CN107729936A - One kind corrects mistakes to inscribe reads and appraises method and system automatically - Google Patents
One kind corrects mistakes to inscribe reads and appraises method and system automatically Download PDFInfo
- Publication number
- CN107729936A CN107729936A CN201710947823.9A CN201710947823A CN107729936A CN 107729936 A CN107729936 A CN 107729936A CN 201710947823 A CN201710947823 A CN 201710947823A CN 107729936 A CN107729936 A CN 107729936A
- Authority
- CN
- China
- Prior art keywords
- answer
- image
- read
- point
- answer image
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The embodiment of the present invention provide one kind correct mistakes topic read and appraise method and device automatically, belong to image processing field.This method includes：Obtain and single-point answer image corresponding to answer is respectively write on answer medium to be read；Image recognition is carried out to each single-point answer image got using pre-set image detection method and/or based on default identification model, obtains recognition result；Obtained to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, each writing answer is read and appraised.This specific embodiment method can overcome the technical problem that the complexity due to topic answer pattern of correcting mistakes causes answer image recognition to be forbidden, so that with quick, the accurate beneficial effect for correct mistakes and inscribing and reading and appraising.
Description
Technical field
The present invention relates to technical field of image information processing, and more particularly, to correcting mistakes, topic reads and appraises method and system automatically.
Background technology
In the class of languages examination examination in the present age, (short essay) topic of correcting mistakes belongs to the objective investigation of subjective paper.Both investigated and answered
The rudimentary knowledge of topic person's each side, the ability of answer person's integrated use language is investigated again.In terms of topic is set, topic is from morphology, sentence
The ability of the corresponding languages language of method and style of writing logic three angles examination candidate's integrated uses in a language piece；During details on pair
State, voice, word predicate verb, noun, article, adjective, conjunction, pronoun and sentence undertaking in terms of examined or check.Answer
The mode that person increased word in examination paper by hand-written mode, is deleted or modified carries out answer.
In recent years, the quick of developing rapidly with computer technology and information technology, especially artificial intelligence technology pushes away
Enter, replace Traditional Man to have become the focus direction of all trades and professions using artificial intelligence.The mould of going over examination papers in exam-oriented education field
Formula also by it is traditional it is pure manually read and appraise, progressively develop into and replace artificial realize in part to read and appraise automatically by artificial intelligence.Existing main flow
Paper read and appraise automatically, except realizing practical objective item earliest, also portion of techniques development station industry forefront product,
Realize that composition, the automatic of gap-filling questions are read and appraised.
However, for correct mistakes topic, be limited to answer person answer middle written handwriting complicated variety (as " increase "
Written mark " ∧ " various handwritten patterns can occur because answer people is different), so topic of correcting mistakes stills need exam paper assessment person and manually commented so far
Read.Particularly in large-scale examination, when such as middle college entrance examination, language grade examination, the mode manually read and appraised takes time and effort, to commenting
The personnel of readding bring very big operating pressure；In addition, manually read and appraise also Chang Rongyi is influenceed by the personnel subjectivity of reading and appraising so that reads and appraises knot
Fruit loses fairness.
In summary, a kind of the automatic of topic of correcting mistakes urgently is provided in the prior art and reads and appraises scheme.
The content of the invention
In order to solve the above problems, the embodiment of the present invention provides one kind and overcomes above mentioned problem or solve at least in part
The topic of correcting mistakes for stating problem reads and appraises method and device automatically.
First aspect according to embodiments of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, and this method includes：
Obtain and single-point answer image corresponding to answer is respectively write on answer medium to be read；
Each single-point answer image got is carried out using pre-set image detection method and/or based on default identification model
Image recognition, obtain recognition result；
Obtained to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, to each book
Answer is write to be read and appraised.
Another aspect according to embodiments of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device automatically, and the device includes：
Acquisition module, single-point answer image corresponding to answer is respectively write on answer medium to be read for obtaining；
Identification module, for using pre-set image detection method and/or based on default identification model to each list for getting
Point answer image carries out image recognition, obtains recognition result；
Module is read and appraised, feature is read and appraised for obtaining to correspond to according to identification information, default comment is based on according to the feature of reading and appraising
Model is read, each writing answer is read and appraised.
The third aspect according to embodiments of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises electronic equipment automatically, including：
At least one processor；And
At least one memory being connected with processor communication, wherein：
Memory storage has the programmed instruction that can be executed by processor, and the instruction of processor caller is able to carry out first party
What any possible implementation was provided in the various possible implementations in face correct mistakes, and topic reads and appraises method automatically.
According to the fourth aspect of the invention, there is provided a kind of non-transient computer readable storage medium storing program for executing, non-transient computer
Readable storage medium storing program for executing stores computer instruction, and computer instruction makes the various possible implementations of computer execution first aspect
In any possible implementation provided correct mistakes topic read and appraise method automatically.
The above embodiment of the present invention provides one kind and corrects mistakes that topic is automatic to read and appraise method and device, methods described can overcome due to
The complexity of topic of correcting mistakes answer pattern causes the inaccurate technical problem of answer image information identification, so as to automatic, quick and accurate
Really carry out the beneficial effect that topic is read and appraised of correcting mistakes.
Brief description of the drawings
Fig. 1 is a kind of automatic schematic flow sheet for reading and appraising method of topic of correcting mistakes of the embodiment of the present invention；
Fig. 2 reads and appraises to treat in method automatically for a kind of topic of correcting mistakes of the embodiment of the present invention reads and appraises what wrong topic image was handled
Schematic diagram；
Fig. 3 is a kind of automatic knot for reading and appraising the specific limited network built in advance in method of topic of correcting mistakes of the embodiment of the present invention
Structure schematic diagram；
Fig. 4 is a kind of automatic block diagram for reading and appraising device of topic of correcting mistakes of the embodiment of the present invention；
Fig. 5 is a kind of automatic block diagram for reading and appraising electronic equipment of topic of correcting mistakes of the embodiment of the present invention.
Embodiment
With reference to the accompanying drawings and examples, the embodiment of the present invention is described in further detail.Implement below
Example is used to illustrate the present invention, but is not limited to the scope of the present invention.
