jiao Shike
To generate multiple operations according to the actual situation, for the ease of distinguishing, unique job identification can be distributed for each operation.
Teacher can print the operation of generation so that student fills in, and student can pass through after fulfiling assignment
Its terminal device sends work correction request to server, include in work correction request the pending image corrected students' papers and
The pending job identification corrected students' papers, wherein the pending image corrected students' papers can pass through its terminal device by student
Camera shoots to obtain.
Step S102, the topic of each topic is determined in the pending image corrected students' papers according to preset sign flag
It answers region.
As shown in Fig. 2, step S102 can specifically include following steps:
Step S1021, the first label and the second label are identified in the pending image corrected students' papers.
It can include a variety of special in finally formed operation as shown in figure 3, being corrected for the ease of carrying out automated job
Sign flag.Described first is labeled as the sign flag of topic starting position, in Fig. 3 ◆, described second is labeled as topic
The sign flag of end position, such as the ◢ in Fig. 3.
Server can carry out image preprocessing to the pending image corrected students' papers first, for example, carrying out first to image
Binaryzation gets rid of text interference by opening operation, then extracts the profile of residual image content, such as according to qualifications
Area, length-width ratio etc. filter out satisfactory image outline, and each sign flag is finally identified from these image outlines.
Below by way of use machine learning model to indicate topic starting position sign flag ◆ identification process for
It is described in detail:
Firstly, carry out the building of sample set, the positive sample and the non-sign flag of a large amount of sign flag are chosen
Negative sample.In order to guarantee that the machine learning model for training has better adaptability, various feelings can be collected as far as possible
Positive negative sample under condition, for example, the positive negative sample printed when printer ink is sufficient, insufficient, in various quality
The positive negative sample etc. that comes out of print on paper.These samples pass through artificial screenshot by scanner typing electronics shelves picture
And mark, to construct the sample set for carrying out machine learning model training.
Then, sample characteristics extraction is carried out.Histograms of oriented gradients (Histogram of can be passed through in this programme
Oriented Gradient, HOG), local binary patterns (Local Binary Patterns, LBP) and other existing skills
Common feature extraction algorithm extracts the characteristics of image of sample in art.
Then, preset machine learning model is trained using the characteristics of image of sample each in sample set, is obtained
Meet the machine learning model of preset error requirements.In the present solution, specifically used machine learning training method include but
It is not limited to: k nearest neighbor, naive Bayesian, logistic regression, support vector machines (Support Vector Machine, SVM), random
The common methods such as forest.
Finally, trained machine learning model can be used to identify candidate image outline, it is screened out from it
The sign flag.
The identification process of other sign flags is similar therewith, specifically can refer to above-mentioned detailed description, details are not described herein again.
Step S1022, it is determined in the pending image corrected students' papers according to first label and second label
Region where each topic.
As shown in figure 4, passing through the first label ◆ can determine topic starting position, can be determined by the first label ◢
Topic end position out, in this way, the region at place of each topic in operation can be determined, as box encloses in Fig. 4
Fixed region is the region where the topic.
Step S1023, third label and the 4th label are identified in the region where each topic.
The third is answered the sign flag of region starting position labeled as topic, as in Fig. 3 [, the 4th label
It answers the sign flag of region end position for topic, in Fig. 3].The identification of the third label and the 4th label
Process is similar with the identification process in step S1021, specifically can refer to above-mentioned detailed description, details are not described herein again.
Step S1024, it is determined in the region where each topic according to third label and the 4th label each
The topic of a topic is answered region.
As shown in figure 4, can by third label [can determine that topic is answered region starting position, pass through the 4th label]
To determine that topic is answered region end position, in this way, the topic of each topic is answered, region can be determined, in Fig. 4
The region that oval frame is drawn a circle to approve is that the topic of the topic is answered region.
It should be noted that actually setting a question the above is only a specific example of sign flag used in operation
Cheng Zhong, can also be arranged according to the actual situation other sign flags as it is described first label, second label, third label and
4th label, these sign flags include but is not limited to@, #, $, &, % etc..
Step S103, using preset deep learning network respectively to the topic of each topic answer region image carry out
Image recognition obtains the characteristics of image of each topic.
