CN109583445A - Character image correction processing method, device, equipment and storage medium - Google Patents
Character image correction processing method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109583445A CN109583445A CN201811416986.5A CN201811416986A CN109583445A CN 109583445 A CN109583445 A CN 109583445A CN 201811416986 A CN201811416986 A CN 201811416986A CN 109583445 A CN109583445 A CN 109583445A
- Authority
- CN
- China
- Prior art keywords
- image
- angle
- detected
- text
- character
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/243—Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to field of computer technology, provide a kind of text character image correction processing method, device, equipment and storage medium, the text character image correction processing method include: obtain include text and text tilt angle sample image；Convolutional neural networks foundation structure is trained based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and using sample image, obtains the angle detection model for the text tilt angle in detection image；Obtain image to be detected comprising target text；In angle detection model, the angle for carrying out tilt angle to the target text in image to be detected is detected, and obtains the angle testing result of target text；According to angle testing result, according to preset correcting mode, processing is corrected to image to be detected, the target image after being corrected.The present invention can be improved the efficiency of correction character image.
Description
Technical field
The present invention relates to character image correction process technical field more particularly to a kind of character image correction processing method,
Device, equipment and storage medium.
Background technique
During identifying to image, it is inverted for being commonly present the angle of text in the image of identification, or figure
The case where other angles is presented in the tilt angle of text as in, if being directly easy to appear image recognition to images to be recognized
The situation of failure or identification inaccuracy, needs to be corrected image, traditional character image correction processing method is usual
The image that there are problems that the tilt angle of text is filtered out from a large amount of image to be identified by the method for artificial screening,
Then the measurement of the tilt angle of the text in image is carried out one by one to the image filtered out, and using other image softwares to depositing
It the correction process such as translated, folded or is overturn in the image of angle problem, a large amount of manpower and material resources, and survey calculation need to be expended
Process is cumbersome, and the whole character image correction process period is long, and the efficiency of the correction character image of image is caused to reduce.
Summary of the invention
Based on this, it is necessary in view of the above technical problems, provide a kind of character image correction processing method, device, equipment
And storage medium, to solve the problems, such as using conventional method to the corrected low efficiency of character image.
One kind being based on character image correction processing method, comprising:
Obtain the sample image comprising text and text tilt angle；
Based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and using the sample image to institute
It states convolutional neural networks foundation structure to be trained, obtains detecting mould for the angle of the text tilt angle in detection image
Type；
Obtain image to be detected comprising target text；
In the angle detection model, the angle for carrying out tilt angle to the target text in described image to be detected is examined
It surveys, obtains the angle testing result of the target text；
According to the angle testing result, according to preset correcting mode, processing is corrected to described image to be detected,
Target image after being corrected.
A kind of character image correction processing device, comprising:
Sample acquisition module, for obtaining the sample image comprising text and text tilt angle；
Model training module for being based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and makes
The convolutional neural networks foundation structure is trained with the sample image, is obtained for the text inclination in detection image
The angle detection model of angle；
Target Acquisition module, for obtaining image to be detected comprising target text；
Module of target detection, in the angle detection model, to the target text in described image to be detected into
The angle of line tilt angle detects, and obtains the angle testing result of the target text；
Image correction module is used for according to the angle testing result, according to preset correcting mode, to described to be detected
Image is corrected processing, the target image after being corrected.
A kind of computer equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, the processor realize above-mentioned character image correction process side when executing the computer program
Method.
A kind of computer readable storage medium, the computer-readable recording medium storage have computer program, the meter
Calculation machine program realizes above-mentioned character image correction processing method when being executed by processor.
Above-mentioned character image correction processing method, device, equipment and storage medium, by being based on deep learning framework establishment
Preset convolutional neural networks foundation structure, and convolutional neural networks foundation structure is instructed using the sample image got
Practice, obtains the angle detection model for the tilt angle of the text in detection image, and then in angle detection model, treat
Target text in detection image carries out the angle detection of tilt angle, obtains the angle value of the tilt angle of the target text,
It can guarantee the accurate extraction to the angle character of target text using trained angle detection model, realize to target text
The accurate detection of the tilt angle of word, can be avoided the method by artificial screening, from largely comprising text and text inclination
In the image of angle, the image that there are problems that the tilt angle of text is filtered out, figure is then carried out one by one to the image filtered out
The case where measurement of the tilt angle of text as in, reduces the testing time to the tilt angle of the text in image, then,
According to preset correcting mode, processing is corrected to image to be detected of the angle value for the tilt angle for obtaining target text,
Target image after being corrected, operating method is easy, and the low calculation amount of complexity is small, to improve the effect of correction character image
Rate.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by institute in the description to the embodiment of the present invention
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is an application environment schematic diagram of text image correction process method in one embodiment of the invention；
Fig. 2 is a flow chart of text image correction process method in one embodiment of the invention；
Fig. 3 is the implementation flow chart of step S4 in text image correction process method in one embodiment of the invention；
Fig. 4 is the implementation flow chart of step S401 in text image correction process method in one embodiment of the invention；
Fig. 5 is the implementation flow chart of step S404 in text image correction process method in one embodiment of the invention；
Fig. 6 is the implementation flow chart of step S5 in text image correction process method in one embodiment of the invention；
Fig. 7 is a schematic diagram of text image correction process device in one embodiment of the invention；
Fig. 8 is a schematic diagram of computer equipment in one embodiment of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
Fig. 1 shows application environment provided in an embodiment of the present invention, which includes server-side and client,
In, it is attached between server-side and client by network, client is used for the image that obtains image data, and will acquire
Data are sent to server-side, and client specifically can be, but not limited to various personal computers, laptop, smart phone, put down
Plate computer and portable wearable device；For server-side for handling image data, server-side can specifically use independent server
Or the server cluster of multiple server compositions is realized.Character image correction processing method application provided in an embodiment of the present invention
In server-side.
Referring to Fig. 2, Fig. 2 shows the implementation processes of character image correction processing method provided in this embodiment.It is described in detail such as
Under:
S1: the sample image comprising text and text tilt angle is obtained.
Specifically, the mode for obtaining the sample image comprising text and text tilt angle specifically can be manually import or
Person's self-timing is obtained from local data base, can also be other acquisition modes, herein with no restrictions.
Wherein, the sample image comprising text and text tilt angle is manually imported, specifically can be by receiving client
The image chained address of user's input at end, the corresponding preservation in the address is obtained from the image chained address includes text and text
The sample image of word tilt angle.
Wherein, self-timing obtains the sample image comprising text and text tilt angle from local data base and specifically may be used
To be to start timing acquisition task to obtain the sample image comprising text and text tilt angle, wherein timing acquisition task tool
Body can be to be read out according to the preset time interval.Preferably, time interval can be set to 20 minutes, 10 minutes or 5
Minute, but it is not limited to this, can specifically be configured according to the needs of practical application, herein with no restrictions.