With the development of image recognition technology, the application field of image recognition technology is more and more wider.At present, image recognition skill
Art is applied to education sector.In the prior art, existed and read and appraised using image recognition technology progress part examination question fraction
Method.However, in class of languages examination examination, (short essay) corrects mistakes topic, due to the regular complexity of its answer (including：Add
Add, delete, change), and diversity and the personalization of the various handwritings of answer people are limited to, in the prior art, not
Have and sentence volume technical scheme in the automation for this kind of topic type.
For said circumstances, correct mistakes to inscribe the embodiments of the invention provide one kind and read and appraise method automatically.This method be applied to pair
Read and appraised in class of languages class (Error Correction class) topic type of correcting mistakes；This method can be applied to the intelligence with IMAQ identification function
Equipment or system, the embodiment of the present invention are not especially limited to this.For the ease of description, the embodiment of the present invention using executive agent as
Exemplified by smart machine.As shown in figure 1, methods described includes：S1, obtain and single-point corresponding to answer is respectively write on answer medium to be read
Answer image；S2, using pre-set image detection method and/or based on default identification model to each single-point answer image for getting
Image recognition is carried out, obtains recognition result；S3, obtained to correspond to according to identification information and read and appraise feature, feature base is read and appraised according to described
Model is read and appraised in default, each writing answer is read and appraised；The identification information includes mistake among the recognition result and identification
Journey information.
In above-mentioned specific embodiment, respectively single-point answer corresponding to writing answer on answer medium to be read described in the S1
Image is to include the image of the hand-written answer of each answer person, and the acquisition of the single-point answer image can be by taking pictures, imaging, sweeping
The mode for retouching or being introduced directly into the image file shot in advance is carried out, and the embodiment of the present invention is not especially limited to this.
Further, in the S2 for correct option for delete when, can be by carrying out figure to the single-point answer image
As detection, and then judge that the single-point answer image is answered with the presence or absence of corresponding deletion；For modification or add for correct option
Added-time, each whole character string of single-point answer image can either be known based on the neural network recognization model previously generated
Not, recognition result is obtained；The cutting of single hand-written character can also be first carried out to each single-point answer image, it is each to being included after cutting
Each subgraph of single hand-written character is identified, and obtains recognition result.
Further, for the S3 according to the recognition result obtained in S2, the neutral net obtained based on training in advance reads and appraises mould
Result is read and appraised in each writing answer of type acquisition.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, described to obtain
Take on answer medium to be read respectively corresponding to writing answer the step of single-point answer image, including：According to answering for answer medium to be read
Image and answer medium template corresponding with the answer medium to be read are inscribed, each writing is answered on answer medium to be read described in acquisition
Single-point answer image corresponding to case；Wherein, the answer medium template includes the positional information of each character in examination question to be answered, and
According to correct option at least one single-point answer region set in advance, the single-point answer image is that the single-point corresponding to is answered
Inscribe region and obtain.
In above-mentioned specific embodiment, the acquisition of the answer image of the answer medium to be read can by taking pictures, imaging,
The mode for scanning or being introduced directly into the image file shot in advance is carried out, and the embodiment of the present invention is not especially limited to this.
Further, by that (can be the answering card of answer medium to be read, also may be used according to the answering pattern plate to the answer medium to be read
To be other answer media for including topic word such as answer paper, electric answer text of the answer medium to be read)
Layout carry out cutting, the answer image of answer medium to be read described in acquisition.It can both be included in the topic image of correcting mistakes whole
The answer image of the answer medium to be read, can also include the answer image of answer medium to be read described in part, and the present invention is real
Example is applied to be not especially limited herein.
Further, in this embodiment of the invention, described treat both had been deposited due to inevitable in the answer image of answer medium to be read
The topic print character image in topic of correcting mistakes is read and appraised, in the answering information image for answering hand-written there is also answer person.So this tool
Body embodiment previously generates the template of examination question to be read and appraised, due to when setting the template, for the difference side of answering for topic of correcting mistakes
Formula correspondingly selectes the corresponding band of position based on rule set in advance, so enabling to include in the template in topic respectively
The chosen position region of correct option (i.e. each single-point answer image).In the present embodiment, the selection in the template is recycled
The band of position, each single-point answer image is obtained from the answer image of the answer medium to be read.
Further, as shown in Fig. 2 in the form of the type of correcting mistakes for topic of correcting mistakes presets answer, such as with college entrance examination English
Exemplified by correcting mistakes, answer type of correcting mistakes includes modification, deletes and add.Wherein modification answers pattern to be marked below word to be modified
Note horizontal line and write out corresponding correct option (type fount parent, crowded in such as figure, on, looks, where,
Telling and a and its corresponding answer shown in image)；The answer pattern of deletion is to rule to delete to show on word to be deleted
(such as type fount very in figure and its corresponding answer shown in image)；The answer pattern of addition is in two character string to be added
Between below position written mark " ∧ " and write out addition answer (in such as figure between type fount promised and her " ∧ " and
Answered corresponding to it shown in image).
Next the examination question template generation process to be read and appraised previously generated in the present embodiment is further described first.
In Fig. 2, floor projection is carried out to the topic examination question image of correcting mistakes for not carrying out answer first, realizes the horizontal cutting to examination question image；
Then upright projection is carried out to every row examination question image so as to realizing monocase cutting in examination question, each character in the examination question that obtains correcting mistakes
The positional information of string.
Further, the single-point answer region is：When correct option is deletes, character string institute to be deleted occupied area in paper
Domain increases the width of preset ratio or preset length, and the height area defined of increase preset ratio or preset length.When
When correct option is changes, the predeterminable area in paper below print character string to be modified.When correct option is adds, paper
In among two print characters string to be added lower section predeterminable area.