It is with the image recognition processes of q-th of topic (1≤q≤Q, Q are the pending topic sum corrected students' papers) below
Example is illustrated:
The answer image in region of the topic of q-th of topic is input in the deep learning network and is handled, is obtained
The characteristics of image of q-th of topic, the deep learning network are made of B L layers of sub-network cascade, it may be assumed that
SbNwOutb=SubNetwk (SbNwOutb-1)
Wherein, b is the serial number of sub-network, and 1≤b≤B, B are positive integer, SbNwOut0Topic for q-th of topic is answered
The image in region, SbNwOutbFor the output of b-th of sub-network, SubNetwk is the Processing Algorithm of each sub-network, it may be assumed that
Wherein, l is the number of plies serial number of sub-network, and 1≤l≤L, L are positive integer, LyOut0For the input of sub-network, LyOutl
For l layers of output of sub-network, [] indicates the merging (concatenation) in feature channel, for example, [LyOut0,
LyOut1,...,LyOutl-1] indicate to the feature channel of the output of l-1 layers before sub-network (from the 1st layer to l-1 layers) into
Row merges, and LayerFunc indicates that the input to current layer carries out feature extraction calculating, for example, can be according to the following formula using default
Feature extraction matrix feature extraction is carried out to the input of the current layer:
Wherein, EtrMtx is the feature extraction matrix, can be configured according to the actual situation, for example, can be setOrEtc., C1 is the line number of the feature extraction matrix, and C2 is the spy
Sign extracts matrix column number, and LyInMtx is the matrix form of the input of the current layer, and FtMtx is characterized the figure for extracting and obtaining
As eigenmatrix, Fts,tFor the element of described image eigenmatrix s row t column, 1≤s≤S, 1≤t≤T, S are described image
The line number of eigenmatrix, T is the columns of described image eigenmatrix, and S=M-C1+1, T=N-C2+1, M are the row of LyInMtx
Number, N be LyInMtx columns, LyInMtx (s, s+C1, t, t+C2) be LyInMtx s row to s+C1 row, t arrange to
The submatrix of t+C2 column region, Hadamard are the Hadamard product for seeking two matrixes.
It preferably, can also be according to the following formula to the feature after the input to the current layer carries out feature extraction
Matrix is extracted to carry out simplifying processing:
Wherein, SpMtx is simplified feature extraction matrix, Spu,vFor SpMtx u row v column element, 1≤u≤U,
1≤v≤V,D1 and D2 is preset simplified coefficient, can be carried out according to the actual situation to its value
Setting, for example, D1=2 can be set, D2=3 obtains D1=D2=4 etc., FtMtx ((u-1) × D1+1, u × D1, (v-1)
× D2+1, v × D2) be FtMtx (u-1) × D1+1 row to u × D1 row, (v-1) × D2+1 is arranged to v × D2 column area
The submatrix in domain, MtxMax are the maximum value for seeking all elements in matrix.
Due to mathematic(al) representation various structures, there is upper and lower, left and right, there are also the situations such as upper subscript, and hand-written character is big
Small disunity, and the case where there are font adhesions, it is not high using simple convolutional network recognition accuracy.Institute in the present embodiment
Each layer of sub-network of output for stating deep learning network is to be comprehensively considered to the output of preceding layers, can
The feature (such as: the network that can enter deeper when identification decimal point extracts feature) of different resolutions is extracted in the picture, thus
More accurately identify different characters.
Step S104, the characteristics of image of each topic is handled respectively using preset serializing model, is obtained each
The content of answering of a topic.
The serializing model is cascaded by more than two gating cycle units (Gated Recurrent Unit, GRU)
Composition.GRU is a kind of good variant of effect of LSTM, it is simpler compared with the structure of LSTM network, and effect is also fine,
It therefore is also a kind of network of current very manifold.Three gate functions are introduced in LSTM: input gate forgets door and out gate
To control input value, memory value and output valve.And only there are two doors in GRU model: being to update door and resetting door respectively.It updates
Status information of the door for controlling previous moment is brought into the degree in current state, update door the bigger explanation of value it is previous when
The status information at quarter bring into it is more, resetting door control how many information of previous state be written on current Candidate Set, reset
Door is smaller, and the information of previous state is written into fewer.A time series, such as text sequence are given, as composed by GRU
The serializing model can capture dependence of two moment between biggish text element (word, word, symbol etc.) and close
System generates text information, to obtain the content of answering of each topic.
