CN106709494A - Coupled spatial learning-based scene character recognition method - Google Patents

Coupled spatial learning-based scene character recognition method Download PDF

Info

Publication number
CN106709494A
CN106709494A CN201710014236.4A CN201710014236A CN106709494A CN 106709494 A CN106709494 A CN 106709494A CN 201710014236 A CN201710014236 A CN 201710014236A CN 106709494 A CN106709494 A CN 106709494A
Authority
CN
China
Prior art keywords
dictionary
scene character
space
dist
character image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710014236.4A
Other languages
Chinese (zh)
Other versions
CN106709494B (en
Inventor
张重
王红
刘爽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhongfang Information Technology Tianjin Co ltd
Original Assignee
Tianjin Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin Normal University filed Critical Tianjin Normal University
Priority to CN201710014236.4A priority Critical patent/CN106709494B/en
Publication of CN106709494A publication Critical patent/CN106709494A/en
Application granted granted Critical
Publication of CN106709494B publication Critical patent/CN106709494B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
    • G06V20/63—Scene text, e.g. street names
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Character Discrimination (AREA)
  • Image Analysis (AREA)

Abstract

The embodiments of the invention disclose a coupled spatial learning-based scene character recognition method. The method includes the following steps that: inputted scene character images are preprocessed, so that trained scene character images are obtained; recognition feature extraction is performed on the trained scene character images, so that a spatial dictionary can be obtained; the spatial dictionary is utilized to perform spatial coding on the recognition features of corresponding images, so that corresponding spatial coding vectors can be obtained; maximization extraction is performed on the spatial coding vectors, so that feature vectors can be obtained; a linear support vector machine is utilized to perform training based on the feature vectors, so that a scene character recognition classification model is obtained; and the feature vectors of a test scene character image is obtained and is inputted into the scene character recognition classification model, so that a scene character recognition result can be obtained. According to the coupled spatial learning-based scene character recognition method of the invention, the spatial dictionary is crated and is utilized to perform spatial coding, and therefore, the textual information of a space can be effectively integrated into the feature vectors, so that spatial information can be effectively mined, and therefore, the correct rate of scene character recognition can be improved.