Preferably, the acquisition modes that the present embodiment uses are that self-timing obtains.
Specifically, it by starting image timing acquisition task, is provided at pre-determined intervals and is spaced automatically from local data base
Read the sample image comprising text and text tilt angle.
S2: it is based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and using sample image to volume
Product neural net base structure is trained, and obtains the angle detection model for the text tilt angle in detection image.
In the present embodiment, deep learning frame includes caffe or caffe2 etc., wherein caffe2 is lightweight, module
Change and an expansible frame, including memory interface (blob), layer structure (layer) and structure link (net).Wherein,
Blob is the structure of arrays and unified memory interface of the standard of entire frame.Layer structure basis of the layer as modeling and calculating,
Set and link of the net as layer.The datail description of blob information is how to store and exchange between layers and nets
's.Caffe2 is particular by using the modeling method of oneself to define network in layer, i.e., by network by inputting
Data carry out bottom-up define of entire model to loss layer.The information such as data and partial derivative forward, backward in a network
Flowing, and caffe2 is stored using blob, is exchanged and manipulate these information.
Wherein, preset convolutional neural networks foundation structure is constructed according to practical business demand, is such as rolled up
The structure of product neural network visual geometric group (Visual Geometry Group-16, VGG-16), for the sample got
This image is learnt, and obtains can be used in the angle detection model of the text tilt angle in detection image, wherein angle inspection
Surveying model includes input layer for pre-processing to image；Convolutional network layer is used to carry out image the angle of the text in image
Spend feature extraction；Fully connected network network layers are according to the angle character extracted, with four kinds of predetermined angle classifications defined in classification layer
Carry out DUAL PROBLEMS OF VECTOR MAPPING, the corresponding angle character vector of output angle feature, i.e., every kind of component corresponding one in angle character vector
Kind predetermined angle classification；Classification layer comprising four kinds of predetermined angle class declarations is used for the angle character vector that basis obtains, into
The calculating of the score value of every kind of angle classification of row, the angle classification of the text in image is determined according to calculated result, wherein four kinds pre-
If angle classification can specifically include but be not limited to the tilt angle of text as 0 degree, 90 degree, 180 degree and 270 degree, do not do herein
Limitation.
Specifically, the present embodiment uses the structure based on caffe2 framework establishment convolutional neural networks visual geometric group, tool
Body can carry out determining for each layer of structure in structure of the layer in caffe2 frame to convolutional neural networks visual geometric group
Justice is linked each layer of structure defined by structure link (net), and will by the blob in caffe2 frame
The layer structure that chain connects is stored, the structure of the VGG-16 put up, in the knot of the VGG-16 of caffe2 framework establishment
Structure, logical construction is well arranged, and parameter definition is clear, can be according to the parameter position in parameter definition quick lock in structure, just
In the modification of the structure of VGG-16 and perfect, then, it is trained, is had using structure of the sample image got to VGG-16
Body can be the feature for extracting sample image in the structure of convolutional neural networks visual geometric group, by traditional random
Gradient descent algorithm, the cost function that will include in each layer of the structure of convolutional neural networks visual geometric group, is tied according to layer
The direction of structure from front to back, calculates the corresponding cost function value of each layer, which can be used in subsequent each layer
Error-sensitivity calculating, then, by traditional back-propagation algorithm, by the structure of convolutional neural networks visual geometric group
Each layer according to layer structure from rear to preceding direction calculate the error-sensitivity in each layer, finally, will be calculated
The weighted value and offset for including in each layer of error-sensitivity, for the original weighted value and offset in update step structure
Amount, obtains angle detection model.
S3: image to be detected comprising target text is obtained.
In the present embodiment, image to be detected comprising target text is the figure for only including text and text tilt angle
Picture.
Specifically, the mode for obtaining image to be detected comprising target text can specifically include but be not limited to receive user
The image to be detected comprising target text or self-timing uploaded is obtained from third party's image data base comprising target text
Image to be detected etc. of word, can also be other acquisition modes, herein with no restrictions, wherein third party's image data base is specific
It can be the database of network image platform.
S4: in angle detection model, the angle for carrying out tilt angle to the target text in image to be detected is detected, and is obtained
To the angle testing result of target text.
Specifically, the image to be detected comprising target text got in step s3 trained angle is inputted to examine
It surveys in model, carrying out pretreatment by image to be detected that the input layer in angle detection model includes target text can reject
Partial redundance image information, the image of the target text clearly shown, convolutional network layer carry out pretreated image
The angle character of target text extracts, and can accurately extract the angle character of target text, then, by the angle extracted spy
Sign carries out DUAL PROBLEMS OF VECTOR MAPPING, output angle feature by fully connected network network layers, with four kinds of predetermined angle classifications defined in classification layer
Corresponding angle character vector, and by carrying out every kind of angle class in the classification layer comprising four kinds of predetermined angle class declarations
The calculating of other score value determines the target angle classification of the target text in image according to calculated result, by the target angle class
Angle testing result of other angle value as target text, such as the inclination that the target angle classification for assuming to obtain is target text
Angle is 180 degree, then using 180 degree as the angle testing result of the target text.
It should be noted that carrying out angle detection to the text in image by trained angle detection model, pass through
Convolutional network layer in angle detection model carries out angle character extraction to the text in image, can accurately distinguish out in image
Text and non-legible background, and then accurately extract the angle character of text, avoid the text in the image manually chosen, and right
The tilt angle of text carries out the measurement error in manual measurement, can be improved the accuracy rate of the angle detection to text.
S5: according to angle testing result, according to preset correcting mode, processing is corrected to image to be detected, is obtained
Target image after correction.
In the present embodiment, preset correcting mode can specifically include but be not limited to coordinate conversion, image is folded or turned over
Turn, herein with no restrictions.
Specifically, according to the angle testing result got in step S4, with the position of the lower-left angular vertex of image to be detected
A vertical rectangular coordinate system is set up, using the lower-left angular vertex as coordinate origin, then, according to the angle detection knot got
Angle value in fruit obtains the predetermined angle direction of the corresponding predetermined angle classification of the angle value, then can be according to the preset angle
Direction and angle value are spent, determines the angle between the y-axis of the target text and rectangular coordinate system in image to be detected, then, with
Origin is rotation center, and according to predetermined angle direction by the degree of image to be detected rotation angle, obtained image is after correcting
Target image.