Step S105, the model answer of each topic is determined according to the pending job identification corrected students' papers, and according to each
The content of answering of each topic is compared in the model answer of a topic, obtains work correction result.
After identifying the content of answering of each topic, it can be determined and be made according to the pending job identification corrected students' papers
In topic serial number (for example, [30] in Fig. 3 represent the 30th topic in operation) in industry and exam pool between topic mark
Corresponding relationship, and the model answer of each topic is inquired in the exam pool, finally according to the model answer of each topic
The content of answering of each topic is compared, work correction result is obtained.If the two compares unanimously, illustrate to answer correct,
If the two comparison is inconsistent, illustrate mistake of answering, to complete to correct process to the automation of operation.
In conclusion without special answering card in the embodiment of the present invention, and need to only be printed in operation default
Sign flag, receive terminal device transmission work correction request after, can be existed according to these sign flags
Determine that the topic of each topic is answered region, and the cost of work correction is greatly reduced in the pending image corrected students' papers.And
And in embodiments of the present invention using preset deep learning network respectively to the topic of each topic answer region image into
Row image recognition obtains the characteristics of image of each topic, reuses preset serializing model respectively to the image of each topic
Feature is handled, and the dependence between word in topic, word, symbol etc. text element can be effectively captured, and is obtained each
The content of answering of a topic, applicable for various scenes such as multiple-choice question, gap-filling questions, simple answers, finally only needing will be preset
Content of answering is compared in model answer, and final work correction result can be obtained.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
The present invention is shown corresponding to a kind of work correction method based on image recognition, Fig. 5 described in foregoing embodiments
A kind of one embodiment structure chart for work correction device that embodiment provides.
In the present embodiment, a kind of work correction device may include:
Work correction request receiving module 501, for the work correction request that receiving terminal apparatus is sent, the job batch
Change includes the pending image corrected students' papers and the pending job identification corrected students' papers in request;
Topic is answered area determination module 502, for according to preset sign flag in the pending image corrected students' papers
The topic of the middle each topic of determination is answered region;
Image characteristics extraction module 503, for using preset deep learning network to make respectively to the topic of each topic
The image for answering region carries out image recognition, obtains the characteristics of image of each topic;
Serialize processing module 504, for use preset serializing model respectively to the characteristics of image of each topic into
Row processing, obtains the content of answering of each topic, the serializing model is formed by more than two gating cycles are unit cascaded;
Content of answering comparison module 505, for determining the mark of each topic according to the pending job identification corrected students' papers
Quasi- answer, and be compared according to answer content of the model answer of each topic to each topic, obtain work correction result.
Further, described image characteristic extracting module may include:
Image characteristics extraction unit, for the answer image in region of the topic of q-th of topic to be input to the depth
It practising and being handled in network, obtain the characteristics of image of q-th of topic, 1≤q≤Q, Q are the pending topic sum corrected students' papers,
The deep learning network is made of B L layers of sub-network cascade, it may be assumed that
SbNwOutb=SubNetwk (SbNwOutb-1)
Wherein, b is the serial number of sub-network, and 1≤b≤B, B are positive integer, SbNwOut0Topic for q-th of topic is answered
The image in region, SbNwOutbFor the output of b-th of sub-network, SubNetwk is the Processing Algorithm of each sub-network, it may be assumed that
Wherein, l is the number of plies serial number of sub-network, and 1≤l≤L, L are positive integer, LyOut0For the input of sub-network, LyOutl
For l layers of output of sub-network, LayerFunc indicates that the input to current layer carries out feature extraction calculating.
Further, described image feature extraction unit is specifically used for using preset feature extraction matrix pair according to the following formula
The input of the current layer carries out feature extraction:
Wherein, EtrMtx is the feature extraction matrix, and C1 is the line number of the feature extraction matrix, and C2 is the feature
Matrix column number is extracted, LyInMtx is the matrix form of the input of the current layer, and FtMtx is characterized the image for extracting and obtaining
Eigenmatrix, Fts,tFor the element of described image eigenmatrix s row t column, 1≤s≤S, 1≤t≤T, S are that described image is special
The line number of matrix is levied, T is the columns of described image eigenmatrix, and S=M-C1+1, T=N-C2+1, M are the row of LyInMtx
Number, N be LyInMtx columns, LyInMtx (s, s+C1, t, t+C2) be LyInMtx s row to s+C1 row, t arrange to
The submatrix of t+C2 column region, Hadamard are the Hadamard product for seeking two matrixes.