Preferably, the chosen position region of each answer image is specially in the embodiment：Delete the answer of answer type
Image chosen position region, region and left and right extension 1/n width shared by word printing symbol string to be deleted in topic, 1/m is extended up and down
Height area defined, the n >=1, m >=1.Preferably, the answer image chosen position region for changing answer type is to inscribe
Printing symbol string lower section to be repaired of altering, width expand 1/2 width by word answer or so, are highly the area that line space surrounds in mesh
Domain.Preferably, the answer image chosen position region for adding answer type is, in topic under to be added two printing symbol strings
Side, width are the width of the most wide character obtained based on statistics, are highly line space area defined.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method, the profit automatically
Image recognition is carried out to each single-point answer image got with pre-set image detection method and/or based on default identification model,
Recognition result is obtained, including：
During for correct option to delete, marked by detecting in corresponding single-point answer image with the presence or absence of at least one delete
Note, obtain the recognition result answered with the presence or absence of deletion；
When correct option is changes or added, based on the limited network built in advance corresponding with correct option, to corresponding
Single-point answer image carries out single hand-written character cutting, obtains each subgraph of the single-point answer image；Based on advance structure
Topic identification model of correcting mistakes, each subgraph is identified, obtains the single hand-written character included in each subgraph
Recognition result.
, can be by detecting in the single-point answer image for the answer pattern of deletion in above-mentioned specific embodiment
Marked on print character with the presence or absence of at least one delete, to judge whether the answer point answers correctly.Wherein, specifically compared with
For such as detection of simple scheme previously generate wait read and appraise in examination question template in examination question target print character string with after examinee's answer
All pixels point registration in each single-point answer image, if misaligned pixel quantity or proportion are more than given threshold,
Then think that the answer point is answered in the presence of deletion.It is further preferred that the detection on deleting answer, can also be used in the prior art
Existing other horizontal lines/oblique line detection method, will not be described here.
Further, when being detected to correct option for the single-point answer image of modification or addition, it is necessary to by single-point answer
Image as input, based on training in advance to correct mistakes topic identification model, the target single-point answer image is identified,
Obtain corresponding recognition result.In this embodiment of the invention, specifically, in order to more accurately realize to target answer image
Identification, it is necessary first to the cutting of single hand-written character image is carried out to the whole character string in the target single-point answer image.
The limited network for being directed to that hand-written character string is established in each answer sample correct option is used in the present embodiment, travels through the target
Single-point answer image, obtain the subgraph of each single hand-written character in the target single-point answer image.By the target single-point
The hand-written character subgraph of each single-point character is as input in answer image, the topic identification mould of correcting mistakes obtained based on training in advance
Type, the target single-point answer image is identified, obtaining each single hand-written character in the target single-point answer image is
The probable value of each character, the maximum corresponding character of probable value is final recognition result.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, described pre-
The limited network first built is built in advance according to its corresponding correct option, including initial layers, end layer, each single hand-written character
Subelement layer and the absorbed layer being present between abovementioned layers：
Each single hand-written character subelement layer respectively according in its corresponding correct option each single character it is hand-written
Sample image is built；
Each absorbed layer redirects arc before existing to each single hand-written character subelement layer and end layer and redirected with after
Arc.
Each level, which jumps to each level and had, in above-mentioned specific embodiment, in the limited network belongs to this layer
Score value and redirect score value (typically belong to absorbed layer and have penalty values, belonging to character subelement has a matching score).It is described
Absorbed layer includes fil (noise absorption) and sil (Jing Yin sound absorption) node.
Further, due to treating the correct option of grade papers, it is known that being known a priori by which of topic character string needs
It is corresponding to be revised as needing which character string added between which character string, any two character strings, thus can utilize known correct
Answer information is handled target single-point answer image, you can by the limited network that builds in advance to target single-point answer figure
Cutting as carrying out single hand-written character, so as to obtain each subgraph comprising single hand-written character.
Specifically, the structure of the limited network built in advance is divided into following steps.The limited network is needed to every
Individual corresponding answer point character string builds a restriction network.As shown in schematic diagram 3, the restriction network includes initial layers S, terminated
Layer E, the absorbed layer FS in character between each single hand-written character subelement layer ai and foregoing all layers, and each absorbed layer is to institute
The single hand-written character subelement and end layer having, which exist, to be redirected arc (have front jumping and then jump), jumps to each level and have
Belong to the score value of this layer and redirect score value and (typically belong to absorbed layer and have penalty values, belonging to character subelement has a matching
Score).Each single hand-written character subelement layer is respectively according to each single character in its corresponding answer point correct option
Handwriting samples picture construction.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method, the base automatically
In the limited network built in advance corresponding with correct option, single hand-written character cutting is carried out to corresponding single-point answer image, obtained
Each subgraph of the single-point answer image is obtained, including：
Along the x-axis direction, the single-point answer is traveled through using the window of presetted pixel width, the single-point answer picture altitude
Image, the pixel for traveling through acquisition is inputted into the limited network built in advance corresponding with correct option, obtains the single-point answer figure
Each subgraph of picture.
In above-mentioned specific embodiment, based on network shown in Fig. 3, (a1, a2, a3, a4 in described Fig. 3 are corresponded to respectively
The image recognition model of " that " four characters), it is assumed that single character first is carried out to target single-point answer image " that " and cut
Point：Along the x-axis direction, presetted pixel width (such as 5 pixel wides), the window of the target single-point answer picture altitude are utilized
The target single-point answer image " that " is traveled through, the pixel input obtained and the target single-point answer image will be traveled through
The limited network of " that " corresponding advance structure；It is sharp respectively to mode of the currently received pixel along Fig. 3 from left to right
Decoded with the a1 to a4 single hand-written character subelement layer, if first single hand-written character confirms as " t ",
Identified by a1；It is not present and redirects if second single hand-written character " h " is successfully identified by a2, between a1 and a2.It is false
If target single-point is " tthat ", after first single hand-written character is really identified by a1, because of second single hand-written character still
It is so " t ", then a1 is carried out from jump.It is " txhat " to assume again that target single-point, when first single hand-written character is really identified by a1
Afterwards, because second single hand-written character is not any one in " that " in four characters, then one before a1 front jumpings to a1
In individual absorbed layer FS；Further assume that target single-point is " tat ", after first single hand-written character is really identified by a1,
Because second single hand-written character is " a ", then a3 is skipped to after a1.