Description

A kind of scene character recognition method based on coupled room study
Technical field
The invention belongs to mode identification technology, and in particular to a kind of scene Text region based on coupled room study Method.
Background technology
Scene Text region plays important role in area of pattern recognition, it can be applied directly to image retrieval, The fields such as intelligent transportation, man-machine interaction.In actual applications, scene Text region is a research direction for very challenging property, Because scene word can be influenceed by extraneous factors such as uneven illumination, distortion, complicated backgrounds.
Recent decades scene Text region is widely studied, and the method for some early stages is carried out using OCR Scene Text region.But, OCR has significant limitation, such as scene character image binarization operation. In recent years, the method for a large amount of scene Text regions is suggested, and achieves larger progress.Wherein, most representative work It is the scene character recognition method based on target identification.Method based on target identification has skipped scene character image binaryzation Process and each scene word is regarded as a special target, and area of pattern recognition achieve it is certain into Work(.Such as：Newell et al. is entered using multiple dimensioned HOG (Histogram of Oriented Gradients, histogram of gradients) Row character representation.Zhang et al. extracts sparse coding histogram (histograms of sparse codes, HSC) feature and enters Row character representation.Shi et al. considers local feature information and global structure information.Although these methods achieve certain Effect, but they largely ignore spatial context information.Because different literals may be wrapped in different positions Characteristic information containing identical, this can cause reconstructed error, therefore in order to solve this problem, Gao et al. in the character representation stage Stroke storehouse is proposed to consider spatial context information.The method that Shi et al. is proposed is the extension of Gao et al. methods, and they use The multiple dimensioned stroke storehouse of identification represents feature.Tian et al. proposes to consider the symbiosis between HOG features to add sky Between contextual information.In addition, Gao et al. also proposes to be embedded in dictionary based on position to consider spatial context information.Although more than Method achieves larger success, but only from the aspect of single spatial context information, i.e. dictionary study stage or Coding stage, so effective spatial context information can not sufficiently be retained.
The content of the invention
Larger technology is influenceed to ask scene Text region result the invention aims to solve spatial context information Topic, therefore, the present invention provides a kind of scene character recognition method based on coupled room study.
In order to realize the purpose, scene character recognition method of the present invention based on coupled room study includes following step Suddenly：
Step S1, pretreatment operation is carried out to N width input scene character images respectively, obtains N width Training scene word graphs Picture；
Step S2, feature extraction is identified for N width Training scene character images respectively, obtains N number of space dictionary；
Step S3, space volume is carried out using the space dictionary of every width Training scene character image to the identification feature of the image Code, obtains corresponding space encoding vector；
Step S4, the space encoding vector for every width Training scene character image carries out maximization extraction, obtains described The corresponding characteristic vector of Training scene character image；
Step S5, based on the characteristic vector of the Training scene character image, is trained using linear SVM, Obtain scene Text region disaggregated model；
Step S6, the characteristic vector of test scene character image, input to the scene are obtained according to the step S1-S4 Text region disaggregated model obtains scene Text region result.
Alternatively, the step S1 is comprised the following steps：
Step S11, gray scale scene character image is converted to by the input scene character image；
Step S12, is H × W by the size normalization of the gray scale scene character image, the gray scale scene text after normalization Word image as the Training scene character image, wherein, H and W represents the height and width of gray scale scene character image respectively.
Alternatively, the step S2 is comprised the following steps：
Step S21, in the P of every width Training scene character imagei(i=1,2 ..., m) position punishment be you can well imagine and take an identification Feature, wherein, m is that the identification feature of every width Training scene character image extracts number of positions；
Step S22, for N width Training scene character images, to from PiAll identification features that extraction is obtained at position are entered Row cluster, obtains sub- dictionary Ci(i=1,2 ..., m), and by the sub- dictionary CiPosition be designated as Pi；
Step S23, the m sub- dictionary series connection that will carry positional information obtains space dictionary.
Alternatively, the identification feature is HOG features.
Alternatively, in the step S22, identification feature is clustered using k-means clustering algorithms.
Alternatively, the space dictionary is expressed as：
D={ C, P }={ (C1,P1),(C2,P2),...,(Cm,Pm),
Wherein, D representation spaces dictionary, C=(C1,C2,…,Cm) it is the m set of sub- dictionary, P=(P1,P2,…,Pm) table Show the positional information set of sub- dictionary set C.
Alternatively, in the step S3, the identification by the object function shown in following formula to Training scene character image is special Levy and be spatially encoded：
Wherein, | | | |2Represent l2Norm, ⊙ represents the point multiplication operation of corresponding element in two matrixes, fjRepresent that identification is special Levy, ajRepresent fjCorresponding space encoding vector, A=[a1,a2,…,aj...] the vectorial set of all space encodings is represented, | | fj-Caj||2Represent that utilization space dictionary is reconstructed the error of generation to identification feature；||djF⊙aj||2It is local regular terms, The distance between code word in expression feature space in identification feature and sub- dictionary restriction relation；||djE⊙aj||2For space just Then item, represents the position relationship between the code word in binding characteristic and sub- dictionary in theorem in Euclid space；α and β is regularization parameter,Representation space coding vector ajIn all elements sum be equal to 1；djFRepresent feature space in identification feature and The distance between code word in sub- dictionary, djERepresent identification feature f in theorem in Euclid spacejCode word pair in corresponding position and sub- dictionary The distance between position answered.
Alternatively, the distance between code word in feature space in identification feature and sub- dictionary djFIt is expressed as：
Wherein, σFIt it is one for adjusting djFThe parameter of weight decrease speed, dist (fj, C) it is defined as：
dist(fj, C) and=[dist (fj,C1),dist(fj,C2),...,dist(fj,Cm)]T
Wherein, dist (fj,Ci) (i=1,2 ..., m) represent feature fjWith sub- dictionary CiIn it is European between all code words Distance.
Alternatively, identification feature f in theorem in Euclid spacejIn corresponding position and sub- dictionary between the corresponding position of code word away from From djEIt is expressed as：
Wherein, σEIt it is one for adjusting djEThe parameter of weight decrease speed, dist (lj, P) it is defined as：
dist(lj, P) and=[dist (lj,P1),…,dist(lj,P1),dist(lj,P2),…,dist(lj,P2),…, dist(lj,Pm),…,dist(lj,Pm)]TWherein, dist (lj,Pi) (i=1,2 ..., m) represent identification feature fjPosition lj With sub- dictionary CiPosition PiBetween Euclidean distance.
Alternatively, in the step S4, entered for the space encoding vector of every width Training scene character image using following formula Row is maximized and extracted：
A=max { a1,a2,...,aj,...,am,
Wherein, a represents the characteristic vector of Training scene character image, aj(j=1,2 ..., m) representation space encode to Amount.