Further, described image characteristic extracting module can also include:
Simplify processing unit, simplify processing for carrying out according to the following formula to the feature extraction matrix:
Wherein, SpMtx is simplified feature extraction matrix, Spu,vFor SpMtx u row v column element, 1≤u≤U,
1≤v≤V,D1 and D2 is preset simplified coefficient, FtMtx ((u-1) × D1+1, u × D1, (v-
1) × D2+1, v × D2) be FtMtx (u-1) × D1+1 row to u × D1 row, (v-1) × D2+1 arrange to v × D2 arrange
The submatrix in region, MtxMax are the maximum value for seeking all elements in matrix.
Further, topic area determination module of answering may include:
First recognition unit, for identifying the first label and the second label, institute in the pending image corrected students' papers
The first sign flag for being labeled as topic starting position is stated, described second is labeled as the sign flag of topic end position;
Topic block determination unit, for pending being corrected students' papers according to first label and second label described
The region where each topic is determined in image;
Second recognition unit, it is described for identifying third label and the 4th label in the region where each topic
Third is answered the sign flag of region starting position labeled as topic, and the described 4th answers region end position labeled as topic
Sign flag;
Topic is answered area determination unit, for according to third label and the 4th label where each topic
Region in determine that the topic of each topic is answered region.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description,
The specific work process of module and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
The schematic block diagram that Fig. 6 shows a kind of server provided in an embodiment of the present invention illustrates only for ease of description
Part related to the embodiment of the present invention.
In the present embodiment, the server 6 may include: processor 60, memory 61 and be stored in the storage
In device 61 and the computer-readable instruction 62 that can be run on the processor 60, such as execute above-mentioned based on image recognition
The computer-readable instruction of work correction method.The processor 60 is realized above-mentioned each when executing the computer-readable instruction 62
Step in a work correction embodiment of the method based on image recognition, such as step S101 to S105 shown in FIG. 1.Alternatively,
The processor 60 realizes the function of each module/unit in above-mentioned each Installation practice when executing the computer-readable instruction 62
Can, such as the function of module 501 to 505 shown in Fig. 5.
Illustratively, the computer-readable instruction 62 can be divided into one or more module/units, one
Or multiple module/units are stored in the memory 61, and are executed by the processor 60, to complete the present invention.Institute
Stating one or more module/units can be the series of computation machine readable instruction section that can complete specific function, the instruction segment
For describing implementation procedure of the computer-readable instruction 62 in the server 6.
The processor 60 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), specific integrated circuit
(Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor is also possible to any conventional processor
Deng.
The memory 61 can be the internal storage unit of the server 6, such as the hard disk or memory of server 6.
The memory 61 is also possible to the External memory equipment of the server 6, such as the plug-in type being equipped on the server 6 is hard
Disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card
(Flash Card) etc..Further, the memory 61 can also both include the internal storage unit of the server 6 or wrap
Include External memory equipment.The memory 61 is for storing needed for the computer-readable instruction and the server 6 it
Its instruction and data.The memory 61 can be also used for temporarily storing the data that has exported or will export.
The functional units in various embodiments of the present invention may be integrated into one processing unit, is also possible to each
Unit physically exists alone, and can also be integrated in one unit with two or more units.Above-mentioned integrated unit both may be used
To use formal implementation of hardware, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, technical solution of the present invention substantially or
Person says that all or part of the part that contributes to existing technology or the technical solution can body in the form of software products
Reveal and, which is stored in a storage medium, including several computer-readable instructions are used so that one
Platform computer equipment (can be personal computer, server or the network equipment etc.) executes described in each embodiment of the present invention
The all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can be with
Store the medium of computer-readable instruction.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations;Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features;And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of work correction method based on image recognition characterized by comprising
The work correction request that receiving terminal apparatus is sent, include in work correction request the pending image corrected students' papers and
The pending job identification corrected students' papers;
Determine that the topic of each topic is answered region in the pending image corrected students' papers according to preset sign flag;
Using preset deep learning network respectively to the topic of each topic answer region image carry out image recognition, obtain
The characteristics of image of each topic;
The characteristics of image of each topic is handled respectively using preset serializing model, it is interior to obtain answering for each topic
Hold, the serializing model is formed by more than two gating cycles are unit cascaded;
The model answer of each topic is determined according to the pending job identification corrected students' papers, and is answered according to the standard of each topic
The content of answering of each topic is compared in case, obtains work correction result.
