CN108846379A - Face list recognition methods, system, terminal device and storage medium - Google Patents
Face list recognition methods, system, terminal device and storage medium Download PDFInfo
- Publication number
- CN108846379A CN108846379A CN201810722796.XA CN201810722796A CN108846379A CN 108846379 A CN108846379 A CN 108846379A CN 201810722796 A CN201810722796 A CN 201810722796A CN 108846379 A CN108846379 A CN 108846379A
- Authority
- CN
- China
- Prior art keywords
- face
- default
- target
- character image
- anchor window
- 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
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The invention discloses a kind of face list recognition methods, system, terminal device and storage mediums to obtain free hand drawing piece in face to be identified by singly carrying out optical character identification to face to be identified；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；Current character image is obtained from the target signature according to target anchor window；The current character image is substituted into default Feature Selection Model, character features sequence is obtained, by the character features sequence inputting to Recognition with Recurrent Neural Network, obtains Text region result, it can be with the text information in precise positioning face list to be identified, invalid information is filtered out, the quality requirement of opposite free hand drawing piece is low, and strong antijamming capability, reduce the workload that face singly identifies, a large amount of face list recognition times are saved, improve face list recognition speed, the user experience is improved.
Description
Technical field
The present invention relates to optical identification fields more particularly to a kind of face list recognition methods, system, terminal device and storage to be situated between
Matter.
Background technique
In all trades and professions, there are a large amount of document, such as bank money, express delivery lists etc..Use pure manual input computer
The disadvantages of system needs a large amount of manpower and time cost, and it is low that there is also timeliness, and error rate is high；Relative to traditional list by hand
According to input method, the recognition speed and precision of optical character identification (Optical Character Recognition, OCR) are all
Have great advantage, a large amount of human cost can be saved, while the processing timeliness of business can be improved.
There is using troubles for existing OCR technique, and anti-interference ability is small, require document picture quality high, need pair
Full text information on document is identified, cannot be accomplished the disadvantages of precisely identifying.Commercial OCR technique currently on the market is big
The method for mostly using pattern match, usually entire OCR system can be divided into scanned picture, and picture pretreatment carries out mould to picture
Formula identification, the several steps of returned text information.Most of document OC technology needs special tool to carry out software deployment, such as hand
Document scanning machine is held, and a very big problem existing for the OCR of use pattern matching technique is the picture quality for requiring typing
Relatively high, line of text keeps horizontal or vertical as far as possible, cannot block to text, keeps text information that can clearly know, but existing
A big chunk document is unsatisfactory for condition in the real world, for example, express delivery singly has folding line in express transportation process, ink marks covers
The problems such as lid, whens many document picture typings, may have that picture is inclination, and existing OCR technique is to these problems
There is no good solution, the picture quality of typing is required very high；Existing OCR system generally be directed to full page picture into
Row Text region, calculation amount is very big, and is extracted many garbages.
Summary of the invention
The main purpose of the present invention is to provide a kind of face list recognition methods, system, terminal device and storage mediums, it is intended to
Solve that pattern match OCR technique using trouble, anti-interference ability in the prior art are small, it is high and can not to require document picture quality
Precisely the technical issues of identification document.
To achieve the above object, the present invention provides a kind of face list recognition methods, and the face list recognition methods includes following step
Suddenly：
Optical character identification is singly carried out to face to be identified, obtains free hand drawing piece in face to be identified；
Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；
Current character image is obtained from the target signature according to target anchor window；
The current character image is substituted into default Feature Selection Model, character features sequence, the default spy are obtained
Sign extracts the corresponding relationship of model reflection current character image and character features sequence；
By the character features sequence inputting to Recognition with Recurrent Neural Network, Text region result is obtained.
Preferably, described concentrated using non-maxima suppression algorithm from default candidate anchor window chooses target location window
Mouthful, it specifically includes：
Obtain each candidate anchor window that default candidate anchor window is concentrated matching degree with the target signature respectively；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses the corresponding candidate of matching degree peak
Anchor window is as target anchor window.
Preferably, described to obtain current character image from the target signature according to target anchor window, it is specific to wrap
It includes：
The corresponding multiple target positions of the target anchor window are found from the target signature；
Image interception is carried out to multiple target positions, using truncated picture as current character image.
Preferably, described by the character features sequence inputting to Recognition with Recurrent Neural Network, Text region is obtained as a result, specific
Including：
By the character features sequence inputting to Recognition with Recurrent Neural Network, so that the Recognition with Recurrent Neural Network is special to the text
It levies the corresponding convolution feature vector of sequence and carries out Classification and Identification, obtain the Text region result of Classification and Identification.
Preferably, described concentrated using non-maxima suppression algorithm from default candidate anchor window chooses target anchor window
Before, the face list recognition methods further includes：
The sample characteristics figure of preset quantity is chosen from default sample characteristics atlas, the sample characteristics figure is and the mesh
It marks characteristic pattern size and differs the characteristic pattern in pre-set dimension threshold range；
Regression analysis is carried out to each sample characteristics figure, obtains multiple positioning four angular coordinate values；
Default candidate window collection is determined according to each positioning four angular coordinate value.
Preferably, described to substitute into the current character image in default Feature Selection Model, obtain character features sequence
Before, the face list recognition methods further includes：
The sample character image for choosing preset quantity is concentrated from default character image, and it is corresponding to obtain this character image of various kinds
Sample character features sequence；
By various kinds, this character image is substituting to default training pattern, obtains training result, and the training pattern is based on depth
Spend the model that convolutional neural networks are established；
The matching result for obtaining this text of various kinds characteristic sequence and the training result, according to the matching result to described
Default training pattern optimizes；
Using the default training pattern after optimization as default Feature Selection Model.
Preferably, described this character image by various kinds is substituting to default training pattern, before obtaining training result, the face
Single recognition methods further includes：
It is chosen from preset model library with the training pattern of the current character images match as default training pattern.
In addition, to achieve the above object, the present invention also proposes that a kind of terminal device, the terminal device include：Memory,
Processor and the face list recognizer that is stored on the memory and can run on the processor, the face singly identify journey
Sequence is arranged for carrying out the step of face list recognition methods as described above.
In addition, to achieve the above object, the present invention also proposes a kind of storage medium, face list is stored on the storage medium
The step of recognizer, the face list recognizer realizes face list recognition methods as described above when being executed by processor.