By the above-mentioned limited network built in advance, can obtain as drawn a conclusion：Front jumping accounts for this from rising space symbol subelement
The ratio of character subelement sum, it is directly proportional to increase character quantity newly to correct option in single-point answer image；Rising space accords with after generation
Subelement accounts for the ratio of character subelement sum in the single-point answer image, directly proportional to the character quantity that correct option fails to write；
The pixel of the target single-point answer image acquisition is traveled through by the absorbed layer using the window of presetted pixel width along the x-axis direction
Absorbed pixel number, the ratio of the single-point answer image slices vegetarian refreshments sum is accounted for, with mistake in the target single-point answer image
Character quantity is also proportional by mistake.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, described to change
Mistake topic identification model is trained by following steps to be obtained：
Sample single-point answer image is subjected to single hand-written character cutting, obtains the various kinds book of sample single-point answer image
Image；
Described this subgraph of various kinds is identified the mark of result；
According to this subgraph of various kinds, using predetermined neural network topology structure and training algorithm, training is changed
Mistake topic identification model.
In above-mentioned specific embodiment, it is described build in advance correct mistakes topic the specific construction method of identification model it is as follows：First
Modification and addition answer view data are collected, and answer view data is labeled (including answer one hand write characters image
The recognition result of cutting and each hand-written character subelement marks)；Then determine to correct mistakes identification model topological structure (using existing
There are common neutral net, such as DNN, CNN)；Training data and the topological structure of determination are finally based on, using existing instruction
Practice algorithm (such as BP algorithm) correct mistakes the training of identification model.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, described
Obtain to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, each writing answer is carried out
Read and appraise, including：When correct option to be identified is deletes, it whether there is the identification knot deleted and answered according to single-point answer image
Fruit, acquisition read and appraise result accordingly；
When correct option to be identified is changes or added, carried out according to each subgraph of corresponding single-point answer image single
Comprising the probability that single hand-written character is respective symbols in jump information and each subgraph during hand-written character cutting, based on pre-
If reading and appraising method, obtain the single-point answer image at least one of read and appraise feature；Feature is read and appraised according to described, based on advance structure
That builds reads and appraises model, and obtain the target single-point answer image reads and appraises result.
In above-mentioned specific embodiment, based in all target single-point answer images each character subsample image read and appraise to
Go out the result of reading and appraising of whole topic of correcting mistakes, average that such as all target single-point answer images are read and appraised, linear weighted function.
Further, feature is read and appraised described in the basis, model is read and appraised based on what is built in advance, obtained the target single-point and answer
The result of reading and appraising of image is inscribed, including：The items are read and appraised into feature, as the input for reading and appraising model built in advance, obtained
Obtain the target single-point answer image reads and appraises result, such as fractional value etc..
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method automatically, described
Obtain to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, each writing answer is carried out
Read and appraise, including：When correct option to be identified is deletes, it whether there is the identification knot deleted and answered according to single-point answer image
Fruit, acquisition read and appraise result accordingly；
When correct option to be identified is changes or added, carried out according to each subgraph of corresponding single-point answer image single
Comprising the probability that single hand-written character is respective symbols in jump information and each subgraph during hand-written character cutting, based on pre-
If reading and appraising method, obtain the single-point answer image at least one of read and appraise feature；Feature is read and appraised according to described, based on advance structure
That builds reads and appraises model, and obtain the target single-point answer image reads and appraises result.
On the basis of any of the above-described specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises method, institute's commentary automatically
Feature is read including at least answer identification posterior probability feature and the coverage feature of character cutting decoding paths, additionally it is possible to including increasing
Write characters ratio characteristic, fail to write character ratio characteristic and by least one of absorption pixel ratio feature；Wherein：
The answer identifies posterior probability feature, is that the single hand-written character included in the single-point answer image is corresponding
The average value of character probabilities；
The coverage feature of the character cutting decoding paths, it is that the single-point answer image is carrying out single hand-written character
During cutting, the ratio of the single hand-written character quantity and hand-written character subelement layer number in the limited network that traverse through；
The increasing write characters ratio characteristic is the single-point answer image when carrying out single hand-written character cutting, front jumping
Or the ratio of character subelement sum in the single-point answer image is accounted for from rising space symbol subelement；
It is described to fail to write character ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, rear jump
Character subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described by absorption pixel ratio feature, be the single-point answer image when carrying out single hand-written character cutting, along x
Direction of principal axis travels through the pixel of the target single-point answer image acquisition by the absorbed layer institute using the window of presetted pixel width
Pixel number is absorbed, accounts for the ratio of the single-point answer image slices vegetarian refreshments sum.