2. work correction method according to claim 1, which is characterized in that described to use preset deep learning network point
The answer image in region of the other topic to each topic carries out image recognition, and the characteristics of image for obtaining each topic includes:
The answer image in region of the topic of q-th of topic is input in the deep learning network and is handled, is obtained q-th
The characteristics of image of topic, 1≤q≤Q, Q are the pending topic sum corrected students' papers, and the deep learning network is by B L layers
Sub-network cascade composition, it may be assumed that
SbNwOutb=SubNetwk (SbNwOutb-1)
Wherein, b is the serial number of sub-network, and 1≤b≤B, B are positive integer, SbNwOut0It answers region for the topic of q-th of topic
Image, SbNwOutbFor the output of b-th of sub-network, SubNetwk is the Processing Algorithm of each sub-network, it may be assumed that
Wherein, l is the number of plies serial number of sub-network, and 1≤l≤L, L are positive integer, LyOut0For the input of sub-network, LyOutlFor son
The output that l layers of network, LayerFunc indicate that the input to current layer carries out feature extraction calculating.
3. work correction method according to claim 2, which is characterized in that the input to current layer carries out feature pumping
The calculating is taken to include:
Feature extraction is carried out to the input of the current layer using preset feature extraction matrix according to the following formula:
Wherein, EtrMtx is the feature extraction matrix, and C1 is the line number of the feature extraction matrix, and C2 is the feature extraction
Matrix column number, LyInMtx are the matrix form of the input of the current layer, and FtMtx is characterized the characteristics of image for extracting and obtaining
Matrix, Fts,tFor the element of described image eigenmatrix s row t column, 1≤s≤S, 1≤t≤T, S are described image feature square
The line number of battle array, T is the columns of described image eigenmatrix, and S=M-C1+1, T=N-C2+1, M are the line number of LyInMtx, and N is
The columns of LyInMtx, LyInMtx (s, s+C1, t, t+C2) are the s row of LyInMtx to s+C1 row, and t is arranged to t+C2
The submatrix of column region, Hadamard are the Hadamard product for seeking two matrixes.
4. work correction method according to claim 3, which is characterized in that carry out feature in the input to the current layer
After extraction, further includes:
The feature extraction matrix is carried out according to the following formula to simplify processing:
Wherein, SpMtx is simplified feature extraction matrix, Spu,vFor the element of SpMtx u row v column, 1≤u≤U, 1≤v
≤ V,D1 and D2 is preset simplified coefficient, FtMtx ((u-1) × D1+1, u × D1, (v-1)
× D2+1, v × D2) be FtMtx (u-1) × D1+1 row to u × D1 row, (v-1) × D2+1 is arranged to v × D2 column area
The submatrix in domain, MtxMax are the maximum value for seeking all elements in matrix.
5. work correction method according to any one of claim 1 to 4, which is characterized in that according to preset symbol mark
Remember and determines that the topic of each topic region of answering includes: in the pending image corrected students' papers
The first label and the second label are identified in the pending image corrected students' papers, described first is labeled as topic start bit
The sign flag set, described second is labeled as the sign flag of topic end position;
It is marked according to first label and described second where determining each topic in the pending image corrected students' papers
Region;
Third label and the 4th label are identified in the region where each topic, the third is answered region labeled as topic
The sign flag of starting position, the described 4th answers the sign flag of region end position labeled as topic;
Determine that the topic of each topic is made in the region where each topic according to third label and the 4th label
Answer region.
6. a kind of work correction device characterized by comprising
Work correction request receiving module, for the work correction request that receiving terminal apparatus is sent, the work correction request
In include the pending image corrected students' papers and the pending job identification corrected students' papers;
Topic is answered area determination module, each for being determined in the pending image corrected students' papers according to preset sign flag
The topic of a topic is answered region;
Image characteristics extraction module, for being answered region to the topic of each topic respectively using preset deep learning network
Image carries out image recognition, obtains the characteristics of image of each topic;
Processing module is serialized, for being handled respectively the characteristics of image of each topic using preset serializing model,
The content of answering of each topic is obtained, the serializing model is formed by more than two gating cycles are unit cascaded;
Content of answering comparison module, for determining the model answer of each topic according to the pending job identification corrected students' papers,
And be compared according to answer content of the model answer of each topic to each topic, obtain work correction result.