In addition, to achieve the above object, the present invention also provides a kind of face list identifying system, the face list identifying system packet
It includes：
Picture obtains module, for singly carrying out optical character identification to face to be identified, obtains free hand drawing piece in face to be identified；
Characteristic extracting module is obtained for carrying out feature extraction to the free hand drawing piece in face to be identified by full convolutional network
Target signature；
Window chooses module, fixed for choosing target from default candidate anchor window concentration using non-maxima suppression algorithm
Position window；
Image collection module, for obtaining current character image from the target signature according to target anchor window；
Characteristic sequence obtains module, for substituting into the current character image in default Feature Selection Model, obtains text
Word characteristic sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；
As a result module is obtained, for obtaining Text region knot for the character features sequence inputting to Recognition with Recurrent Neural Network
Fruit.
List recognition methods in face proposed by the present invention is obtained to be identified by singly carrying out optical character identification to face to be identified
Face free hand drawing piece；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；Using non-
Maximum restrainable algorithms are concentrated from default candidate anchor window and choose target anchor window；According to target anchor window from the mesh
It marks and obtains current character image in characteristic pattern；The current character image is substituted into default Feature Selection Model, text is obtained
Characteristic sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；It will be described
Character features sequence inputting obtains Text region as a result, can be with the text in precise positioning face list to be identified to Recognition with Recurrent Neural Network
This information filters out invalid information, not only supports block letter, also supports handwriting recongnition, and the quality requirement of opposite free hand drawing piece is low,
And anti-inclination, rotation is reflective and a variety of interference such as block, and reduces the workload that face singly identifies, saves a large amount of faces and singly know
The other time improves face list recognition speed, and the user experience is improved.
Detailed description of the invention
Fig. 1 is the terminal device structural schematic diagram for the hardware running environment that the embodiment of the present invention is related to；
Fig. 2 is the flow diagram of list recognition methods first embodiment in face of the present invention；
Fig. 3 is the flow diagram of list recognition methods second embodiment in face of the present invention；
Fig. 4 is the flow diagram of list recognition methods 3rd embodiment in face of the present invention；
Fig. 5 is the functional block diagram of list identifying system first embodiment in face of the present invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The solution of the embodiment of the present invention is mainly：The present invention by singly carrying out optical character identification to face to be identified,
Obtain free hand drawing piece in face to be identified；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, it is special to obtain target
Sign figure；It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；It is positioned according to target
Window obtains current character image from the target signature；The current character image is substituted into default Feature Selection Model
In, character features sequence is obtained, the default Feature Selection Model reflection current character image is corresponding with character features sequence
Relationship；By the character features sequence inputting to Recognition with Recurrent Neural Network, Text region is obtained as a result, can be to be identified with precise positioning
Text information in the list of face filters out invalid information, not only supports block letter, also supports handwriting recongnition, opposite free hand drawing piece
Quality requirement is low, and anti-inclination, rotation, reflective and a variety of interference such as block, and reduces the workload that face singly identifies, saves
A large amount of faces list recognition time, improves face list recognition speed, the user experience is improved, solves pattern match in the prior art
OCR technique using trouble, anti-interference ability are small, require high to document picture quality and can not precisely identify that the technology of document is asked
Topic.
Referring to Fig.1, Fig. 1 is the terminal device structural schematic diagram for the hardware running environment that the embodiment of the present invention is related to.
As shown in Figure 1, the terminal device may include：Processor 1001, such as CPU, communication bus 1002, user's termination
Mouth 1003, network interface 1004, memory 1005.Wherein, communication bus 1002 is logical for realizing the connection between these components
Letter.User's end interface 1003 may include display screen (Display), input unit such as keyboard (Keyboard), optional user
End interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 optionally may include standard
Wireline interface, wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory, be also possible to stable deposit
Reservoir (non-volatile memory), such as magnetic disk storage.Memory 1005 optionally can also be independently of aforementioned place
Manage the storage device of device 1001.
It will be understood by those skilled in the art that terminal device structure shown in Fig. 1 is not constituted to the terminal device
It limits, may include perhaps combining certain components or different component layouts than illustrating more or fewer components.
As shown in Figure 1, as may include operating system, network communication mould in a kind of memory 1005 of storage medium
Block, user terminal interface module and face list recognizer.
Terminal device of the present invention calls the face list recognizer stored in memory 1005 by processor 1001, and executes
It operates below：
Optical character identification is singly carried out to face to be identified, obtains free hand drawing piece in face to be identified；
Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；
Current character image is obtained from the target signature according to target anchor window；
The current character image is substituted into default Feature Selection Model, character features sequence, the default spy are obtained
Sign extracts the corresponding relationship of model reflection current character image and character features sequence；
By the character features sequence inputting to Recognition with Recurrent Neural Network, Text region result is obtained.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
Obtain each candidate anchor window that default candidate anchor window is concentrated matching degree with the target signature respectively；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses the corresponding candidate of matching degree peak
Anchor window is as target anchor window.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
The corresponding multiple target positions of the target anchor window are found from the target signature；
Image interception is carried out to multiple target positions, using truncated picture as current character image.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
By the character features sequence inputting to Recognition with Recurrent Neural Network, so that the Recognition with Recurrent Neural Network is special to the text
It levies the corresponding convolution feature vector of sequence and carries out Classification and Identification, obtain the Text region result of Classification and Identification.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
The sample characteristics figure of preset quantity is chosen from default sample characteristics atlas, the sample characteristics figure is and the mesh
It marks characteristic pattern size and differs the characteristic pattern in pre-set dimension threshold range；
Regression analysis is carried out to each sample characteristics figure, obtains multiple positioning four angular coordinate values；
Default candidate window collection is determined according to each positioning four angular coordinate value.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
The sample character image for choosing preset quantity is concentrated from default character image, and it is corresponding to obtain this character image of various kinds
Sample character features sequence；
By various kinds, this character image is substituting to default training pattern, obtains training result, and the training pattern is based on depth
Spend the model that convolutional neural networks are established；
The matching result for obtaining this text of various kinds characteristic sequence and the training result, according to the matching result to described
Default training pattern optimizes；
Using the default training pattern after optimization as default Feature Selection Model.
Further, processor 1001 can call the face list recognizer stored in memory 1005, also execute following
Operation：
It is chosen from preset model library with the training pattern of the current character images match as default training pattern.