Beneficial effects of the present invention are：The present invention is by creating space dictionary and carrying out space using the space dictionary for creating , can be effectively combined spatial context information in characteristic vector by coding, reach the purpose of effective excavated space information, from And improve the accuracy of scene Text region.
It should be noted that the present invention obtained project of national nature science fund project No.61401309, No.61501327, Tianjin application foundation and cutting edge technology research plan youth fund project No.15JCQNJC01700, Tianjin Normal University doctor The subsidy of fund project No.5RL134, No.52XB1405.
Brief description of the drawings
Fig. 1 is the flow of the scene character recognition method learnt based on coupled room proposed according to one embodiment of the invention Figure.
Specific embodiment
To make the object, technical solutions and advantages of the present invention of greater clarity, with reference to specific embodiment and join According to accompanying drawing, the present invention is described in more detail.It should be understood that these descriptions are merely illustrative, and it is not intended to limit this hair Bright scope.Additionally, in the following description, the description to known features and technology is eliminated, to avoid unnecessarily obscuring this The concept of invention.
Fig. 1 is the flow of the scene character recognition method learnt based on coupled room proposed according to one embodiment of the invention Figure, illustrates more of the invention to implement flow below by taking Fig. 1 as an example.The method of the present invention is a kind of based on coupled room The scene character recognition method of study, its specific steps include：
Step S1, pretreatment operation is carried out to N width input scene character images respectively, obtains N width Training scene word graphs Picture；
Wherein, the pretreatment operation is comprised the following steps：
Step S11, gray scale scene character image is converted to by the input scene character image；
Step S12, is H × W by the size normalization of the gray scale scene character image, the gray scale scene text after normalization Word image as the Training scene character image, wherein, H and W represents the height and width of gray scale scene character image respectively.
Step S2, feature extraction is identified for N width Training scene character images respectively, obtains N number of space dictionary；
Further, the step S2 is comprised the following steps：
Step S21, in the P of every width Training scene character imagei(i=1,2 ..., m) position punishment be you can well imagine and take an identification Feature, wherein, m is the feature extraction number of positions of every width Training scene character image, so every width Training scene character image M identification feature can be obtained；
Wherein, the identification feature can be HOG features, or other identification feature.
Step S22, for N width Training scene character images, to from PiAll identification features that extraction is obtained at position are entered Row cluster, obtains sub- dictionary Ci(i=1,2 ..., m), and by the sub- dictionary CiPosition be designated as Pi, so, for m feature Extract position and obtain m sub- dictionary；
Wherein, cluster operation is carried out using clustering algorithms such as k-means.
Step S23, the m sub- dictionary series connection that will carry positional information obtains space dictionary.
Wherein, the space dictionary can be expressed as：
D={ C, P }={ (C1,P1),(C2,P2),...,(Cm,Pm),
Wherein, D representation spaces dictionary, C=(C1,C2,…,Cm) it is the m set of sub- dictionary, corresponding P=(P1, P2,…,Pm) represent the positional information set of sub- dictionary set C.
Step S3, sky is carried out using the space dictionary of every width Training scene character image to m identification feature of the image Between encode, obtain corresponding m space encoding vectorial；
In the step S3, by following object function using the space dictionary to every width Training scene character image M identification feature is spatially encoded：
Wherein, | | | |2Represent l2Norm, ⊙ represents the point multiplication operation of corresponding element in two matrixes, fjRepresent that identification is special Levy, ajRepresent fjCorresponding space encoding vector, corresponding A=[a1,a2,…,aj...] and represent the vectorial collection of all space encodings Close, | | fj-Caj||2Represent that utilization space dictionary is reconstructed the error of generation to identification feature；||djF⊙aj||2For part just Then item, represents the distance between code word in feature space in identification feature and sub- dictionary restriction relation；||djE⊙aj||2It is sky Between regular terms, represent the position relationship between code word in binding characteristic and sub- dictionary in theorem in Euclid space；α and β is regularization Parameter,Representation space coding vector ajIn all elements sum be equal to 1；djFRepresent identification in feature space The distance between code word in feature and sub- dictionary, embodies form as follows：
Wherein, σFIt it is one for adjusting djFThe parameter of weight decrease speed, dist (fj, C) it is defined as follows：
dist(fj, C) and=[dist (fj,C1),dist(fj,C2),...,dist(fj,Cm)]T
Wherein, dist (fj,Ci) (i=1,2 ..., m) represent feature fjWith sub- dictionary CiIn it is European between all code words Distance.
djERepresent identification feature f in theorem in Euclid spacejCorresponding position ljThe distance between with P, embody form as follows It is shown：
Wherein, σEIt it is one for adjusting djEThe parameter of weight decrease speed.dist(lj, P) it is defined as follows：dist(lj, P)=[dist (lj,P1),…,dist(lj,P1),dist(lj,P2),…,dist(lj,P2),…,dist(lj,Pm),…, dist(lj,Pm)]TWherein, dist (lj,Pi) (i=1,2 ..., m) represent identification feature fjPosition ljWith sub- dictionary CiPosition Pi Between Euclidean distance.
Above-mentioned object function is chosen one group of code word using local regular terms in feature space and weight is carried out to identification feature Structure, at the same in theorem in Euclid space utilization space regular terms come constrain the position between the code word in identification feature and sub- dictionary close System.
To above-mentioned object function derivation, an analytic solutions can be obtained, it is as follows：
Wherein, Aj=(CT-1fj T)(CT-1fj T)TCovariance matrix is represented, using formulaCan be to solving 'sIt is normalized operation.
The optimization process direct solution that can avoid complexity by above-mentioned analytic solutions goes out the corresponding space encoding of identification feature Vector.
Step S4, the space encoding vector for every width Training scene character image carries out maximization extraction, obtains described The corresponding characteristic vector of Training scene character image；
In the step S4, carried out most for the space encoding vector of every width Training scene character image using equation below Bigization is extracted：
A=max { a1,a2,...,aj,...,am,
Wherein, aj(j=1,2 ..., m) representation space coding vector, a represent the feature of Training scene character image to Amount.
By above-mentioned formula to m space encoding of width Training scene character image vector it is every one-dimensional take maximum come Obtain the characteristic vector a of the Training scene character image.
Step S5, based on the characteristic vector of the Training scene character image, is trained using linear SVM, Obtain scene Text region disaggregated model；
Step S6, the characteristic vector of test scene character image, input to the scene are obtained according to the step S1-S4 Text region disaggregated model obtains scene Text region result.
Using online disclosed scene text image data storehouse as test object, such as on ICDAR2003 databases, when H × W=64 × 32, when position m is 128, the accuracy of scene Text region is 83.2%, it can be seen that the inventive method has Effect property.
It should be appreciated that above-mentioned specific embodiment of the invention is used only for exemplary illustration or explains of the invention Principle, without being construed as limiting the invention.Therefore, that is done without departing from the spirit and scope of the present invention is any Modification, equivalent, improvement etc., should be included within the scope of the present invention.Additionally, appended claims purport of the present invention In the whole changes covered in the equivalents for falling into scope and border or this scope and border and repair Change example.