7. work correction device according to claim 6, which is characterized in that described image characteristic extracting module includes:
Image characteristics extraction unit, for the answer image in region of the topic of q-th of topic to be input to the deep learning net
It is handled in network, obtains the characteristics of image of q-th of topic, 1≤q≤Q, Q are the pending topic sum corrected students' papers, described
Deep learning network is made of B L layers of sub-network cascade, it may be assumed that
SbNwOutb=SubNetwk (SbNwOutb-1)
Wherein, b is the serial number of sub-network, and 1≤b≤B, B are positive integer, SbNwOut0It answers region for the topic of q-th of topic
Image, SbNwOutbFor the output of b-th of sub-network, SubNetwk is the Processing Algorithm of each sub-network, it may be assumed that
Wherein, l is the number of plies serial number of sub-network, and 1≤l≤L, L are positive integer, LyOut0For the input of sub-network, LyOutlFor son
The output that l layers of network, LayerFunc indicate that the input to current layer carries out feature extraction calculating.
8. work correction device according to claim 6 or 7, which is characterized in that the topic is answered area determination module
Include:
First recognition unit, for identifying the first label and the second label in the pending image corrected students' papers, described the
One is labeled as the sign flag of topic starting position, and described second is labeled as the sign flag of topic end position;
Topic block determination unit, for being marked according to first label and described second in the pending image corrected students' papers
Region where the middle each topic of determination;
Second recognition unit, for identifying third label and the 4th label, the third in the region where each topic
It answers the sign flag of region starting position labeled as topic, the described 4th answers the symbol of region end position labeled as topic
Label;
Topic is answered area determination unit, for according to third label and the 4th label in the area where each topic
Determine that the topic of each topic is answered region in domain.
9. a kind of computer readable storage medium, the computer-readable recording medium storage has computer-readable instruction, special
Sign is, the job batch as described in any one of claims 1 to 5 is realized when the computer-readable instruction is executed by processor
The step of changing method.
10. a kind of server, including memory, processor and storage can transport in the memory and on the processor
Capable computer-readable instruction, which is characterized in that realized when the processor executes the computer-readable instruction as right is wanted
Described in asking any one of 1 to 5 the step of work correction method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910603800.5A CN110490180B (en) | 2019-07-05 | 2019-07-05 | Image recognition-based job correction method and device, storage medium and server |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910603800.5A CN110490180B (en) | 2019-07-05 | 2019-07-05 | Image recognition-based job correction method and device, storage medium and server |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110490180A true CN110490180A (en) | 2019-11-22 |
CN110490180B CN110490180B (en) | 2022-06-17 |
Family
ID=68546798
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910603800.5A Active CN110490180B (en) | 2019-07-05 | 2019-07-05 | Image recognition-based job correction method and device, storage medium and server |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110490180B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111079641A (en) * | 2019-12-13 | 2020-04-28 | 科大讯飞股份有限公司 | Answering content identification method, related device and readable storage medium |
CN111104883A (en) * | 2019-12-09 | 2020-05-05 | 平安国际智慧城市科技股份有限公司 | Job answer extraction method, device, equipment and computer readable storage medium |
CN111192171A (en) * | 2019-12-27 | 2020-05-22 | 创而新(北京)教育科技有限公司 | Teaching assistance method, teaching assistance device, teaching assistance equipment and storage medium |
CN111507292A (en) * | 2020-04-22 | 2020-08-07 | 广州光大教育软件科技股份有限公司 | Handwriting board correction method and device, computer equipment and storage medium |