In the present embodiment, by being based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and make
Convolutional neural networks foundation structure is trained with the sample image got, obtains inclining for the text in detection image
The angle detection model of rake angle, and then in angle detection model, inclination angle is carried out to the target text in image to be detected
The angle of degree detects, and obtains the angle value of the tilt angle of the target text, can be protected using trained angle detection model
The accurate extraction to the angle character of target text is demonstrate,proved, realizes to the accurate detection of the tilt angle of the target text, can keep away
Exempt from the method by artificial screening, from a large amount of image comprising text and text tilt angle, filters out that there are texts
Then the image of tilt angle problem carries out the feelings of the measurement of the tilt angle of the text in image one by one to the image filtered out
Condition reduces the testing time to the tilt angle of the text in image, then, according to preset correcting mode, to obtaining target
Image to be detected of the angle value of the tilt angle of text is corrected processing, the target image after being corrected, operating method
Simplicity, the low calculation amount of complexity is small, to improve the efficiency of correction character image.
In one embodiment, as shown in figure 3, in step S4, angle detection model includes input layer, convolutional network layer, complete
Network layer and the classification layer comprising four kinds of predetermined angle class declarations are connected, i.e., in angle detection model, to image to be detected
In target text carry out the angle detection of tilt angle, the angle testing result for obtaining target text specifically includes following step
It is rapid:
S401: according to the definition of input layer in angle detection model, image to be detected is pre-processed, is pre-processed
Original image afterwards.
Specifically, due to the influence of the factors such as the writing of the angle of shooting, distance and text, what is got is to be detected
Image may exist it is of low quality, directly in image to be detected target text carry out angle detection, be easy to cause angle
The efficiency and accuracy rate of detection reduce, therefore in order to realize the quickly and effectively angle detection to image to be detected, the present embodiment
By the input layer of preparatory trained convolutional neural networks model define in image processing method, image to be detected is carried out
Pretreatment obtains that treated only and includes the original image of text, can enhance the detectability and maximum limit to target text
Degree ground simplifies data, reduces the operand of angle detection of the subsequent step to target text, improves to target in image to be detected
The efficiency and accuracy rate of the angle detection of text, to improve the correction efficiency to character image.
Wherein, carrying out pretreatment to image to be detected can specifically include random noise, normalization in smooth elimination image
The image processing process such as image and recovery correction image degeneration.
Wherein, random noise in image is smoothly eliminated to refer to and become image outline or lines while eliminating noise
It is smudgy, it is to not destroy the essence of the important angle information in image after guarantee processing while eliminating redundancy
Degree, common smoothing method have median method, local averaging method and k neighbour's method of average, sometimes also application space frequency domain band logical
Filtering method, wherein the regional area size of image can be fixed in local averaging method, be also possible to point by point with gray scale
It is worth size variation.
Wherein, normalized image refers to that certain factors or the influence converted to image properties can pass through normalized
It is eliminated or weakens, so that a kind of these properties of image graphics standard form with invariance under given transformation,
For example, certain properties of image, such as the area and perimeter of image, transform normalization is carried out to image so that the area of image and
Perimeter has constant property under the transformation that coordinate rotates, so that the area and perimeter of image can be selected as measurement image
Important evidence.Common method for normalizing includes gray scale normalization, geometrical normalization and transform normalization.
S402: the inclination according to the definition of convolutional network layer in angle detection model, to the target text in original image
Angle carries out angle character extraction, obtains the angle character matrix of the target text comprising angle character value.
Specifically, according to the definition of convolutional network layer in convolutional neural networks model, target text is carried out to original image
Angle character extract be to be able to effectively extract original image in target text tilt angle important feature information,
The information for rejecting the non-angled feature in original image, is then combined the feature extracted, and it is corresponding to obtain original image
N*N angle character matrix, be to avoid important angle special in order to which the information to the important angle character extracted is protected
The information of sign is lost, and the feature extracted is saved with a matrix type, so that the important angle in image is special
Digitization is levied, can be improved computational efficiency, to improve the efficiency corrected to character image.
Wherein, according to the definition of convolutional network layer in convolutional neural networks model, the feature of angle is carried out to original image
It extracts and feature combination specifically can be original image carrying out convolutional calculation, the numerical value being calculated is as the original image
Angle character value, then, the angle character value that will be calculated, according to the pre-set feature in the definition of convolutional network layer
Combination carries out the combination of angle character value, obtains the angle character matrix of the N*N of original image.
S403: according to the definition of fully connected network network layers in angle detection model, according to preset dimension map condition, diagonally
It spends eigenmatrix and carries out dimension map, obtain corresponding four dimensional feature vector of angle character matrix, wherein in four dimensional feature vectors
The position of each component correspond to a kind of predetermined angle classification.
In the present embodiment, preset dimension map condition is the angle comprising a large amount of angle character values for reducing acquisition
The dimension for spending eigenmatrix, can buffer a large amount of angle character information, improve the extraction accuracy to angle character, tool
Body can be configured according to practical application request, herein with no restrictions.
When specifically, by angle character Input matrix fully connected network network layers, since fully connected network network layers compare convolutional network
Layer more considers global information, it is therefore desirable to which the angle character matrix for having local message originally is all mapped to preset dimension
In, a large amount of angle character information can be buffered, reduce subsequent step to the calculation amount of angle character, improve angle inspection
The efficiency of survey, to improve the correction efficiency to character image, such as common dimension is 4096 dimensions.
For example, in one embodiment, input fully connected network network layers is the angle character square of a 7*7*512 dimension
The eigenmatrix of the 7*7*512 dimension first can be converted to the 1*1 dimensional characteristics vector that length is 25088 by battle array, if it is default
Dimension map condition be " eigenmatrix that will acquire is mapped in 4096 dimensions ", then by length be 25088 1*1 dimension
It, i.e., can be by angle character by angle character matrix compression to original 1/5th on maps feature vectors to 4096 dimensions
Characteristic information in matrix is shared.
Further, due to only carrying out primary preset dimension map, the process of existing characteristics buffering is shorter, is unfavorable for mould
Accurate refinement of the type to the angle character of target text, so a dimension map can be carried out in fully connected network network layers again, into
One step angle eigenmatrix is buffered, the dimension of angle character matrix is reduced, reduces the operand of angle character.
Further, since the purpose of fully connected network network layers is and the classification layer comprising four kinds of predetermined angle class declarations
Classification task is associated, in order to reduce the learning pressure of classification layer, is established by fully connected network network layers and is defined with classification layer
Four kinds of predetermined angle classifications mapping relations, specifically can be by by dimension twice buffering after dimension have already decreased to one
The angle character vector for determining degree carries out DUAL PROBLEMS OF VECTOR MAPPING according to four kinds of predetermined angle classifications, establishes every in angle character vector
Kind component corresponds to a kind of relationship of preset angle classification, and obtaining a dimension is four-dimensional angle character vector.
S404: according to the definition for layer of classifying in angle detection model, the score of each component of four dimensional feature vectors is calculated.