Claims (10)

1. it is a kind of based on coupled room study scene character recognition method, it is characterised in that the method is comprised the following steps：
Step S1, pretreatment operation is carried out to N width input scene character images respectively, obtains N width Training scene character images；
Step S2, feature extraction is identified for N width Training scene character images respectively, obtains N number of space dictionary；
Step S3, is spatially encoded using the space dictionary of every width Training scene character image to the identification feature of the image, Obtain corresponding space encoding vector；
Step S4, the space encoding vector for every width Training scene character image carries out maximization extraction, obtains the training The corresponding characteristic vector of scene character image；
Step S5, based on the characteristic vector of the Training scene character image, is trained using linear SVM, is obtained Scene Text region disaggregated model；
Step S6, the characteristic vector of test scene character image, input to the scene word are obtained according to the step S1-S4 Identification disaggregated model obtains scene Text region result.
2. method according to claim 1, it is characterised in that the step S1 is comprised the following steps：
Step S11, gray scale scene character image is converted to by the input scene character image；
Step S12, is H × W, the gray scale scene word graph after normalization by the size normalization of the gray scale scene character image As the Training scene character image, wherein, H and W represents the height and width of gray scale scene character image respectively.
3. method according to claim 1, it is characterised in that the step S2 is comprised the following steps：
Step S21, in the P of every width Training scene character imagei(i=1,2 ..., m) position punishment be you can well imagine and take an identification feature, Wherein, m is the identification feature extraction number of positions of every width Training scene character image；
Step S22, for N width Training scene character images, to from PiAll identification features that extraction is obtained at position are gathered Class, obtains sub- dictionary Ci(i=1,2 ..., m), and by the sub- dictionary CiPosition be designated as Pi；
Step S23, the m sub- dictionary series connection that will carry positional information obtains space dictionary.
4. method according to claim 3, it is characterised in that the identification feature is HOG features.
5. method according to claim 3, it is characterised in that in the step S22, using k-means clustering algorithms pair Identification feature is clustered.
6. method according to claim 3, it is characterised in that the space dictionary is expressed as：
D={ C, P }={ (C1,P1),(C2,P2),...,(Cm,Pm),
Wherein, D representation spaces dictionary, C=(C1,C2,…,Cm) it is the m set of sub- dictionary, P=(P1,P2,…,Pm) represent son The positional information set of dictionary set C.
7. method according to claim 1, it is characterised in that in the step S3, by the object function shown in following formula Identification feature to Training scene character image is spatially encoded：
Wherein, | | | |2Represent l2Norm, ⊙ represents the point multiplication operation of corresponding element in two matrixes, fjRepresent identification feature, aj Represent fjCorresponding space encoding vector, A=[a1,a2,…,aj...] the vectorial set of all space encodings is represented, | | fj-Caj ||2Represent that utilization space dictionary is reconstructed the error of generation to identification feature；||djF⊙aj||2It is local regular terms, represents special Levy the distance between code word in space in identification feature and sub- dictionary restriction relation；||djE⊙aj||2It is space regular terms, table Show the position relationship between the code word in theorem in Euclid space in binding characteristic and sub- dictionary；α and β is regularization parameter,Representation space coding vector ajIn all elements sum be equal to 1；djFRepresent feature space in identification feature and The distance between code word in sub- dictionary, djERepresent identification feature f in theorem in Euclid spacejCode word pair in corresponding position and sub- dictionary The distance between position answered.
8. method according to claim 7, it is characterised in that code word in feature space in identification feature and sub- dictionary it Between apart from djFIt is expressed as：
d j F = exp ( d i s t ( f j , C ) σ F ) ,
Wherein, σFIt it is one for adjusting djFThe parameter of weight decrease speed, dist (fj, C) it is defined as：
dist(fj, C) and=[dist (fj,C1),dist(fj,C2),...,dist(fj,Cm)]T
Wherein, dist (fj,Ci) (i=1,2 ..., m) represent feature fjWith sub- dictionary CiIn Euclidean distance between all code words.
9. method according to claim 7, it is characterised in that identification feature f in theorem in Euclid spacejCorresponding position and sub- word The distance between corresponding position of code word d in allusion quotationjEIt is expressed as：
d j E = exp ( d i s t ( l j , P ) σ E ) ,
Wherein, σEIt it is one for adjusting djEThe parameter of weight decrease speed, dist (lj, P) it is defined as：
dist(lj, P) and=[dist (lj,P1),…,dist(lj,P1),dist(lj,P2),…,dist(lj,P2),…,dist (lj,Pm),…,dist(lj,Pm)]T
Wherein, dist (lj,Pi) (i=1,2 ..., m) represent identification feature fjPosition ljWith sub- dictionary CiPosition PiBetween Europe Formula distance.
10. method according to claim 1, it is characterised in that in the step S4, using following formula for every width training place The space encoding vector of scape character image carries out maximization extraction：
A=max { a1,a2,...,aj,...,am,
Wherein, a represents the characteristic vector of Training scene character image, aj(j=1,2 ..., m) representation space coding vector.
CN201710014236.4A 2017-01-10 2017-01-10 Scene character recognition method based on coupling space learning Active CN106709494B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710014236.4A CN106709494B (en) 2017-01-10 2017-01-10 Scene character recognition method based on coupling space learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710014236.4A CN106709494B (en) 2017-01-10 2017-01-10 Scene character recognition method based on coupling space learning