As shown in figure 4, on the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises dress automatically
Put, including：
Acquisition module A1, single-point answer image corresponding to answer is respectively write on answer medium to be read for obtaining；Identify mould
Block A2, for entering using pre-set image detection method and/or based on default identification model to each single-point answer image got
Row image recognition, obtain recognition result；Modules A 3 is read and appraised, feature is read and appraised for obtaining to correspond to according to identification information, according to described
Read and appraise feature and read and appraise model based on default, each writing answer is read and appraised；The identification information include the recognition result and
Identify pilot process information.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device automatically, described to obtain
Modulus block is further used for：
According to the answer image of answer medium to be read and answer medium template corresponding with the answer medium to be read, obtain
Take single-point answer image corresponding to each writing answer on the answer medium to be read；Wherein, the answer medium template includes treating
The positional information of each character in examination question is answered, and according to correct option at least one single-point answer region set in advance, it is described
Single-point answer image corresponding to single-point answer region and obtain.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device, the knowledge automatically
Other module, is further used for：
During for correct option to delete, marked by detecting in corresponding single-point answer image with the presence or absence of at least one delete
Note, obtain the recognition result answered with the presence or absence of deletion；
When correct option is changes or added, based on the limited network built in advance corresponding with correct option, to corresponding
Single-point answer image carries out single hand-written character cutting, obtains each subgraph of the single-point answer image；Based on advance structure
Topic identification model of correcting mistakes, each subgraph is identified, obtains the single hand-written character included in each subgraph
Recognition result.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device automatically, described pre-
The limited network first built is built in advance according to its corresponding correct option, including initial layers, end layer, each single hand-written character
Subelement layer and the absorbed layer being present between abovementioned layers：
Each single hand-written character subelement layer respectively according in its corresponding correct option each single character it is hand-written
Sample image is built；
Each absorbed layer redirects arc before existing to each single hand-written character subelement layer and end layer and redirected with after
Arc.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device, the knowledge automatically
Other module, is further used for：
Along the x-axis direction, the single-point answer is traveled through using the window of presetted pixel width, the single-point answer picture altitude
Image, the pixel for traveling through acquisition is inputted into the limited network built in advance corresponding with correct option, obtains the single-point answer figure
Each subgraph of picture.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device automatically, described to change
Mistake topic identification model is trained by following steps to be obtained：
Sample single-point answer image is subjected to single hand-written character cutting, obtains the various kinds book of sample single-point answer image
Image；
Described this subgraph of various kinds is identified the mark of result；
According to this subgraph of various kinds, using predetermined neural network topology structure and training algorithm, training is changed
Mistake topic identification model.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device, institute's commentary automatically
Module is read, is further used for：
When correct option to be identified is deletes, it whether there is the recognition result deleted and answered according to single-point answer image,
Acquisition reads and appraises result accordingly；
When correct option to be identified is changes or added, carried out according to each subgraph of corresponding single-point answer image single
Comprising the probability that single hand-written character is respective symbols in jump information and each subgraph during hand-written character cutting, based on pre-
If reading and appraising method, obtain the single-point answer image at least one of read and appraise feature；Feature is read and appraised according to described, based on advance structure
That builds reads and appraises model, and obtain the target single-point answer image reads and appraises result.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises device, institute's commentary automatically
Feature is read including at least answer identification posterior probability feature and the coverage feature of character cutting decoding paths, additionally it is possible to including increasing
Write characters ratio characteristic, fail to write character ratio characteristic and by least one of absorption pixel ratio feature；Wherein：
The answer identifies posterior probability feature, is that the single hand-written character included in the single-point answer image is corresponding
The average value of character probabilities；
The coverage feature of the character cutting decoding paths, it is that the single-point answer image is carrying out single hand-written character
During cutting, the ratio of the single hand-written character quantity and hand-written character subelement layer number in the limited network that traverse through；
The increasing write characters ratio characteristic is the single-point answer image when carrying out single hand-written character cutting, front jumping
Or the ratio of character subelement sum in the single-point answer image is accounted for from rising space symbol subelement；
It is described to fail to write character ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, rear jump
Character subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described by absorption pixel ratio feature, be the single-point answer image when carrying out single hand-written character cutting, along x
Direction of principal axis travels through the pixel of the target single-point answer image acquisition by the absorbed layer institute using the window of presetted pixel width
Pixel number is absorbed, accounts for the ratio of the single-point answer image slices vegetarian refreshments sum.
On the basis of any above-mentioned specific embodiment of the present invention, there is provided one kind corrects mistakes to inscribe reads and appraises electronic equipment automatically.Ginseng
See Fig. 5, the topic of correcting mistakes reads and appraises electronic equipment automatically to be included：Processor (processor) 501, memory (memory) 502 and total
Line 503；
Wherein, processor 501 and memory 502 complete mutual communication by bus 503 respectively；
Processor 501 is used to call the programmed instruction in memory 502, to perform the topic of correcting mistakes that above-described embodiment is provided
It is automatic to read and appraise method, such as including：Obtain and single-point answer image corresponding to answer is respectively write on answer medium to be read；Using default
Image detecting method and/or each single-point answer image progress image recognition based on default identification model to getting, are known
Other result；Obtained to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, to each writing
Answer is read and appraised.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium storing program for executing, the non-transient computer readable storage medium
Matter stores computer instruction, and the computer instruction makes computer perform the automatic side of reading and appraising of topic of correcting mistakes that above-described embodiment is provided
Method, such as including：Obtain and single-point answer image corresponding to answer is respectively write on answer medium to be read；Utilize pre-set image detection side
Method and/or each single-point answer image progress image recognition based on default identification model to getting, obtain recognition result；According to
Identification information, which obtains to correspond to, reads and appraises feature, reads and appraises model based on default according to the feature of reading and appraising, each writing answer is commented
Read.
One of ordinary skill in the art will appreciate that：Realizing all or part of step of above method embodiment can pass through
Programmed instruction related hardware is completed, and foregoing program can be stored in a computer read/write memory medium, the program
Upon execution, the step of execution includes above method embodiment；And foregoing storage medium includes：ROM, RAM, magnetic disc or light
Disk etc. is various can be with the medium of store program codes.
The embodiments such as information exchange equipment described above are only schematical, wherein illustrate as separating component
Unit can be or may not be physically separate, can be as the part that unit is shown or may not be thing
Manage unit, you can with positioned at a place, or can also be distributed on multiple NEs.It can select according to the actual needs
Some or all of module therein is selected to realize the purpose of this embodiment scheme.Those of ordinary skill in the art are not paying wound
In the case of the work for the property made, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
Realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on such understanding, on
The part that technical scheme substantially in other words contributes to prior art is stated to embody in the form of software product, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including some fingers
Make to cause a computer equipment (can be personal computer, server, or network equipment etc.) to perform each implementation
Some Part Methods of example or embodiment.
Finally, the present processes are only preferable embodiment, are not intended to limit the protection model of the embodiment of the present invention
Enclose.All spirit in the embodiment of the present invention any modification, equivalent substitution and improvements made etc., should be included in within principle
Within the protection domain of the embodiment of the present invention.
Claims (18)
1. one kind is corrected mistakes, topic reads and appraises method automatically, it is characterised in that including：
Obtain and single-point answer image corresponding to answer is respectively write on answer medium to be read；
Image is carried out to each single-point answer image got using pre-set image detection method and/or based on default identification model
Identification, obtain recognition result；
Obtained to correspond to according to identification information and read and appraise feature, model is read and appraised based on default according to the feature of reading and appraising, each writing is answered
Case is read and appraised；The identification information includes the recognition result and identification pilot process information.