Specifically, according to the definition for layer of classifying in angle detection model, obtaining for each component of four dimensional feature vectors is calculated
Dividing specifically can be the four-dimension angle character vector according to obtained in step S403, each component in the four-dimension angle character vector
Position correspond to a kind of predetermined angle classification, therefore a preset angle weighted value can be arranged to each component positions, so
Afterwards, it calculates the product between each component of four dimensional feature vector and corresponding preset angle weighted value, and will obtain
Score of the product as each component of four dimensional feature vector, can also be by other calculations calculate four dimensional features to
The score of each component of amount, is not particularly limited herein.
S405: the angle value of the corresponding predetermined angle classification of the maximum score of numerical value, the detection as target text are chosen
As a result.
Specifically, the score of each characteristic component is compared, obtains the maximum score of numerical value in each characteristic component,
It then obtains the corresponding component of the maximum score of the numerical value, by the angle value of the corresponding predetermined angle classification in the position of the component,
As the inclined angle value of the target text, the i.e. testing result of target text.
In the present embodiment, image to be detected is carried out pre- by elder generation according to the definition of input layer in angle detection model
Processing, obtains pretreated original image, can be improved the quality of image to be detected, and then roll up according in angle detection model
The definition of product network layer carries out angle character extraction to the tilt angle of the target text in original image, obtains comprising angle
The angle character matrix of the target text of characteristic value can effectively extract the weight of the tilt angle of the target text in original image
Characteristic information is wanted, the information of the non-angled feature in original image is rejected, guarantees the accuracy rate of angle character, further, root
It defines according to the fully connected network network layers of angle detection model, according to preset dimension map condition, angle eigenmatrix is tieed up
Degree mapping, obtains corresponding four dimensional feature vector of angle character matrix, reduces the angle comprising a large amount of angle character values of acquisition
The dimension of eigenmatrix can buffer a large amount of angle character information, be defined according to the classification layer of angle detection model,
The score of each component of four dimensional feature vectors is calculated, the angle of the corresponding predetermined angle classification of the maximum score of numerical value is chosen
Value on the basis of fully connected network network layers have carried out dimension map, reduces classification layer as the testing result of target text
Pressure is practised, reduces operand, improves the angle detection efficiency of target text, from the correction efficiency for improving character image.
In one embodiment, as shown in figure 4, in step 401, i.e., according to the definition of input layer in angle detection model,
Image to be detected is pre-processed, pretreated original image is obtained and specifically comprises the following steps:
S4011: according to preset image scaling mode, image scaling is carried out to image to be detected, obtains base image.
Specifically, according to preset image scaling mode, carrying out image scaling processing to image to be detected specifically be can be
Equal proportion scaling is carried out to image to be detected, i.e., long and width zooms in and out in the same scale, obtains the scaling figure of h*j pixel
Picture, the zoomed image of the h*j pixel are the base image obtained after handling, wherein h is the side length of preset short side.
It is readily appreciated that ground, image to be detected is rectangular image, is scaled by equal proportion, and the short side of image to be detected is scaled
To h pixel size, feature extraction is rapidly and accurately carried out in order to subsequent.
Wherein, the size of h can be configured according to actual needs, be not specifically limited herein.
For example, in a specific embodiment, image to be detected size got is 640 × 360, unit is a pixel
Point, i.e. long side length are 640 pixels, and bond length is the rectangular image of 360 pixels, and preset h is 128 pixels
Point passes through uniform zoom to the rectangular image, and obtaining size is 228 × 128, and unit is the base image of a pixel.
S4012: by base image, image cropping is carried out according to preset cutting method, obtains standard picture.
Specifically, the base image of h*j pixel obtained in step S4011 is carried out according to preset image cropping mode
Image cropping specifically can be base image carrying out equal proportion cutting, i.e., long and width is cut in the same scale, obtain
The cutting image of t*s pixel, as standard picture, wherein t is length of short sides.
Continue to obtain the size of base image in step S4011 as 228 × 128 pixels, unit is the base of a pixel
It is illustrated for plinth image, the long side for being 228 to pixel number is cut, and is cut out according to 220 × 120 ratio
It cuts, obtaining 1 size is 220 × 120, and unit is the standard picture of a pixel.
S4013: carrying out mean value and normalized to standard picture, and will treated standard picture as to be detected
The original image of image.
Specifically, it before extracting feature, needs to carry out data prediction to each standard picture, in the embodiment of the present invention
In, preferred data preprocessing method is first to carry out average value processing, then be normalized, and is accelerated after normalized
Gradient declines the speed for seeking optimal solution, that is, improves the efficiency of processing, meanwhile, be conducive to improve detection accuracy, improve to be checked
The efficiency of the angle detection of target text in altimetric image, to improve the efficiency of character image correction.
Wherein, normalized is including but not limited to simple scalability；Sample-by-sample mean value abatement, also referred to as removal direct current point
Amount；Feature normalization makes all features in data acquisition system all have zero-mean and unit variance etc..
Preferably, method for normalizing used in the embodiment of the present invention is to be normalized by feature normalization.
Wherein, it goes average value processing to refer to that the data of every dimension subtract itself mean value, data can be made in each dimension in this way
It is upper that there is similar width, certain increase data distribution range can be played.
For example, in a specific embodiment, the grapholect image got includes the feature of two dimensions, respectively
Feature A and feature B, the range of feature A are 8 to 1000, and the range of feature B is 23 to 10006, by going at mean value and normalization
It is 0 to 1 by the range reduction of feature A and feature B after reason.
In the present embodiment, by carrying out equal proportion scaling to image to be detected, the base that picture size is h*j pixel is obtained
Plinth image, and base image is cut, the image of t*s pixel size is obtained, as standard picture, and then to standard picture
Mean value and normalized are carried out, the original image of standard picture is obtained, so that standard picture has unified image size
With reasonable parameter area, is conducive to the subsequent angle character for rapidly and accurately carrying out target text in image to be detected and extracts,
The efficiency for improving the angle detection of target text in image to be detected, to improve the efficiency corrected to character image.
In one embodiment, as shown in figure 5, in step s 404, i.e., according to the definition for layer of classifying in angle detection model,
The score for calculating each component of four dimensional feature vectors specifically comprises the following steps:
S4041: each component of four dimensional feature vectors is obtained.
Specifically, according to obtaining four dimensional feature vectors in step S403, according to each component position by four dimensional features to
Each component in amount extracts, the calculating for convenience of subsequent step to the score of component, can be according to the position of each component
Each component that setting will acquire is marked, such as " the first component " or " third component ", herein with no restrictions.
S4042: using following calculation formula, calculates the score p of each component in four dimensional feature vectors:
Wherein, xiFor the component in four dimensional feature vectors, i is i-th of component in four dimensional feature vectors.