The present embodiment through the above scheme, by singly carrying out optical character identification to face to be identified, it is single to obtain face to be identified
Picture；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；Using it is non-greatly
It is worth restrainable algorithms and concentrates selection target anchor window from default candidate anchor window；It is special from the target according to target anchor window
It levies and obtains current character image in figure；The current character image is substituted into default Feature Selection Model, character features are obtained
Sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；By the text
Characteristic sequence is input to Recognition with Recurrent Neural Network, obtains Text region as a result, can be with the text envelope in precise positioning face list to be identified
Breath, filters out invalid information, not only supports block letter, also support handwriting recongnition, and the quality requirement of opposite free hand drawing piece is low, and
Anti- inclination, rotation is reflective and a variety of interference such as block, and reduces the workload that face singly identifies, when saving a large amount of faces and singly identifying
Between, face list recognition speed is improved, the user experience is improved.
Based on above-mentioned hardware configuration, list recognition methods embodiment in face of the present invention is proposed.
It is the flow diagram of list recognition methods first embodiment in face of the present invention referring to Fig. 2, Fig. 2.
In the first embodiment, the face list recognition methods includes the following steps：
Step S10, optical character identification is singly carried out to face to be identified, obtains free hand drawing piece in face to be identified.
It should be noted that the face to be identified is singly the face list identified, the face in the present embodiment singly refers to
All kinds of papery documents, generally can be bank money used in all trades and professions, can also be express delivery document, naturally it is also possible to be
Other kinds of document, the present embodiment are without restriction to this；Optical character identification (Optical Character
Recognition, OCR) refer to that electronic equipment (such as scanner or digital camera) checks the character printed on paper, pass through detection
Secretly, bright mode determines its shape, then shape is translated into the process of computword with character identifying method；That is, being directed to
Text conversion in paper document, is become the image file of black and white lattice using optical mode by printed character, and is passed through
Identification software by the text conversion in image at text formatting, the technology further edited and processed for word processor.It can be with
Understand, by singly carrying out optical character identification to face to be identified, preliminary ground free hand drawing piece in face to be identified can be obtained, to connect
Singly identification is prepared in the face got off.
Step S20, feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature
Figure.
It should be noted that the full convolutional network (Fully Convolutional Networks, FCN) is by network
Full articulamentum become convolutional layer after, whole network becomes the network of only convolutional layer and pond layer；Pass through full convolution net
Network carries out feature extraction to the free hand drawing piece in face to be identified, i.e., carries out feature extraction to the picture being originally inputted, and can obtain pair
The characteristic pattern answered, i.e., the described target signature.
Step S30, it is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window.
It is understood that the non-maxima suppression algorithm (Non Maximum Suppression, NMS) is a kind of
The efficient algorithm for obtaining local maximum, can eliminate extra frame, find the position of optimal object detection, using described non-
Maximum restrainable algorithms can be concentrated from default candidate anchor window and choose target anchor window, to improve the essence of Text region
Degree.
Step S40, current character image is obtained from the target signature according to target anchor window.
It should be understood that corresponding text figure can be obtained from the target signature according to target anchor window
Picture effectively reduces to ensure that the precision of character image identification and calculates consumption caused by incoherent image.
Further, the step S40 specifically includes following steps：
The corresponding multiple target positions of the target anchor window are found from the target signature；
Image interception is carried out to multiple target positions, using truncated picture as current character image.
It is understood that by finding the corresponding multiple targets of the target anchor window from the target signature
Position can select some and target location window that is, by corresponding target anchor window in the target signature
The corresponding position of mouth carries out image interception as the target position, then to multiple target positions, can obtain the figure of interception
Picture, the truncated picture are used as current character image.
Step S50, the current character image is substituted into default Feature Selection Model, obtains character features sequence, institute
State the corresponding relationship of default Feature Selection Model reflection current character image and character features sequence.
It should be noted that the default Feature Selection Model is pre-set for reflecting current character image and text
The model of the corresponding relationship of word characteristic sequence, the default Feature Selection Model are the mould established based on depth convolutional neural networks
Type can obtain character features sequence by the way that the current character image to be substituting in default Feature Selection Model.
Step S60, by the character features sequence inputting to Recognition with Recurrent Neural Network, Text region result is obtained.
It should be noted that the Recognition with Recurrent Neural Network RNN is the general name of two kinds of artificial neural networks.One is the times to pass
Return neural network (recurrent neural network), another kind is structure recurrent neural network (recursive
neural network).Digraph is connected and composed between the neuron of time recurrent neural network, and structure recurrent neural network is sharp
With the increasingly complex depth network of similar neural network structure recurrence Construction.The algorithm of the two training is different, but belongs to same
Algorithm variations；It is input to Recognition with Recurrent Neural Network by the way that the character features blow gently, can be obtained and the character features sequence
The corresponding Text region result of vector.
Further, the step S60 specifically includes following steps：
By the character features sequence inputting to Recognition with Recurrent Neural Network, so that the Recognition with Recurrent Neural Network is special to the text
It levies the corresponding convolution feature vector of sequence and carries out Classification and Identification, obtain the Text region result of Classification and Identification.
It is understood that the circulation can be passed through by after the character features sequence inputting to Recognition with Recurrent Neural Network
Neural network carries out Classification and Identification to the corresponding convolution feature vector of the character features sequence, in general, can be by same
One time point carried out operation, output category identification to the convolution feature vector according to the weight parameter of preset both direction
Text region afterwards is as a result, to improve recognition speed in the case where ensureing Text region precision.
The present embodiment through the above scheme, by singly carrying out optical character identification to face to be identified, it is single to obtain face to be identified
Picture；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；Using it is non-greatly
It is worth restrainable algorithms and concentrates selection target anchor window from default candidate anchor window；It is special from the target according to target anchor window
It levies and obtains current character image in figure；The current character image is substituted into default Feature Selection Model, character features are obtained
Sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；By the text
Characteristic sequence is input to Recognition with Recurrent Neural Network, obtains Text region as a result, can be with the text envelope in precise positioning face list to be identified
Breath, filters out invalid information, not only supports block letter, also support handwriting recongnition, and the quality requirement of opposite free hand drawing piece is low, and
Anti- inclination, rotation is reflective and a variety of interference such as block, and reduces the workload that face singly identifies, when saving a large amount of faces and singly identifying
Between, face list recognition speed is improved, and more intelligent, the user experience is improved.
Further, Fig. 3 is the flow diagram of list recognition methods second embodiment in face of the present invention, as shown in figure 3, being based on
First embodiment proposes list recognition methods second embodiment in face of the present invention, in the present embodiment, the step S30 specifically include with
Lower step：
Step S31, obtain each candidate anchor window that default candidate anchor window is concentrated respectively with the target signature
Matching degree.