2. according to the method for claim 1, it is characterised in that described obtain respectively writes answer correspondingly on answer medium to be read
Single-point answer image the step of, including：
According to the answer image of answer medium to be read and answer medium template corresponding with the answer medium to be read, institute is obtained
State and single-point answer image corresponding to answer is respectively write on answer medium to be read；Wherein, the answer medium template includes waiting to answer examination
The positional information of each character in topic, and according to correct option at least one single-point answer region set in advance, the single-point
Answer image corresponding to single-point answer region and obtain.
3. according to the method for claim 1, it is characterised in that described using pre-set image detection method and/or based on pre-
If identification model carries out image recognition to each single-point answer image got, recognition result is obtained, including：
During for correct option to delete, marked by detecting in corresponding single-point answer image with the presence or absence of at least one delete,
Obtain the recognition result answered with the presence or absence of deletion；
When correct option is changes or added, based on the limited network built in advance corresponding with correct option, to corresponding single-point
Answer image carries out single hand-written character cutting, obtains each subgraph of the single-point answer image；Changed based on what is built in advance
Mistake topic identification model, is identified to each subgraph, obtains the knowledge of the single hand-written character included in each subgraph
Other result.
4. according to the method for claim 3, it is characterised in that the limited network built in advance according to corresponding to it just
True answer is built in advance, including initial layers, end layer, each single hand-written character subelement layer and the suction being present between abovementioned layers
Receive layer：
Each single hand-written character subelement layer is respectively according to the handwriting samples of each single character in its corresponding correct option
Picture construction；
Each absorbed layer redirects arc before existing to each single hand-written character subelement layer and end layer and redirects arc with after.
5. according to the method for claim 4, it is characterised in that it is described based on it is corresponding with correct option build in advance be limited
Network, single hand-written character cutting is carried out to corresponding single-point answer image, obtains each subgraph of the single-point answer image, bag
Include：
Along the x-axis direction, the single-point answer figure is traveled through using the window of presetted pixel width, the single-point answer picture altitude
Picture, the pixel for traveling through acquisition is inputted into the limited network built in advance corresponding with correct option, obtains the single-point answer image
Each subgraph.
6. according to the method for claim 3, it is characterised in that the topic identification model of correcting mistakes is obtained by following steps training
：
Sample single-point answer image is subjected to single hand-written character cutting, this subgraph of the various kinds of acquisition sample single-point answer image
Picture；
Described this subgraph of various kinds is identified the mark of result；
According to this subgraph of various kinds, using predetermined neural network topology structure and training algorithm, training obtains topic of correcting mistakes
Identification model.
7. according to the method for claim 3, it is characterised in that described to read and appraise feature, root according to identification information acquisition is corresponding
Model is read and appraised based on default according to the feature of reading and appraising, each writing answer is read and appraised, including：When correct option to be identified is to delete
Except when, result is read and appraised according to single-point answer image accordingly with the presence or absence of the recognition result answered, acquisition is deleted；
When correct option to be identified is changes or added, carried out according to each subgraph of corresponding single-point answer image single hand-written
Comprising the probability that single hand-written character is respective symbols in jump information and each subgraph during character cutting, commented based on default
Read method, obtain the single-point answer image at least one of read and appraise feature；Feature is read and appraised according to described, based on what is built in advance
Model is read and appraised, obtain the target single-point answer image reads and appraises result.
8. according to the method for claim 7, it is characterised in that the feature of reading and appraising is including at least answer identification posterior probability
The coverage feature of feature and character cutting decoding paths, additionally it is possible to including increasing write characters ratio characteristic, failing to write character ratio spy
Seek peace by least one of absorption pixel ratio feature；Wherein：
The answer identifies posterior probability feature, is that the single hand-written character included in the single-point answer image is respective symbols
The average value of probability；
The coverage feature of the character cutting decoding paths, it is that the single-point answer image is carrying out single hand-written character cutting
When, the ratio of the single hand-written character quantity and hand-written character subelement layer number in the limited network that traverse through；
It is described increasing write characters ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, front jumping or oneself
Rising space symbol subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described to fail to write character ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, rear rising space accords with
Subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described by absorption pixel ratio feature, be the single-point answer image when carrying out single hand-written character cutting, along x-axis side
The pixel that the target single-point answer image acquisition is traveled through to the window using presetted pixel width is absorbed by the absorbed layer
Pixel number, account for the ratio of the single-point answer image slices vegetarian refreshments sum.
9. one kind is corrected mistakes, topic reads and appraises device automatically, it is characterised in that including：
Acquisition module, single-point answer image corresponding to answer is respectively write on answer medium to be read for obtaining；
Identification module, for being answered using pre-set image detection method and/or based on default identification model each single-point got
Inscribe image and carry out image recognition, obtain recognition result；
Module is read and appraised, feature is read and appraised for obtaining to correspond to according to identification information, mould is read and appraised based on default according to the feature of reading and appraising
Type, each writing answer is read and appraised；The identification information includes the recognition result and identification pilot process information.
10. device according to claim 9, it is characterised in that the acquisition module is further used for：
According to the answer image of answer medium to be read and answer medium template corresponding with the answer medium to be read, institute is obtained
State and single-point answer image corresponding to answer is respectively write on answer medium to be read；Wherein, the answer medium template includes waiting to answer examination
The positional information of each character in topic, and according to correct option at least one single-point answer region set in advance, the single-point
Answer image corresponding to single-point answer region and obtain.
11. device according to claim 9, it is characterised in that the identification module, be further used for：
During for correct option to delete, marked by detecting in corresponding single-point answer image with the presence or absence of at least one delete,
Obtain the recognition result answered with the presence or absence of deletion；
When correct option is changes or added, based on the limited network built in advance corresponding with correct option, to corresponding single-point
Answer image carries out single hand-written character cutting, obtains each subgraph of the single-point answer image；Changed based on what is built in advance
Mistake topic identification model, is identified to each subgraph, obtains the knowledge of the single hand-written character included in each subgraph
Other result.