Specifically, by each component of four dimensional feature vectors got from step S4041, substitute into respectively in formula into
Row calculates, first according to the numerical value of each component got, can calculate four components index and, then, according to calculating
The index that arrives and, calculate separately the score of each component, the score p of each component of four-dimensional feature vector can be accessed.
For example, in one embodiment, if the one-component got is 1, then by the component substitute into formula into
Row calculate, if the index for four components being calculated and be 0.078, and then can be calculated the component be 1 score
It is 0.0287.
In the present embodiment, by obtaining each component of four dimensional feature vectors, then, the component that will acquire generation respectively
Enter and calculated into formula, the score p of each component can be quickly obtained by scoring function formula, improves target text
Angle detection efficiency, to improve the correction efficiency of character image.
In one embodiment, as shown in fig. 6, in step s 5, i.e., according to angle testing result, according to preset correction side
Formula is corrected processing to image to be detected, and the target image after being corrected specifically comprises the following steps:
S501: using the central point of image to be detected as axis, according to the angle value in angle testing result to image to be detected
The angle rotation for carrying out preset direction, obtains postrotational base correction image.
In the present embodiment, preset direction specifically can be counterclockwise, or clockwise, specifically can be according to reality
Border application demand is configured, herein with no restrictions.
When specifically, according to pre-processing in step S401 to image to be detected, image to be detected can detecte
The position of central point, then can be using the central point of image to be detected as the rotary shaft of image to be detected, then, according to step S4
In angle testing result in angle value, obtain the preset direction of the definition of the predetermined angle classification of the angle value, then according to
The preset direction, image to be detected is rotated, and the degree of rotation is the angle value in angle testing result, after obtaining rotation
Image, and image will be corrected based on the image.
S502: base correction image and preset blank background template are synthesized, and using the image after synthesis as
Target image, wherein the size of blank background template and the size of image to be detected are identical.
Specifically, the base correction image obtained in step S501 is synthesized with preset blank background template,
It specifically can be and base correction image be inserted into size preset blank background identical with the size of image to be detected
In template, base correction image in template is then selected, Transparent color is set by the background color of the image, target can be obtained
Text fusion target image in preset blank background template.
In the present embodiment, by using the central point of image to be detected as axis, according to the angle value in angle testing result
The angle rotation that preset direction is carried out to image to be detected, then carries out postrotational image and preset blank background template
Synthesis, the target image of correct direction needed for obtaining user realize the quick correction to image to be detected, improve to text figure
The correction efficiency of picture.
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.
In one embodiment, a kind of character image correction processing device is provided, the text image correction process device with it is upper
Text image correction process method in embodiment is stated to correspond.As shown in fig. 7, the text image correction process device includes sample
This acquisition module 701, model training module 702, Target Acquisition module 703, module of target detection 704 and image correction module
705.Detailed description are as follows for each functional module:
Sample acquisition module 701, for obtaining the sample image comprising text and text tilt angle；
Model training module 702, for being based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and
Convolutional neural networks foundation structure is trained using sample image, is obtained for the text tilt angle in detection image
Angle detection model；
Target Acquisition module 703, for obtaining image to be detected comprising target text；
Module of target detection 704, for being tilted to the target text in image to be detected in angle detection model
The angle of angle detects, and obtains the angle testing result of target text；
Image correction module 705 is used for according to angle testing result, according to preset correcting mode, to image to be detected
It is corrected processing, the target image after being corrected.
Further, model training module 702 includes:
Image processing unit 7041 carries out image to be detected for the definition according to input layer in angle detection model
Pretreatment, obtains pretreated original image；
Feature extraction unit 7042, for the definition according to convolutional network layer in angle detection model, in original image
Target text tilt angle carry out angle character extraction, obtain comprising angle character value target text angle character square
Battle array；
Dimension map unit 7043, for the definition according to fully connected network network layers in angle detection model, according to preset
Dimension map condition carries out dimension map to angle eigenmatrix, obtains corresponding four dimensional feature vector of angle character matrix,
In, the position of each component in four dimensional feature vectors corresponds to a kind of predetermined angle classification；
Score calculation unit 7044 calculates four dimensional feature vectors for the definition according to layer of classifying in angle detection model
Each component score；
As a result make for choosing the angle value of the corresponding predetermined angle classification of the maximum score of numerical value output unit 7045
For the testing result of target text.
Further, image processing unit 7041 includes:
Image scaling subelement 70411, for carrying out image contracting to image to be detected according to preset image scaling mode
It puts, obtains base image；
Image cropping subelement 70412, for carrying out base image image cropping according to preset cutting method, obtaining
To standard picture；
Image generates subelement 70413, for carrying out mean value and normalized to standard picture, and by treated
Original image of the standard picture as image to be detected.
Further, score calculation unit 7044 includes:
Component obtains subelement 70441, for obtaining each component of four dimensional feature vectors；
Formula computation subunit 70442 calculates each component in four dimensional feature vectors for using following calculation formula
Score p:
Wherein, xiFor the component in four dimensional feature vectors, i is i-th of component in four dimensional feature vectors.
Further, image correction module 705 includes:
Image rotation unit 7051, for using the central point of image to be detected as axis, according to the angle in angle testing result
The angle that angle value carries out preset direction to image to be detected rotates, and obtains postrotational base correction image；
Image composing unit 7052, for base correction image and preset blank background template to be synthesized, and will
Image after synthesis is as target image, wherein the size of blank background template and the size of image to be detected are identical.
Specific restriction about character image correction processing device may refer to above for character image correction process
The restriction of method, details are not described herein.Modules in above-mentioned character image correction processing device can be fully or partially through
Software, hardware and combinations thereof are realized.Above-mentioned each module can be embedded in the form of hardware or independently of the place in computer equipment
It manages in device, can also be stored in a software form in the memory in computer equipment, in order to which processor calls execution or more
The corresponding operation of modules.
In one embodiment, a kind of computer equipment is provided, which can be server, internal junction
Composition can be as shown in Figure 8.The computer equipment include by system bus connect processor, memory, network interface and
Database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory packet of the computer equipment
Include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data
Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The database of machine equipment is for saving image data.The network interface of the computer equipment is used to pass through network with external terminal
Connection communication.To realize a kind of character image correction processing method when the computer program is executed by processor.
In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, processor realize above-described embodiment character image school when executing computer program
The step of positive processing method, such as step S1 shown in Fig. 2 to step S5.Alternatively, processor is realized when executing computer program
The function of each module/unit of text image correction process device in above-described embodiment, such as module 701 shown in Fig. 7 is to module
705 function.To avoid repeating, which is not described herein again.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Text image correction process method in above method embodiment is realized when machine program is executed by processor, alternatively, the computer journey
The function of each module/unit in text image correction process device in above-mentioned apparatus embodiment is realized when sequence is executed by processor.
To avoid repeating, which is not described herein again.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink), DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.