CN111539369A (en) * | 2020-04-30 | 2020-08-14 | 联想万像(深圳)科技有限公司 | Job processing method, device and equipment and readable storage medium |
CN111768170A (en) * | 2020-06-23 | 2020-10-13 | 北京字节跳动网络技术有限公司 | Method and device for displaying operation correction result |
CN111767883A (en) * | 2020-07-07 | 2020-10-13 | 北京猿力未来科技有限公司 | Title correction method and device |
CN111967479A (en) * | 2020-07-27 | 2020-11-20 | 广东工业大学 | Image target identification method based on convolutional neural network idea |
CN112183253A (en) * | 2020-09-15 | 2021-01-05 | 北京大米科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN113065988A (en) * | 2021-03-31 | 2021-07-02 | 北京有竹居网络技术有限公司 | Operation correction method, system and device |
CN113140210A (en) * | 2021-04-30 | 2021-07-20 | 武汉悦学帮网络技术有限公司 | Audio correction method and device, electronic equipment and storage medium |
CN113221877A (en) * | 2020-02-06 | 2021-08-06 | 北京字节跳动网络技术有限公司 | Method and apparatus for processing image |
CN113469147A (en) * | 2021-09-02 | 2021-10-01 | 北京世纪好未来教育科技有限公司 | Answer sheet identification method and device, electronic equipment and storage medium |
CN113793112A (en) * | 2021-07-29 | 2021-12-14 | 深圳点猫科技有限公司 | Automatic correcting method, system and medium for programming operation |
CN114662621A (en) * | 2022-05-24 | 2022-06-24 | 灵枭科技(武汉)有限公司 | Agricultural machinery working area calculation method and system based on machine learning |
CN113140210B (en) * | 2021-04-30 | 2024-06-07 | 武汉悦学帮网络技术有限公司 | Audio correction method, device, electronic equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007233888A (en) * | 2006-03-03 | 2007-09-13 | Fuji Xerox Co Ltd | Image processor and image processing program |
CN106023698A (en) * | 2016-07-29 | 2016-10-12 | 李铧 | Automatic reading and amending method for homework and exercise books |
CN106504599A (en) * | 2016-11-29 | 2017-03-15 | 华中师范大学 | Method that online assignment is corrected, apparatus and system |
CN107798321A (en) * | 2017-12-04 | 2018-03-13 | 海南云江科技有限公司 | A kind of examination paper analysis method and computing device |
CN108764201A (en) * | 2018-06-06 | 2018-11-06 | 宁波宁大教育设备有限公司 | A kind of intelligence correction method |
CN109712043A (en) * | 2018-12-28 | 2019-05-03 | 杭州大拿科技股份有限公司 | Method and device is corrected in a kind of answer |
-
2019
- 2019-07-05 CN CN201910603800.5A patent/CN110490180B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007233888A (en) * | 2006-03-03 | 2007-09-13 | Fuji Xerox Co Ltd | Image processor and image processing program |
CN106023698A (en) * | 2016-07-29 | 2016-10-12 | 李铧 | Automatic reading and amending method for homework and exercise books |
CN106504599A (en) * | 2016-11-29 | 2017-03-15 | 华中师范大学 | Method that online assignment is corrected, apparatus and system |
CN107798321A (en) * | 2017-12-04 | 2018-03-13 | 海南云江科技有限公司 | A kind of examination paper analysis method and computing device |
CN108764201A (en) * | 2018-06-06 | 2018-11-06 | 宁波宁大教育设备有限公司 | A kind of intelligence correction method |
CN109712043A (en) * | 2018-12-28 | 2019-05-03 | 杭州大拿科技股份有限公司 | Method and device is corrected in a kind of answer |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111104883A (en) * | 2019-12-09 | 2020-05-05 | 平安国际智慧城市科技股份有限公司 | Job answer extraction method, device, equipment and computer readable storage medium |
CN111104883B (en) * | 2019-12-09 | 2023-06-06 | 平安国际智慧城市科技股份有限公司 | Job answer extraction method, apparatus, device and computer readable storage medium |
CN111079641A (en) * | 2019-12-13 | 2020-04-28 | 科大讯飞股份有限公司 | Answering content identification method, related device and readable storage medium |
CN111079641B (en) * | 2019-12-13 | 2024-04-16 | 科大讯飞股份有限公司 | Answer content identification method, related device and readable storage medium |
CN111192171A (en) * | 2019-12-27 | 2020-05-22 | 创而新(北京)教育科技有限公司 | Teaching assistance method, teaching assistance device, teaching assistance equipment and storage medium |
CN113221877A (en) * | 2020-02-06 | 2021-08-06 | 北京字节跳动网络技术有限公司 | Method and apparatus for processing image |
CN111507292A (en) * | 2020-04-22 | 2020-08-07 | 广州光大教育软件科技股份有限公司 | Handwriting board correction method and device, computer equipment and storage medium |
CN111507292B (en) * | 2020-04-22 | 2023-05-12 | 广东光大信息科技股份有限公司 | Handwriting board correction method, handwriting board correction device, computer equipment and storage medium |