It should be noted that each candidate anchor window that the default candidate anchor window is concentrated is special with the target respectively
The matching degree of sign figure refers to each candidate anchor window and the type of the target signature and the matching degree of size, generally
, the matching corresponding with having between different types of characteristic pattern or the characteristic pattern of different sizes of different anchor windows is closed
System, correspondingly, different anchor windows and the matching degree of the target signature refer to different anchor windows and different type or
The direct matching degree of the characteristic pattern of size, matching degree height then illustrate that the anchor window can accurately grab characteristic pattern
In character image, matching degree is low, illustrate the anchor window obtain character image in characteristic pattern effect it is poor, it is certainly each
The candidate anchor window matching degree with the target signature respectively, in addition to can be type or ruler with the target signature
Except the matching degree of very little size, the matching degree of other feature graph parameters can also be, the present embodiment is without restriction to this.
Step S32, it is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses matching degree peak pair
The candidate anchor window answered is as target anchor window.
It is understood that by the way that using non-maxima suppression algorithm selection can be concentrated from default candidate anchor window
With the corresponding candidate anchor window of degree peak, thus using candidate's anchor window as target anchor window, it can be further
Reduction calculate consumption, save the time of Text region, improve the accuracy and speed of Text region.
Further, before the step S30, the face list recognition methods is further comprising the steps of：
Step S301, the sample characteristics figure of preset quantity, the sample characteristics figure are chosen from default sample characteristics atlas
To differ the characteristic pattern in pre-set dimension threshold range with the target signature size.
It should be noted that the default sample characteristics atlas is pre-set for storing the collection of various characteristic patterns
It closes, is differed with the target signature size in pre-set dimension threshold value by that can be chosen from the default sample characteristics atlas
Characteristic pattern in range, using this feature figure as the sample characteristics figure.
Step S302, regression analysis is carried out to each sample characteristics figure, obtains multiple positioning four angular coordinate values.
It is understood that each sample characteristics figure is carried out regression analysis, it is corresponding fixed that each sample characteristics figure can be obtained
Position four angular coordinate value can carry out regression analysis by 5 various sizes of sample characteristics figures, be used in the concrete realization
In the positioning four angular coordinate value of localization of text information, to effectively reduce calculating consumption, the speed of Text region is improved, is saved
The time of Text region is saved.
Step S303, default candidate window collection is determined according to each positioning four angular coordinate value.
It should be understood that after each sample characteristics figure of acquisition corresponding positioning four angular coordinate value, it can be according to each positioning
Four angular coordinate value determines corresponding default candidate window, and multiple default candidate windows are gathered, and can determine default candidate
Window collection.
The present embodiment through the above scheme, passes through the sample characteristics of the selection preset quantity from default sample characteristics atlas
Figure, the sample characteristics figure is the characteristic pattern differed in pre-set dimension threshold range with the target signature size；To each
Sample characteristics figure carries out regression analysis, obtains multiple positioning four angular coordinate values；Default wait is determined according to each positioning four angular coordinate value
Select window collection；Obtain each candidate anchor window matching with the target signature respectively that default candidate anchor window is concentrated
Degree；It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses the corresponding candidate location window of matching degree peak
Mouth is used as target anchor window, and the workload that the face that further reduces singly identifies saves a large amount of face list recognition times, improves
Face list recognition speed, the user experience is improved.
Further, Fig. 4 is the flow diagram of list recognition methods 3rd embodiment in face of the present invention, as shown in figure 4, being based on
Second embodiment proposes list recognition methods 3rd embodiment in face of the present invention, in the present embodiment, described before the step S50
Face list recognition methods is further comprising the steps of：
Step S501, the sample character image for choosing preset quantity is concentrated from default character image, and obtains various kinds herein
The corresponding sample character features sequence of word image.
It should be noted that the default character image collection is the collection of the pre-set character image for training pattern
It closes, may include a large amount of various types of other character image, the sample for choosing preset quantity is concentrated from the default character image
Character image, the mode for obtaining sample character image can be special specified, be also possible to randomly select, naturally it is also possible to be it
His mode, the present embodiment are without restriction to this；Correspondingly, the default character image collection further includes and this character image of various kinds
Corresponding sample character features sequence.
Step S502, by various kinds, this character image is substituting to default training pattern, obtains training result, the training pattern
For the model established based on depth convolutional neural networks.
It is understood that the default training pattern is the model established based on depth convolutional neural networks, for anti-
The relationship between character image and character features sequence is reflected, it can be with by the way that various kinds this character image is substituting to default training pattern
Obtain training result corresponding with this character image of various kinds.
Further, further comprising the steps of before the step S502：
It is chosen from preset model library with the training pattern of the current character images match as default training pattern.
It should be noted that the preset model library is database for storing all kinds of training patterns, by choose with
The accuracy of Text region can be greatly improved as default training pattern in the training pattern of current character images match, reduce
Calculation amount, accelerates the speed of Text region.
Step S503, the matching result for obtaining this text of various kinds characteristic sequence and the training result, according to the matching
As a result the default training pattern is optimized.
It should be understood that this text of various kinds characteristic sequence is matched with the training result, various kinds can be obtained
The matching result of this text characteristic sequence and the training result, i.e. this text of various kinds characteristic sequence whether with training result one
It causes, is found if inconsistent at its difference, generate corresponding matching result, it can be to default trained mould according to the matching result
Type optimizes, i.e., preset weight is adjusted and is adjusted other operational parameters, until the matching result is matching
Unanimously.
Step S504, using the default training pattern after optimization as default Feature Selection Model.
It is understood that can more accurately to obtain the character image corresponding for the default training pattern after optimization
Character features sequence, using the default training pattern after optimization as default Feature Selection Model can be improved face list recognition speed and
Efficiency, and ensure that the accuracy of Text region.
The present embodiment through the above scheme, the sample text figure of preset quantity is chosen by concentrating from default character image
Picture, and obtain the corresponding sample character features sequence of various kinds this character image；From preset model library choose with it is described ought be above
The training pattern of word images match is as default training pattern, and by various kinds, this character image is substituting to default training pattern, obtains
Training result, the training pattern are the model established based on depth convolutional neural networks；Obtain this text of various kinds characteristic sequence
With the matching result of the training result, the default training pattern is optimized according to the matching result；After optimizing
Default training pattern as default Feature Selection Model, the workload that the face that further reduces singly identifies saves a large amount of faces
Single recognition time improves face list recognition speed and efficiency, and ensure that the accuracy of Text region.
The present invention further provides a kind of face list identifying systems.
It is the functional block diagram of list identifying system first embodiment in face of the present invention referring to Fig. 5, Fig. 5.