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 should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of character image correction processing method, which is characterized in that the character image correction processing method includes:
Obtain the sample image comprising text and text tilt angle；
Based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and using the sample image to the volume
Product neural net base structure is trained, and obtains the angle detection model for the text tilt angle in detection image；
Obtain image to be detected comprising target text；
In the angle detection model, the angle for carrying out tilt angle to the target text in described image to be detected is detected,
Obtain the angle testing result of the target text；
According to the angle testing result, according to preset correcting mode, processing is corrected to described image to be detected, is obtained
Target image after correction.
2. character image correction processing method as described in claim 1, which is characterized in that the angle detection model includes defeated
Enter layer, convolutional network layer, fully connected network network layers and the classification layer comprising four kinds of predetermined angle class declarations, it is described in the angle
In detection model, the angle for carrying out tilt angle to the target text in described image to be detected is detected, and obtains the target text
The angle testing result of word includes:
According to the definition of input layer described in the angle detection model, described image to be detected is pre-processed, is obtained pre-
Treated original image；
According to the definition of convolutional network layer described in the angle detection model, to the target text in the original image
Tilt angle carry out angle character extraction, obtain comprising angle character value the target text angle character matrix；
According to the definition of fully connected network network layers described in the angle detection model, according to preset dimension map condition, to institute
It states angle character matrix and carries out dimension map, obtain corresponding four dimensional feature vector of the angle character matrix, wherein described four
The position of each component in dimensional feature vector corresponds to a kind of predetermined angle classification；
The definition of classification layer according to the angle detection model, calculates each of described four dimensional feature vector component
Score；
Choose the angle value of the corresponding predetermined angle classification of the maximum score of numerical value, the inspection as the target text
Survey result.
3. character image correction processing method as claimed in claim 2, which is characterized in that described to detect mould according to the angle
The definition of input layer described in type pre-processes described image to be detected, obtains pretreated original image and includes:
According to preset image scaling mode, image scaling is carried out to described image to be detected, obtains base image；
By the base image, image cropping is carried out according to preset cutting method, obtains standard picture；
Mean value and normalized carried out to the standard picture, and will treated standard picture as the mapping to be checked
The original image of picture.
4. character image correction processing method as claimed in claim 2, which is characterized in that described to detect mould according to the angle
The definition of classification layer described in type, the score for calculating each of described four dimensional feature vector component include:
Obtain each of four dimensional feature vector component；
Using following calculation formula, the score p of each component in four dimensional feature vector is calculated:
Wherein, xiFor the component in four dimensional feature vector, i is i-th of component in four dimensional feature vector.
5. character image correction processing method as described in claim 1, which is characterized in that described detected according to the angle is tied
Fruit is corrected processing to described image to be detected according to preset correcting mode, and the target image after being corrected includes:
Using the central point of described image to be detected as axis, according to the angle value in the angle testing result to described to be checked
Altimetric image carries out the angle rotation of preset direction, obtains postrotational base correction image；
The base correction image and preset blank background template are synthesized, and using the image after synthesis as the mesh
Logo image, wherein the size of the blank background template is identical as the size of described image to be detected.
6. a kind of character image correction processing device, which is characterized in that the character image correction processing device includes:
Sample acquisition module, for obtaining the sample image comprising text and text tilt angle；
Model training module for being based on the preset convolutional neural networks foundation structure of deep learning framework establishment, and uses institute
It states sample image to be trained the convolutional neural networks foundation structure, obtain for the text tilt angle in detection image
Angle detection model；
Target Acquisition module, for obtaining image to be detected comprising target text；
Module of target detection, for being inclined in the angle detection model to the target text in described image to be detected
The angle of rake angle detects, and obtains the angle testing result of the target text；
Image correction module is used for according to the angle testing result, according to preset correcting mode, to described image to be detected
It is corrected processing, the target image after being corrected.
7. character image correction processing device as claimed in claim 6, which is characterized in that the module of target detection includes:
Image processing unit, for the definition of the input layer according to the angle detection model, to described image to be detected
It is pre-processed, obtains pretreated original image；
Feature extraction unit, for the definition of the convolutional network layer according to the angle detection model, to the original graph
The tilt angle of the target text as in carries out angle character extraction, obtains the target text comprising angle character value
Angle character matrix；
Dimension map unit, for the definition of the fully connected network network layers according to the angle detection model, according to preset
Dimension map condition carries out dimension map to the angle character matrix, and it is corresponding four-dimensional special to obtain the angle character matrix
Levy vector, wherein the position of each component in four dimensional feature vector corresponds to a kind of predetermined angle classification；
Score calculation unit calculates four dimensional feature for the definition for layer of classifying according to the angle detection model
The score of each of vector component；
As a result output unit is made for choosing the angle value of the corresponding predetermined angle classification of the maximum score of numerical value
For the testing result of the target text.
8. character image correction processing device as claimed in claim 7, which is characterized in that described image processing unit includes:
Image scaling subelement, for carrying out image scaling to described image to be detected, obtaining according to preset image scaling mode
To base image；
Image cropping subelement, for carrying out the base image image cropping according to preset cutting method, obtaining standard
Image；
Image generates subelement, for carrying out mean value and normalized to the standard picture, and will treated standard
Original image of the image as described image to be detected.