12. device as claimed in claim 11, it is characterised in that the limited network built in advance according to corresponding to it just
True answer is built in advance, including initial layers, end layer, each single hand-written character subelement layer and the suction being present between abovementioned layers
Receive layer：
Each single hand-written character subelement layer is respectively according to the handwriting samples of each single character in its corresponding correct option
Picture construction；
Each absorbed layer redirects arc before existing to each single hand-written character subelement layer and end layer and redirects arc with after.
13. device according to claim 12, it is characterised in that the identification module, be further used for：
Along the x-axis direction, the single-point answer figure is traveled through using the window of presetted pixel width, the single-point answer picture altitude
Picture, the pixel for traveling through acquisition is inputted into the limited network built in advance corresponding with correct option, obtains the single-point answer image
Each subgraph.
14. device according to claim 11, it is characterised in that the topic identification model of correcting mistakes is trained by following steps
Obtain：
Sample single-point answer image is subjected to single hand-written character cutting, this subgraph of the various kinds of acquisition sample single-point answer image
Picture；
Described this subgraph of various kinds is identified the mark of result；
According to this subgraph of various kinds, using predetermined neural network topology structure and training algorithm, training obtains topic of correcting mistakes
Identification model.
15. device according to claim 11, it is characterised in that it is described to read and appraise module, it is further used for：
When correct option to be identified is deletes, it whether there is the recognition result deleted and answered according to single-point answer image, obtain
Read and appraise result accordingly；
When correct option to be identified is changes or added, carried out according to each subgraph of corresponding single-point answer image single hand-written
Comprising the probability that single hand-written character is respective symbols in jump information and each subgraph during character cutting, commented based on default
Read method, obtain the single-point answer image at least one of read and appraise feature；Feature is read and appraised according to described, based on what is built in advance
Model is read and appraised, obtain the target single-point answer image reads and appraises result.
16. device according to claim 15, it is characterised in that the feature of reading and appraising identifies that posteriority is general including at least answer
The coverage feature of rate feature and character cutting decoding paths, additionally it is possible to including increasing write characters ratio characteristic, failing to write character ratio
Feature and by least one of absorption pixel ratio feature；Wherein：
The answer identifies posterior probability feature, is that the single hand-written character included in the single-point answer image is respective symbols
The average value of probability；
The coverage feature of the character cutting decoding paths, it is that the single-point answer image is carrying out single hand-written character cutting
When, the ratio of the single hand-written character quantity and hand-written character subelement layer number in the limited network that traverse through；
It is described increasing write characters ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, front jumping or oneself
Rising space symbol subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described to fail to write character ratio characteristic, be the single-point answer image when carrying out single hand-written character cutting, rear rising space accords with
Subelement accounts for the ratio of character subelement sum in the single-point answer image；
It is described by absorption pixel ratio feature, be the single-point answer image when carrying out single hand-written character cutting, along x-axis side
The pixel that the target single-point answer image acquisition is traveled through to the window using presetted pixel width is absorbed by the absorbed layer
Pixel number, account for the ratio of the single-point answer image slices vegetarian refreshments sum.
17. one kind is corrected mistakes, topic reads and appraises electronic equipment automatically, it is characterised in that including：
At least one processor；And
At least one memory being connected with the processor communication, wherein：
The memory storage has can be by the programmed instruction of the computing device, and the processor calls described program instruction energy
Enough perform method according to any one of claims 1 to 8.
18. a kind of non-transient computer readable storage medium storing program for executing, it is characterised in that the non-transient computer readable storage medium storing program for executing is deposited
Computer instruction is stored up, the computer instruction makes the computer perform method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710947823.9A CN107729936B (en) | 2017-10-12 | 2017-10-12 | Automatic error correction review method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710947823.9A CN107729936B (en) | 2017-10-12 | 2017-10-12 | Automatic error correction review method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107729936A true CN107729936A (en) | 2018-02-23 |
| CN107729936B CN107729936B (en) | 2020-12-08 |
Family
ID=61211192
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710947823.9A Active CN107729936B (en) | 2017-10-12 | 2017-10-12 | Automatic error correction review method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107729936B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108388895A (en) * | 2018-03-04 | 2018-08-10 | 南京理工大学 | A kind of paper answering card automatic processing method based on machine learning |
| CN110969052A (en) * | 2018-09-29 | 2020-04-07 | 杭州萤石软件有限公司 | Operation correction method and equipment |
| CN111008594A (en) * | 2019-12-04 | 2020-04-14 | 科大讯飞股份有限公司 | Error correction evaluation method, related equipment and readable storage medium |
| CN111104881A (en) * | 2019-12-09 | 2020-05-05 | 科大讯飞股份有限公司 | Image processing method and related device |
| WO2020177531A1 (en) * | 2019-03-04 | 2020-09-10 | 杭州大拿科技股份有限公司 | Question assistance method and system |
| CN112257703A (en) * | 2020-12-24 | 2021-01-22 | 北京世纪好未来教育科技有限公司 | Image recognition method, device, equipment and readable storage medium |
| CN112507879A (en) * | 2020-12-08 | 2021-03-16 | 科大讯飞股份有限公司 | Evaluation method, evaluation device, electronic equipment and storage medium |