In list identifying system first embodiment in face of the present invention, which includes：
Picture obtains module 10, for singly carrying out optical character identification to face to be identified, obtains free hand drawing piece in face to be identified.
It should be noted that the face to be identified is singly the face list identified, the face in the present embodiment singly refers to
All kinds of papery documents, generally can be bank money used in all trades and professions, can also be express delivery document, naturally it is also possible to be
Other kinds of document, the present embodiment are without restriction to this；Optical character identification (Optical Character
Recognition, OCR) refer to that electronic equipment (such as scanner or digital camera) checks the character printed on paper, pass through detection
Secretly, bright mode determines its shape, then shape is translated into the process of computword with character identifying method；That is, being directed to
Text conversion in paper document, is become the image file of black and white lattice using optical mode by printed character, and is passed through
Identification software by the text conversion in image at text formatting, the technology further edited and processed for word processor.
It is understood that it is to be identified that preliminary ground can be obtained by singly carrying out optical character identification to face to be identified
Face free hand drawing piece, for next face, singly identification is prepared.
Characteristic extracting module 20 is obtained for carrying out feature extraction to the free hand drawing piece in face to be identified by full convolutional network
Obtain target signature.
It should be noted that the full convolutional network (Fully Convolutional Networks, FCN) is by network
Full articulamentum become convolutional layer after, whole network becomes the network of only convolutional layer and pond layer；Pass through full convolution net
Network carries out feature extraction to the free hand drawing piece in face to be identified, i.e., carries out feature extraction to the picture being originally inputted, and can obtain pair
The characteristic pattern answered, i.e., the described target signature.
Window chooses module 30, chooses target for concentrating using non-maxima suppression algorithm from default candidate anchor window
Anchor window.
It is understood that the non-maxima suppression algorithm (Non Maximum Suppression, NMS) is a kind of
The efficient algorithm for obtaining local maximum, can eliminate extra frame, find the position of optimal object detection, using described non-
Maximum restrainable algorithms can be concentrated from default candidate anchor window and choose target anchor window, to improve the essence of Text region
Degree.
Image collection module 40, for obtaining current character figure from the target signature according to target anchor window
Picture.
It should be understood that corresponding text figure can be obtained from the target signature according to target anchor window
Picture effectively reduces to ensure that the precision of character image identification and calculates consumption caused by incoherent image.
Characteristic sequence obtains module 50, for substituting into the current character image in default Feature Selection Model, obtains
Character features sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence.
It should be noted that the default Feature Selection Model is pre-set for reflecting current character image and text
The model of the corresponding relationship of word characteristic sequence, the default Feature Selection Model are the mould established based on depth convolutional neural networks
Type can obtain character features sequence by the way that the current character image to be substituting in default Feature Selection Model.
As a result module 60 is obtained, for obtaining Text region for the character features sequence inputting to Recognition with Recurrent Neural Network
As a result.
It should be noted that the Recognition with Recurrent Neural Network RNN is the general name of two kinds of artificial neural networks.One is the times to pass
Return neural network (recurrent neural network), another kind is structure recurrent neural network (recursive
neural network).Digraph is connected and composed between the neuron of time recurrent neural network, and structure recurrent neural network is sharp
With the increasingly complex depth network of similar neural network structure recurrence Construction.The algorithm of the two training is different, but belongs to same
Algorithm variations；It is input to Recognition with Recurrent Neural Network by the way that the character features blow gently, can be obtained and the character features sequence
The corresponding Text region result of vector.
In the concrete realization, the embodiment based on above-mentioned face list recognition methods, can be overall the face list identifying system
Structure divides into scheduler module, training module, identification module, picture database and model library, and it is whole that scheduler module is responsible for control
It is flowed in system operation and data flow in systems all components, and the trigger event when corresponding actions occur.Training module is negative
The model training of deep learning is blamed, model optimization needs to call the picture in picture database as training data, and training
Model after optimization is stored in model library.Identification module provides identification service, can provide the identifying call service based on API, fits
With optimal models in model library, identification service in real time is provided, optionally the new image data encountered in identification service is added
To picture database.Picture database is responsible for storing picture file and tab file, includes two class data：The data of manual markings
The data returned with identification module.Model library is responsible for storing the model that training module generates, and provides model clothes for identification module
Business, and in actual operation, the database of MongoDB distributed document storage can be used, mass data storage is provided
Service, for storing picture and corresponding mark information；Using Tensorflow even depth learning database, the instruction that building face singly identifies
Practice module, acquisition meets the requirements model on identification accuracy and recognition efficiency；In model library long-lasting useization technology and point
Cloth dispatching technique, can quick calling model file for training module and identification module provide service.Pass through abstract API, packaging
Identification service is the service of a convenient calling, can be with the external offer service of very comfortable.Convenient and fast control method is provided, it can
With by web terminal controlled training, identification is serviced, and model calls the monitoring for selecting and providing to whole system.Detailed function mould
It is automatic that block designs the data set management service for including, the function of model training, the identification service function for providing encapsulation, model file
Scheduling feature and system management function；Wherein, data set management service mainly completes the various necessary datas of system normal operation
Maintenance and management function.It is main to service the data management returned, in conjunction with tagging system including the management of flag data collection, identification
Data storage service etc..Model training service be mainly using distributed deep learning system and the data set enriched constantly into
The model training of row automation, Optimized model performance provide better support for identification service.Identify that service module function is main
Including the identifying system based on deep learning, packaged application programming interface (Application Programming
Interface, API) service, the access process service of high concurrent.Model library service is used in the chain of command list identifying system
Deep learning model, main includes the function of model persistent storage and support model distributed scheduling.System administration services
It is to provide the monitoring to whole system resource (including memory, graphics processor for managing entire surface list identifying system
(Graphics Processing Unit, GPU)), while model training can be carried out, identify the management of service, also offer pair
The management of data set and model file.
The present embodiment through the above scheme, obtains module by picture and singly carries out optical character identification to face to be identified, obtain
Obtain free hand drawing piece in face to be identified；Characteristic extracting module carries out feature extraction to the free hand drawing piece in face to be identified by full convolutional network,
Obtain target signature；Window chooses module and concentrates selection target from default candidate anchor window using non-maxima suppression algorithm
Anchor window；Image collection module obtains current character image according to target anchor window from the target signature；Feature
Retrieval module substitutes into the current character image in default Feature Selection Model, obtains character features sequence, described pre-
If the corresponding relationship of Feature Selection Model reflection current character image and character features sequence；As a result module is obtained by the text
Characteristic sequence is input to Recognition with Recurrent Neural Network, obtains Text region as a result, can be with the text envelope in precise positioning face list to be identified
Breath, filters out invalid information, not only supports block letter, also support handwriting recongnition, and the quality requirement of opposite free hand drawing piece is low, and
Anti- inclination, rotation is reflective and a variety of interference such as block, and reduces the workload that face singly identifies, when saving a large amount of faces and singly identifying
Between, face list recognition speed is improved, and more intelligent, the user experience is improved.