9. a kind of computer equipment, including memory, processor and storage are in the memory and can be in the processor
The computer program of upper operation, which is characterized in that the processor realized when executing the computer program as claim 1 to
The step of any one of 5 character image correction processing method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization character image correction process side as described in any one of claim 1 to 5 when the computer program is executed by processor
The step of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416986.5A CN109583445A (en) | 2018-11-26 | 2018-11-26 | Character image correction processing method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416986.5A CN109583445A (en) | 2018-11-26 | 2018-11-26 | Character image correction processing method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109583445A true CN109583445A (en) | 2019-04-05 |
Family
ID=65924545
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811416986.5A Pending CN109583445A (en) | 2018-11-26 | 2018-11-26 | Character image correction processing method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583445A (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188747A (en) * | 2019-04-28 | 2019-08-30 | 广州华多网络科技有限公司 | A kind of sloped correcting method of text image, device and image processing equipment |
| CN110211092A (en) * | 2019-04-25 | 2019-09-06 | 上海上湖信息技术有限公司 | Imaging direction model training method and device, detection method and device, medium |
| CN110213605A (en) * | 2019-05-28 | 2019-09-06 | 广州华多网络科技有限公司 | Image rectification method, device and equipment |
| CN110223263A (en) * | 2019-04-23 | 2019-09-10 | 上海上湖信息技术有限公司 | Imaging direction detection method and device, readable storage medium storing program for executing |
| CN110378249A (en) * | 2019-06-27 | 2019-10-25 | 腾讯科技（深圳）有限公司 | The recognition methods of text image tilt angle, device and equipment |
| CN110456026A (en) * | 2019-08-13 | 2019-11-15 | 北京农业信息技术研究中心 | A kind of soil moisture content monitoring method and device |
| CN110490232A (en) * | 2019-07-18 | 2019-11-22 | 北京捷通华声科技股份有限公司 | Method, apparatus, the equipment, medium of training literal line direction prediction model |
| CN110490198A (en) * | 2019-08-12 | 2019-11-22 | 上海眼控科技股份有限公司 | Text orientation bearing calibration, device, computer equipment and storage medium |
| CN111062374A (en) * | 2019-12-10 | 2020-04-24 | 爱信诺征信有限公司 | Identification method, device, system, equipment and readable medium of identity card information |
| CN111079760A (en) * | 2019-08-02 | 2020-04-28 | 广东小天才科技有限公司 | Character recognition method and electronic equipment |
| CN111104942A (en) * | 2019-12-09 | 2020-05-05 | 熵智科技(深圳)有限公司 | Template matching network training method, template matching network recognition method and template matching network recognition device |
| CN111340040A (en) * | 2020-02-26 | 2020-06-26 | 五八有限公司 | Paper character recognition method and device, electronic equipment and storage medium |
| CN111435445A (en) * | 2019-12-24 | 2020-07-21 | 珠海大横琴科技发展有限公司 | Training method and device of character recognition model and character recognition method and device |
| CN111445386A (en) * | 2020-04-15 | 2020-07-24 | 深源恒际科技有限公司 | Image correction method based on four-point detection of text content |
| CN111507908A (en) * | 2020-03-11 | 2020-08-07 | 平安科技（深圳）有限公司 | Image correction processing method, device, storage medium and computer equipment |
| CN111583099A (en) * | 2020-04-14 | 2020-08-25 | 上海联影智能医疗科技有限公司 | Image rectification method, computer device, and storage medium |
| CN112199946A (en) * | 2020-09-15 | 2021-01-08 | 北京大米科技有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| CN112241737A (en) * | 2020-11-12 | 2021-01-19 | 瞬联软件科技（北京）有限公司 | Character image correction method and device |
| CN111814802B (en) * | 2020-08-31 | 2021-01-26 | 创新奇智(南京)科技有限公司 | Character image processing method and device, storage medium and electronic equipment |
| CN113128564A (en) * | 2021-03-23 | 2021-07-16 | 武汉泰沃滋信息技术有限公司 | Typical target detection method and system based on deep learning under complex background |
| CN113362427A (en) * | 2021-06-29 | 2021-09-07 | 深圳市创想三维科技股份有限公司 | Contour generation method and device for bevel edge three-dimensional character, computer equipment and medium |
| CN113436079A (en) * | 2021-06-23 | 2021-09-24 | 平安科技（深圳）有限公司 | Certificate image detection method and device, electronic equipment and storage medium |
| CN113554558A (en) * | 2020-04-26 | 2021-10-26 | 北京金山数字娱乐科技有限公司 | Image processing method and device |
| CN113785305A (en) * | 2019-05-05 | 2021-12-10 | 华为技术有限公司 | Method, device and equipment for detecting inclined characters |
| CN116151532A (en) * | 2022-09-29 | 2023-05-23 | 河北数微信息技术有限公司 | Self-service government service handling method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503687A (en) * | 2016-11-09 | 2017-03-15 | 合肥工业大学 | The monitor video system for identifying figures of fusion face multi-angle feature and its method |
| CN107180413A (en) * | 2017-05-05 | 2017-09-19 | 平安科技（深圳）有限公司 | Car damages picture angle correcting method, electronic installation and readable storage medium storing program for executing |
| CN108121986A (en) * | 2017-12-29 | 2018-06-05 | 深圳云天励飞技术有限公司 | Object detection method and device, computer installation and computer readable storage medium |
| CN108304821A (en) * | 2018-02-14 | 2018-07-20 | 广东欧珀移动通信有限公司 | Image-recognizing method and device, image acquiring method and equipment, computer equipment and non-volatile computer readable storage medium storing program for executing |
-
2018
- 2018-11-26 CN CN201811416986.5A patent/CN109583445A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503687A (en) * | 2016-11-09 | 2017-03-15 | 合肥工业大学 | The monitor video system for identifying figures of fusion face multi-angle feature and its method |
| CN107180413A (en) * | 2017-05-05 | 2017-09-19 | 平安科技（深圳）有限公司 | Car damages picture angle correcting method, electronic installation and readable storage medium storing program for executing |
| CN108121986A (en) * | 2017-12-29 | 2018-06-05 | 深圳云天励飞技术有限公司 | Object detection method and device, computer installation and computer readable storage medium |
| CN108304821A (en) * | 2018-02-14 | 2018-07-20 | 广东欧珀移动通信有限公司 | Image-recognizing method and device, image acquiring method and equipment, computer equipment and non-volatile computer readable storage medium storing program for executing |
Non-Patent Citations (1)
| Title |
|---|
| 沙周阳 等: "基于Mojette变换的倾斜文本图像方向校正", 计算机技术与发展, vol. 22, no. 07, pages 107 - 112 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110223263B (en) * | 2019-04-23 | 2021-07-13 | 上海上湖信息技术有限公司 | Imaging direction detection method and device and readable storage medium |
| CN110223263A (en) * | 2019-04-23 | 2019-09-10 | 上海上湖信息技术有限公司 | Imaging direction detection method and device, readable storage medium storing program for executing |
| CN110211092A (en) * | 2019-04-25 | 2019-09-06 | 上海上湖信息技术有限公司 | Imaging direction model training method and device, detection method and device, medium |
| CN110211092B (en) * | 2019-04-25 | 2021-08-27 | 上海上湖信息技术有限公司 | Imaging direction model training method and device, detection method and device, and medium |
| CN110188747A (en) * | 2019-04-28 | 2019-08-30 | 广州华多网络科技有限公司 | A kind of sloped correcting method of text image, device and image processing equipment |