| CN112926587A (en) * | 2021-02-19 | 2021-06-08 | 北京大米未来科技有限公司 | Text recognition method and device, readable storage medium and electronic equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102568475A (en) * | 2011-12-31 | 2012-07-11 | 安徽科大讯飞信息科技股份有限公司 | System and method for assessing proficiency in Putonghua |
| CN103903491A (en) * | 2014-02-14 | 2014-07-02 | 广东小天才科技有限公司 | Method and device for realizing writing check |
| CN104268603A (en) * | 2014-09-16 | 2015-01-07 | 科大讯飞股份有限公司 | Intelligent marking method and system for text objective questions |
| CN104364815A (en) * | 2012-10-31 | 2015-02-18 | Sk电信有限公司 | Apparatus and method for automatic scoring |
| CN105160342A (en) * | 2015-08-11 | 2015-12-16 | 成都数联铭品科技有限公司 | HMM-GMM-based automatic word picture splitting method and system |
| CN105183712A (en) * | 2015-08-27 | 2015-12-23 | 北京时代焦点国际教育咨询有限责任公司 | Method and apparatus for scoring English composition |
| CN106781784A (en) * | 2017-01-04 | 2017-05-31 | 王骁乾 | A kind of intelligence correction system |
| CN106846203A (en) * | 2017-03-09 | 2017-06-13 | 常州工学院 | A kind of intelligently reading method and system |
-
2017
- 2017-10-12 CN CN201710947823.9A patent/CN107729936B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102568475A (en) * | 2011-12-31 | 2012-07-11 | 安徽科大讯飞信息科技股份有限公司 | System and method for assessing proficiency in Putonghua |
| CN104364815A (en) * | 2012-10-31 | 2015-02-18 | Sk电信有限公司 | Apparatus and method for automatic scoring |
| US20150093737A1 (en) * | 2012-10-31 | 2015-04-02 | Sk Telecom Co., Ltd. | Apparatus and method for automatic scoring |
| CN103903491A (en) * | 2014-02-14 | 2014-07-02 | 广东小天才科技有限公司 | Method and device for realizing writing check |
| CN104268603A (en) * | 2014-09-16 | 2015-01-07 | 科大讯飞股份有限公司 | Intelligent marking method and system for text objective questions |
| CN105160342A (en) * | 2015-08-11 | 2015-12-16 | 成都数联铭品科技有限公司 | HMM-GMM-based automatic word picture splitting method and system |
| CN105183712A (en) * | 2015-08-27 | 2015-12-23 | 北京时代焦点国际教育咨询有限责任公司 | Method and apparatus for scoring English composition |
| CN106781784A (en) * | 2017-01-04 | 2017-05-31 | 王骁乾 | A kind of intelligence correction system |
| CN106846203A (en) * | 2017-03-09 | 2017-06-13 | 常州工学院 | A kind of intelligently reading method and system |
Non-Patent Citations (1)
| Title |
|---|
| YA ZHOU ET AL.: "An Automatic English Composition scoring model based on neural network algorithm", 《 2014 IEEE/ACIS 13TH INTERNATIONAL CONFERENCE ON COMPUTER AND INFORMATION SCIENCE (ICIS)》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108388895A (en) * | 2018-03-04 | 2018-08-10 | 南京理工大学 | A kind of paper answering card automatic processing method based on machine learning |
| CN108388895B (en) * | 2018-03-04 | 2022-02-18 | 南京理工大学 | Machine learning-based automatic processing method for test paper answer sheet |
| CN110969052A (en) * | 2018-09-29 | 2020-04-07 | 杭州萤石软件有限公司 | Operation correction method and equipment |
| WO2020177531A1 (en) * | 2019-03-04 | 2020-09-10 | 杭州大拿科技股份有限公司 | Question assistance method and system |
| CN111008594A (en) * | 2019-12-04 | 2020-04-14 | 科大讯飞股份有限公司 | Error correction evaluation method, related equipment and readable storage medium |
| CN111008594B (en) * | 2019-12-04 | 2024-04-19 | 科大讯飞股份有限公司 | Error-correction question review method, related device and readable storage medium |
| CN111104881A (en) * | 2019-12-09 | 2020-05-05 | 科大讯飞股份有限公司 | Image processing method and related device |
| CN111104881B (en) * | 2019-12-09 | 2023-12-01 | 科大讯飞股份有限公司 | Image processing method and related device |
| CN112507879A (en) * | 2020-12-08 | 2021-03-16 | 科大讯飞股份有限公司 | Evaluation method, evaluation device, electronic equipment and storage medium |
| CN112257703A (en) * | 2020-12-24 | 2021-01-22 | 北京世纪好未来教育科技有限公司 | Image recognition method, device, equipment and readable storage medium |
| CN112926587A (en) * | 2021-02-19 | 2021-06-08 | 北京大米未来科技有限公司 | Text recognition method and device, readable storage medium and electronic equipment |
| CN112926587B (en) * | 2021-02-19 | 2024-03-29 | 北京大米未来科技有限公司 | Text recognition method and device, readable storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107729936B (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107729936A (en) | One kind corrects mistakes to inscribe reads and appraises method and system automatically | |
| CN106781784A (en) | A kind of intelligence correction system | |
| CN105574090B (en) | A kind of filtering sensitive words method and system | |
| Davis et al. | End-to-end document recognition and understanding with dessurt | |
| CN110516248A (en) | Method for correcting error of voice identification result, device, storage medium and electronic equipment | |
| CN109101469A (en) | The information that can search for is extracted from digitized document | |
| CN109614944A (en) | A kind of method for identifying mathematical formula, device, equipment and readable storage medium storing program for executing | |
| WO2020218512A1 (en) | Learning model generating device, character recognition device, learning model generating method, character recognition method, and program | |
| US20120197908A1 (en) | Method and apparatus for associating a table of contents and headings | |
| CN111274239B (en) | Test paper structuring processing method, device and equipment | |
| CN101622620A (en) | The method and apparatus of control access computer systems and annotating media file | |
| CN107944718A (en) | A kind of business policy assessment system and method | |
| JP2010015571A (en) | Automated evaluation of overly repetitive word use in essay | |
| Valy et al. | A new khmer palm leaf manuscript dataset for document analysis and recognition: Sleukrith set | |
| CN111753120A (en) | Method and device for searching questions, electronic equipment and storage medium | |
| CN112541095A (en) | Video title generation method and device, electronic equipment and storage medium | |
| US11675966B2 (en) | Table of contents generation | |
| CN110490237B (en) | Data processing method and device, storage medium and electronic equipment | |
| CN114730241B (en) | Gesture and stroke recognition in touch user interface input | |
| CN110135225A (en) | Sample mask method and computer storage medium | |
| CN114419636A (en) | Text recognition method, device, equipment and storage medium | |
| Yu et al. | Aggregating crowdsourced and automatic judgments to scale up a corpus of anaphoric reference for fiction and Wikipedia texts | |
| Nederhof et al. | OCR of handwritten transcriptions of Ancient Egyptian hieroglyphic text | |
| CN106339726A (en) | Method and device for handwriting recognition | |
| EP3992923A2 (en) | System and method for identifying non-standard user interface object |
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 |