In addition, the embodiment of the present invention also proposes a kind of storage medium, face list recognizer is stored on the storage medium,
Following operation is realized when the face list recognizer is executed by processor：
Optical character identification is singly carried out to face to be identified, obtains free hand drawing piece in face to be identified；
Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；
Current character image is obtained from the target signature according to target anchor window；
The current character image is substituted into default Feature Selection Model, character features sequence, the default spy are obtained
Sign extracts the corresponding relationship of model reflection current character image and character features sequence；
By the character features sequence inputting to Recognition with Recurrent Neural Network, Text region result is obtained.
Further, following operation is also realized when the face list recognizer is executed by processor：
Obtain each candidate anchor window that default candidate anchor window is concentrated matching degree with the target signature respectively；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses the corresponding candidate of matching degree peak
Anchor window is as target anchor window.
Further, following operation is also realized when the face list recognizer is executed by processor：
The corresponding multiple target positions of the target anchor window are found from the target signature；
Image interception is carried out to multiple target positions, using truncated picture as current character image.
Further, following operation is also realized when the face list recognizer is executed by processor：
By the character features sequence inputting to Recognition with Recurrent Neural Network, so that the Recognition with Recurrent Neural Network is special to the text
It levies the corresponding convolution feature vector of sequence and carries out Classification and Identification, obtain the Text region result of Classification and Identification.
Further, following operation is also realized when the face list recognizer is executed by processor：
The sample characteristics figure of preset quantity is chosen from default sample characteristics atlas, the sample characteristics figure is and the mesh
It marks characteristic pattern size and differs the characteristic pattern in pre-set dimension threshold range；
Regression analysis is carried out to each sample characteristics figure, obtains multiple positioning four angular coordinate values；
Default candidate window collection is determined according to each positioning four angular coordinate value.
Further, following operation is also realized when the face list recognizer is executed by processor：
The sample character image for choosing preset quantity is concentrated from default character image, and it is corresponding to obtain this character image of various kinds
Sample character features sequence；
By various kinds, this character image is substituting to default training pattern, obtains training result, and the training pattern is based on depth
Spend the model that convolutional neural networks are established；
The matching result for obtaining this text of various kinds characteristic sequence and the training result, according to the matching result to described
Default training pattern optimizes；
Using the default training pattern after optimization as default Feature Selection Model.
Further, following operation is also realized when the face list recognizer is executed by processor：
It is chosen from preset model library with the training pattern of the current character images match as default training pattern.
The present embodiment through the above scheme, by singly carrying out optical character identification to face to be identified, it is single to obtain face to be identified
Picture；Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；Using it is non-greatly
It is worth restrainable algorithms and concentrates selection target anchor window from default candidate anchor window；It is special from the target according to target anchor window
It levies and obtains current character image in figure；The current character image is substituted into default Feature Selection Model, character features are obtained
Sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；By the text
Characteristic sequence is input to Recognition with Recurrent Neural Network, obtains Text region as a result, can be with the text envelope in precise positioning face list to be identified
Breath, filters out invalid information, not only supports block letter, also support handwriting recongnition, and the quality requirement of opposite free hand drawing piece is low, and
Anti- inclination, rotation is reflective and a variety of interference such as block, and reduces the workload that face singly identifies, when saving a large amount of faces and singly identifying
Between, face list recognition speed is improved, and more intelligent, the user experience is improved.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the system that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or system institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or system.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of face list recognition methods, which is characterized in that the face list recognition methods includes：
Optical character identification is singly carried out to face to be identified, obtains free hand drawing piece in face to be identified；
Feature extraction is carried out to the free hand drawing piece in face to be identified by full convolutional network, obtains target signature；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses target anchor window；
Current character image is obtained from the target signature according to target anchor window；
The current character image is substituted into default Feature Selection Model, character features sequence is obtained, the default feature mentions
The corresponding relationship of modulus type reflection current character image and character features sequence；
By the character features sequence inputting to Recognition with Recurrent Neural Network, Text region result is obtained.
2. list recognition methods in face as described in claim 1, which is characterized in that it is described using non-maxima suppression algorithm from default
Candidate anchor window, which is concentrated, chooses target anchor window, specifically includes：
Obtain each candidate anchor window that default candidate anchor window is concentrated matching degree with the target signature respectively；
It is concentrated using non-maxima suppression algorithm from default candidate anchor window and chooses the corresponding candidate positioning of matching degree peak
Window is as target anchor window.
3. list recognition methods in face as claimed in claim 2, which is characterized in that it is described according to target anchor window from the target
Current character image is obtained in characteristic pattern, is specifically included：
The corresponding multiple target positions of the target anchor window are found from the target signature；
Image interception is carried out to multiple target positions, using truncated picture as current character image.
4. list recognition methods in face as claimed in claim 3, which is characterized in that it is described by the character features sequence inputting to following
Ring neural network obtains Text region as a result, specifically including：
By the character features sequence inputting to Recognition with Recurrent Neural Network, so that the Recognition with Recurrent Neural Network is to the character features sequence
It arranges corresponding convolution feature vector and carries out Classification and Identification, obtain the Text region result of Classification and Identification.
5. such as list recognition methods in face of any of claims 1-4, which is characterized in that described to utilize non-maxima suppression
Algorithm is concentrated before choosing target anchor window from default candidate anchor window, and the face list recognition methods further includes：
The sample characteristics figure of preset quantity is chosen from default sample characteristics atlas, the sample characteristics figure is special with the target
It levies figure size and differs the characteristic pattern in pre-set dimension threshold range；
Regression analysis is carried out to each sample characteristics figure, obtains multiple positioning four angular coordinate values；
Default candidate window collection is determined according to each positioning four angular coordinate value.
6. such as list recognition methods in face of any of claims 1-4, which is characterized in that described by the current character figure
As substituting into default Feature Selection Model, before obtaining character features sequence, the face list recognition methods further includes：
The sample character image for choosing preset quantity is concentrated from default character image, and obtains the corresponding sample of various kinds this character image
This text characteristic sequence；
By various kinds, this character image is substituting to default training pattern, obtains training result, and the training pattern is to be rolled up based on depth
The model of product neural network；
The matching result for obtaining this text of various kinds characteristic sequence and the training result, according to the matching result to described default
Training pattern optimizes；
Using the default training pattern after optimization as default Feature Selection Model.