| CN113785305B (en) * | 2019-05-05 | 2024-04-16 | 华为云计算技术有限公司 | Method, device and equipment for detecting inclined characters |
| CN113785305A (en) * | 2019-05-05 | 2021-12-10 | 华为技术有限公司 | Method, device and equipment for detecting inclined characters |
| CN110213605A (en) * | 2019-05-28 | 2019-09-06 | 广州华多网络科技有限公司 | Image rectification method, device and equipment |
| CN110213605B (en) * | 2019-05-28 | 2022-09-13 | 广州方硅信息技术有限公司 | Image correction method, device and equipment |
| CN110378249A (en) * | 2019-06-27 | 2019-10-25 | 腾讯科技（深圳）有限公司 | The recognition methods of text image tilt angle, device and equipment |
| CN110378249B (en) * | 2019-06-27 | 2024-01-12 | 腾讯科技（深圳）有限公司 | Text image inclination angle recognition method, device and equipment |
| CN110490232A (en) * | 2019-07-18 | 2019-11-22 | 北京捷通华声科技股份有限公司 | Method, apparatus, the equipment, medium of training literal line direction prediction model |
| CN111079760A (en) * | 2019-08-02 | 2020-04-28 | 广东小天才科技有限公司 | Character recognition method and electronic equipment |
| CN111079760B (en) * | 2019-08-02 | 2023-11-28 | 广东小天才科技有限公司 | Character recognition method and electronic equipment |
| CN110490198A (en) * | 2019-08-12 | 2019-11-22 | 上海眼控科技股份有限公司 | Text orientation bearing calibration, device, computer equipment and storage medium |
| CN110456026A (en) * | 2019-08-13 | 2019-11-15 | 北京农业信息技术研究中心 | A kind of soil moisture content monitoring method and device |
| CN111104942A (en) * | 2019-12-09 | 2020-05-05 | 熵智科技(深圳)有限公司 | Template matching network training method, template matching network recognition method and template matching network recognition device |
| CN111104942B (en) * | 2019-12-09 | 2023-11-03 | 熵智科技(深圳)有限公司 | Template matching network training method, recognition method and device |
| CN111062374A (en) * | 2019-12-10 | 2020-04-24 | 爱信诺征信有限公司 | Identification method, device, system, equipment and readable medium of identity card information |
| CN111435445A (en) * | 2019-12-24 | 2020-07-21 | 珠海大横琴科技发展有限公司 | Training method and device of character recognition model and character recognition method and device |
| CN111340040B (en) * | 2020-02-26 | 2023-09-12 | 五八有限公司 | Paper character recognition method and device, electronic equipment and storage medium |
| CN111340040A (en) * | 2020-02-26 | 2020-06-26 | 五八有限公司 | Paper character recognition method and device, electronic equipment and storage medium |
| WO2021179485A1 (en) * | 2020-03-11 | 2021-09-16 | 平安科技（深圳）有限公司 | Image rectification processing method and apparatus, storage medium, and computer device |
| CN111507908A (en) * | 2020-03-11 | 2020-08-07 | 平安科技（深圳）有限公司 | Image correction processing method, device, storage medium and computer equipment |
| CN111507908B (en) * | 2020-03-11 | 2023-10-20 | 平安科技（深圳）有限公司 | Image correction processing method, device, storage medium and computer equipment |
| CN111583099A (en) * | 2020-04-14 | 2020-08-25 | 上海联影智能医疗科技有限公司 | Image rectification method, computer device, and storage medium |
| CN111445386A (en) * | 2020-04-15 | 2020-07-24 | 深源恒际科技有限公司 | Image correction method based on four-point detection of text content |
| CN113554558A (en) * | 2020-04-26 | 2021-10-26 | 北京金山数字娱乐科技有限公司 | Image processing method and device |
| CN111814802B (en) * | 2020-08-31 | 2021-01-26 | 创新奇智(南京)科技有限公司 | Character image processing method and device, storage medium and electronic equipment |
| CN112199946B (en) * | 2020-09-15 | 2024-05-07 | 北京大米科技有限公司 | Data processing method, device, electronic equipment and readable storage medium |
| CN112199946A (en) * | 2020-09-15 | 2021-01-08 | 北京大米科技有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| CN112241737B (en) * | 2020-11-12 | 2024-01-26 | 瞬联软件科技（北京）有限公司 | Text image correction method and device |
| CN112241737A (en) * | 2020-11-12 | 2021-01-19 | 瞬联软件科技（北京）有限公司 | Character image correction method and device |
| CN113128564A (en) * | 2021-03-23 | 2021-07-16 | 武汉泰沃滋信息技术有限公司 | Typical target detection method and system based on deep learning under complex background |
| CN113436079A (en) * | 2021-06-23 | 2021-09-24 | 平安科技（深圳）有限公司 | Certificate image detection method and device, electronic equipment and storage medium |
| CN113362427B (en) * | 2021-06-29 | 2024-03-15 | 深圳市创想三维科技股份有限公司 | Contour generation method, device, computer equipment and medium for hypotenuse three-dimensional word |
| CN113362427A (en) * | 2021-06-29 | 2021-09-07 | 深圳市创想三维科技股份有限公司 | Contour generation method and device for bevel edge three-dimensional character, computer equipment and medium |
| CN116151532A (en) * | 2022-09-29 | 2023-05-23 | 河北数微信息技术有限公司 | Self-service government service handling method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109583445A (en) | Character image correction processing method, device, equipment and storage medium | |
| CN109389030A (en) | Facial feature points detection method, apparatus, computer equipment and storage medium | |
| CN106778928B (en) | Image processing method and device | |
| CN109492643A (en) | Certificate recognition methods, device, computer equipment and storage medium based on OCR | |
| WO2022000862A1 (en) | Method and apparatus for detecting object in fisheye image, and storage medium | |
| CN110490212A (en) | Molybdenum target image processing arrangement, method and apparatus | |
| CN110210276A (en) | A kind of motion track acquisition methods and its equipment, storage medium, terminal | |
| CN109271870A (en) | Pedestrian recognition methods, device, computer equipment and storage medium again | |
| CN109840524A (en) | Kind identification method, device, equipment and the storage medium of text | |
| CN110211035A (en) | Merge the image super-resolution method of the deep neural network of mutual information | |
| CN109711419A (en) | Image processing method, device, computer equipment and storage medium | |
| CN109711416A (en) | Target identification method, device, computer equipment and storage medium | |
| WO2012030384A1 (en) | Recognition of digital images | |
| CN111163265A (en) | Image processing method, image processing device, mobile terminal and computer storage medium | |
| CN110490232A (en) | Method, apparatus, the equipment, medium of training literal line direction prediction model | |
| CN108830175A (en) | Iris image local enhancement methods, device, equipment and storage medium | |
| CN110147787A (en) | Bank's card number automatic identifying method and system based on deep learning | |
| CN109063720A (en) | Handwritten word training sample acquisition methods, device, computer equipment and storage medium | |
| CN111027545A (en) | Card picture mark detection method and device, computer equipment and storage medium | |
| Tsigkas et al. | Markerless detection of ancient rock carvings in the wild: rock art in Vathy, Astypalaia | |
| US20240212393A1 (en) | Face anti-spoofing method, device, and computer-readable storage medium | |
| Cai et al. | An adaptive symmetry detection algorithm based on local features | |
| CN116597317A (en) | Remote sensing image change detection data generation method, device, equipment and medium | |
| Wang et al. | Spatial-invariant convolutional neural network for photographic composition prediction and automatic correction | |
| CN113012030A (en) | Image splicing method, device and equipment |
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 |