CN111539369A (en) * | 2020-04-30 | 2020-08-14 | 联想万像(深圳)科技有限公司 | Job processing method, device and equipment and readable storage medium |
CN111768170A (en) * | 2020-06-23 | 2020-10-13 | 北京字节跳动网络技术有限公司 | Method and device for displaying operation correction result |
CN111767883A (en) * | 2020-07-07 | 2020-10-13 | 北京猿力未来科技有限公司 | Title correction method and device |
CN111767883B (en) * | 2020-07-07 | 2024-04-12 | 北京猿力未来科技有限公司 | Question correction method and device |
CN111967479A (en) * | 2020-07-27 | 2020-11-20 | 广东工业大学 | Image target identification method based on convolutional neural network idea |
CN112183253A (en) * | 2020-09-15 | 2021-01-05 | 北京大米科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN113065988A (en) * | 2021-03-31 | 2021-07-02 | 北京有竹居网络技术有限公司 | Operation correction method, system and device |
CN113140210A (en) * | 2021-04-30 | 2021-07-20 | 武汉悦学帮网络技术有限公司 | Audio correction method and device, electronic equipment and storage medium |
CN113140210B (en) * | 2021-04-30 | 2024-06-07 | 武汉悦学帮网络技术有限公司 | Audio correction method, device, electronic equipment and storage medium |
CN113793112A (en) * | 2021-07-29 | 2021-12-14 | 深圳点猫科技有限公司 | Automatic correcting method, system and medium for programming operation |
CN113469147B (en) * | 2021-09-02 | 2021-12-17 | 北京世纪好未来教育科技有限公司 | Answer sheet identification method and device, electronic equipment and storage medium |
CN113469147A (en) * | 2021-09-02 | 2021-10-01 | 北京世纪好未来教育科技有限公司 | Answer sheet identification method and device, electronic equipment and storage medium |
CN114662621A (en) * | 2022-05-24 | 2022-06-24 | 灵枭科技(武汉)有限公司 | Agricultural machinery working area calculation method and system based on machine learning |
Also Published As
Publication number | Publication date |
---|---|
CN110490180B (en) | 2022-06-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110490180A (en) | Work correction method, apparatus, storage medium and server based on image recognition | |
US11410407B2 (en) | Method and device for generating collection of incorrectly-answered questions | |
CN108171297B (en) | Answer sheet identification method | |
CN109815932B (en) | Test paper correcting method and device, electronic equipment and storage medium | |
CN107798321B (en) | Test paper analysis method and computing device | |
US8494273B2 (en) | Adaptive optical character recognition on a document with distorted characters | |
CN109634961B (en) | Test paper sample generation method and device, electronic equipment and storage medium | |
CN109242856A (en) | A kind of papery operation electronization reads and makes comments method and device | |
CN110705233B (en) | Note generation method and device based on character recognition technology and computer equipment | |
CN107516096A (en) | A kind of character identifying method and device | |
CN110046622B (en) | Targeted attack sample generation method, device, equipment and storage medium | |
CN109784342A (en) | A kind of OCR recognition methods and terminal based on deep learning model | |
CN113011144A (en) | Form information acquisition method and device and server | |
WO2021232670A1 (en) | Pcb component identification method and device | |
CN110210309A (en) | The recognition methods of mistake topic, system, readable storage medium storing program for executing and equipment | |
CN113762274B (en) | Answer sheet target area detection method, system, storage medium and equipment | |
CN107220610A (en) | A kind of subjective item fraction recognition methods applied to marking system | |
JPH06282686A (en) | Method for character recognition | |
CN112686263A (en) | Character recognition method and device, electronic equipment and storage medium | |
CN114648771A (en) | Character recognition method, electronic device and computer readable storage medium | |
CN116030469A (en) | Processing method, processing device, processing equipment and computer readable storage medium | |
CN114332898A (en) | Automatic correcting method and device for connection test questions and storage medium | |
CN107045635A (en) | A kind of paper image paging sub title processing method of online paper-marking system | |
CN110751140A (en) | Character batch recognition method and device and computer equipment | |
CN113223025B (en) | Image processing method and device, and neural network training method and device |
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 |