7. list recognition methods in face as claimed in claim 6, which is characterized in that it is described various kinds this character image is substituting to it is default
Training pattern, before obtaining training result, the face list recognition methods further includes：
It is chosen from preset model library with the training pattern of the current character images match as default training pattern.
8. a kind of face list identifying system, which is characterized in that the face list identifying system includes：
Picture obtains module, for singly carrying out optical character identification to face to be identified, obtains free hand drawing piece in face to be identified；
Characteristic extracting module obtains target for carrying out feature extraction to the free hand drawing piece in face to be identified by full convolutional network
Characteristic pattern；
Window chooses module, chooses target location window for concentrating using non-maxima suppression algorithm from default candidate anchor window
Mouthful；
Image collection module, for obtaining current character image from the target signature according to target anchor window；
Characteristic sequence obtains module, and for substituting into the current character image in default Feature Selection Model, it is special to obtain text
Levy sequence, the corresponding relationship of default Feature Selection Model reflection the current character image and character features sequence；
As a result module is obtained, for obtaining Text region result for the character features sequence inputting to Recognition with Recurrent Neural Network.
9. a kind of terminal device, which is characterized in that the terminal device includes：Memory, processor and it is stored in the storage
On device and the face list recognizer that can run on the processor, the face list recognizer are arranged for carrying out such as claim
Described in any one of 1 to 7 the step of the recognition methods of face list.
10. a kind of storage medium, which is characterized in that be stored with face list recognizer on the storage medium, the face singly identifies
The step of face list recognition methods as described in any one of claims 1 to 7 is realized when program is executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810722796.XA CN108846379A (en) | 2018-07-03 | 2018-07-03 | Face list recognition methods, system, terminal device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810722796.XA CN108846379A (en) | 2018-07-03 | 2018-07-03 | Face list recognition methods, system, terminal device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108846379A true CN108846379A (en) | 2018-11-20 |
Family
ID=64201067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810722796.XA Pending CN108846379A (en) | 2018-07-03 | 2018-07-03 | Face list recognition methods, system, terminal device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108846379A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840524A (en) * | 2019-01-04 | 2019-06-04 | 平安科技（深圳）有限公司 | Kind identification method, device, equipment and the storage medium of text |
| CN110309796A (en) * | 2019-07-05 | 2019-10-08 | 上海中通吉网络技术有限公司 | Face list recognition methods, device and equipment |
| CN110334818A (en) * | 2019-05-22 | 2019-10-15 | 广州中船文冲船坞有限公司 | A kind of method and system of pipeline automatic identification |
| CN110675121A (en) * | 2019-09-23 | 2020-01-10 | 珠海市新德汇信息技术有限公司 | Method for collecting picture type file material |
| CN110705382A (en) * | 2019-09-12 | 2020-01-17 | 武汉儒松科技有限公司 | Electronic invoice management method, device, equipment and medium based on invoice categories |
| CN110738203A (en) * | 2019-09-06 | 2020-01-31 | 中国平安财产保险股份有限公司 | Method and device for outputting field structuralization and computer readable storage medium |
| CN111079709A (en) * | 2019-12-31 | 2020-04-28 | 广州市昊链信息科技股份有限公司 | Electronic document generation method and device, computer equipment and storage medium |
| CN111144484A (en) * | 2019-12-26 | 2020-05-12 | 深圳集智数字科技有限公司 | Image identification method and device |
| WO2020113561A1 (en) * | 2018-12-07 | 2020-06-11 | 华为技术有限公司 | Method for extracting structural data from image, apparatus and device |
| CN111275051A (en) * | 2020-02-28 | 2020-06-12 | 上海眼控科技股份有限公司 | Character recognition method, character recognition device, computer equipment and computer-readable storage medium |
| CN111325194A (en) * | 2018-12-13 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
| CN111767732A (en) * | 2020-06-09 | 2020-10-13 | 上海交通大学 | Document content understanding method and system based on graph attention model |
| CN111898603A (en) * | 2020-08-10 | 2020-11-06 | 上海瑞美锦鑫健康管理有限公司 | Physical examination order recognition method and system based on deep neural network |
| CN112541489A (en) * | 2019-09-23 | 2021-03-23 | 顺丰科技有限公司 | Image detection method and device, mobile terminal and storage medium |
| CN112818961A (en) * | 2021-03-26 | 2021-05-18 | 北京东方金朔信息技术有限公司 | Image feature identification method and device |
| CN112906687A (en) * | 2019-12-03 | 2021-06-04 | 上海智臻智能网络科技股份有限公司 | Picture character positioning method and device, computer equipment and storage medium |
| CN112990172A (en) * | 2019-12-02 | 2021-06-18 | 阿里巴巴集团控股有限公司 | Text recognition method, character recognition method and device |
| WO2022006829A1 (en) * | 2020-07-09 | 2022-01-13 | 国网电子商务有限公司 | Bill image recognition method and system, electronic device, and storage medium |
| CN114187586A (en) * | 2022-02-16 | 2022-03-15 | 中航信移动科技有限公司 | Travel itinerary identification method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407971A (en) * | 2016-09-14 | 2017-02-15 | 北京小米移动软件有限公司 | Text recognition method and device |
| CN106570497A (en) * | 2016-10-08 | 2017-04-19 | 中国科学院深圳先进技术研究院 | Text detection method and device for scene image |
| CN107133616A (en) * | 2017-04-02 | 2017-09-05 | 南京汇川图像视觉技术有限公司 | A kind of non-division character locating and recognition methods based on deep learning |
| CN107563387A (en) * | 2017-09-14 | 2018-01-09 | 成都掌中全景信息技术有限公司 | Frame method is selected in a kind of image object detection based on Recognition with Recurrent Neural Network |
| CN107798299A (en) * | 2017-10-09 | 2018-03-13 | 平安科技（深圳）有限公司 | Billing information recognition methods, electronic installation and readable storage medium storing program for executing |
| CN108229524A (en) * | 2017-05-25 | 2018-06-29 | 北京航空航天大学 | A kind of chimney and condensing tower detection method based on remote sensing images |
-
2018
- 2018-07-03 CN CN201810722796.XA patent/CN108846379A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407971A (en) * | 2016-09-14 | 2017-02-15 | 北京小米移动软件有限公司 | Text recognition method and device |
| CN106570497A (en) * | 2016-10-08 | 2017-04-19 | 中国科学院深圳先进技术研究院 | Text detection method and device for scene image |
| CN107133616A (en) * | 2017-04-02 | 2017-09-05 | 南京汇川图像视觉技术有限公司 | A kind of non-division character locating and recognition methods based on deep learning |
| CN108229524A (en) * | 2017-05-25 | 2018-06-29 | 北京航空航天大学 | A kind of chimney and condensing tower detection method based on remote sensing images |
| CN107563387A (en) * | 2017-09-14 | 2018-01-09 | 成都掌中全景信息技术有限公司 | Frame method is selected in a kind of image object detection based on Recognition with Recurrent Neural Network |
| CN107798299A (en) * | 2017-10-09 | 2018-03-13 | 平安科技（深圳）有限公司 | Billing information recognition methods, electronic installation and readable storage medium storing program for executing |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020113561A1 (en) * | 2018-12-07 | 2020-06-11 | 华为技术有限公司 | Method for extracting structural data from image, apparatus and device |
| CN111325194B (en) * | 2018-12-13 | 2023-12-29 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
| CN111325194A (en) * | 2018-12-13 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
| CN109840524B (en) * | 2019-01-04 | 2023-07-11 | 平安科技（深圳）有限公司 | Text type recognition method, device, equipment and storage medium |
| CN109840524A (en) * | 2019-01-04 | 2019-06-04 | 平安科技（深圳）有限公司 | Kind identification method, device, equipment and the storage medium of text |
| CN110334818A (en) * | 2019-05-22 | 2019-10-15 | 广州中船文冲船坞有限公司 | A kind of method and system of pipeline automatic identification |
| CN110334818B (en) * | 2019-05-22 | 2022-04-26 | 广州文冲船舶修造有限公司 | Method and system for automatically identifying pipeline |
| CN110309796A (en) * | 2019-07-05 | 2019-10-08 | 上海中通吉网络技术有限公司 | Face list recognition methods, device and equipment |
| CN110738203A (en) * | 2019-09-06 | 2020-01-31 | 中国平安财产保险股份有限公司 | Method and device for outputting field structuralization and computer readable storage medium |
| CN110738203B (en) * | 2019-09-06 | 2024-04-05 | 中国平安财产保险股份有限公司 | Field structured output method, device and computer readable storage medium |
| CN110705382A (en) * | 2019-09-12 | 2020-01-17 | 武汉儒松科技有限公司 | Electronic invoice management method, device, equipment and medium based on invoice categories |
| CN110675121A (en) * | 2019-09-23 | 2020-01-10 | 珠海市新德汇信息技术有限公司 | Method for collecting picture type file material |
| CN112541489A (en) * | 2019-09-23 | 2021-03-23 | 顺丰科技有限公司 | Image detection method and device, mobile terminal and storage medium |
| CN112990172B (en) * | 2019-12-02 | 2023-12-22 | 阿里巴巴集团控股有限公司 | Text recognition method, character recognition method and device |
| CN112990172A (en) * | 2019-12-02 | 2021-06-18 | 阿里巴巴集团控股有限公司 | Text recognition method, character recognition method and device |
| CN112906687A (en) * | 2019-12-03 | 2021-06-04 | 上海智臻智能网络科技股份有限公司 | Picture character positioning method and device, computer equipment and storage medium |
| CN111144484A (en) * | 2019-12-26 | 2020-05-12 | 深圳集智数字科技有限公司 | Image identification method and device |
| CN111079709A (en) * | 2019-12-31 | 2020-04-28 | 广州市昊链信息科技股份有限公司 | Electronic document generation method and device, computer equipment and storage medium |
| CN111275051A (en) * | 2020-02-28 | 2020-06-12 | 上海眼控科技股份有限公司 | Character recognition method, character recognition device, computer equipment and computer-readable storage medium |
| CN111767732B (en) * | 2020-06-09 | 2024-01-26 | 上海交通大学 | Document content understanding method and system based on graph attention model |
| CN111767732A (en) * | 2020-06-09 | 2020-10-13 | 上海交通大学 | Document content understanding method and system based on graph attention model |
| WO2022006829A1 (en) * | 2020-07-09 | 2022-01-13 | 国网电子商务有限公司 | Bill image recognition method and system, electronic device, and storage medium |
| CN111898603A (en) * | 2020-08-10 | 2020-11-06 | 上海瑞美锦鑫健康管理有限公司 | Physical examination order recognition method and system based on deep neural network |
| CN112818961A (en) * | 2021-03-26 | 2021-05-18 | 北京东方金朔信息技术有限公司 | Image feature identification method and device |
| CN114187586A (en) * | 2022-02-16 | 2022-03-15 | 中航信移动科技有限公司 | Travel itinerary identification method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108846379A (en) | Face list recognition methods, system, terminal device and storage medium | |
| CN109308476B (en) | Billing information processing method, system and computer readable storage medium | |
| CN108985214A (en) | The mask method and device of image data | |
| CN113313111B (en) | Text recognition method, device, equipment and medium | |
| CN106529380A (en) | Image identification method and device | |
| CN109934181A (en) | Text recognition method, device, equipment and computer-readable medium | |
| JPH10260993A (en) | Title, headline and photograph from document scan image | |
| CN109948549A (en) | OCR data creation method, device, computer equipment and storage medium | |
| CN110874618B (en) | OCR template learning method and device based on small sample, electronic equipment and medium | |
| CN109190630A (en) | Character identifying method | |
| CN110321788A (en) | Training data processing method, device, equipment and computer readable storage medium | |
| CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
| CN113762269A (en) | Chinese character OCR recognition method, system, medium and application based on neural network | |
| CN109635805A (en) | Image text location method and device, image text recognition methods and device | |
| CN109800750A (en) | A kind of character area mask method of the text based on morphological image for rule composing | |
| CN109710788A (en) | Image pattern mark and management method and equipment | |
| US20220406084A1 (en) | Text recognition for a neural network | |
| CN112766255A (en) | Optical character recognition method, device, equipment and storage medium | |
| CN110298353A (en) | A kind of character identifying method and system | |
| US20230215125A1 (en) | Data identification method and apparatus | |
| CN110378351A (en) | Seal discrimination method and device | |
| CN108460418A (en) | A kind of invoice sorting technique based on Text region and semantic analysis | |
| CN110276352A (en) | Index identification method, device, electronic equipment and computer readable storage medium | |
| WO2023035558A1 (en) | Anchor point cut-based image processing method and apparatus, device, and medium | |
| CN114565927A (en) | Table identification method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181120 |
|
| RJ01 | Rejection of invention patent application after publication |