Publications (2)

Publication Number Publication Date
CN106709494A true CN106709494A (en) 2017-05-24
CN106709494B CN106709494B (en) 2019-12-24

Family

ID=58908090

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710014236.4A Active CN106709494B (en) 2017-01-10 2017-01-10 Scene character recognition method based on coupling space learning

Country Status (1)

Country Link
CN (1) CN106709494B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107679505A (en) * 2017-10-13 2018-02-09 成都准星云学科技有限公司 A kind of method realized to handwritten character rejection
CN108764233A (en) * 2018-05-08 2018-11-06 天津师范大学 A kind of scene character recognition method based on continuous convolution activation

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104537362A (en) * 2015-01-16 2015-04-22 中国科学院自动化研究所 Domain-based self-adaptive English scene character recognition method
CN105760821A (en) * 2016-01-31 2016-07-13 中国石油大学(华东) Classification and aggregation sparse representation face identification method based on nuclear space

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104537362A (en) * 2015-01-16 2015-04-22 中国科学院自动化研究所 Domain-based self-adaptive English scene character recognition method
CN105760821A (en) * 2016-01-31 2016-07-13 中国石油大学(华东) Classification and aggregation sparse representation face identification method based on nuclear space

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107679505A (en) * 2017-10-13 2018-02-09 成都准星云学科技有限公司 A kind of method realized to handwritten character rejection
CN107679505B (en) * 2017-10-13 2020-04-21 林辉 Method for realizing rejection of handwritten character
CN108764233A (en) * 2018-05-08 2018-11-06 天津师范大学 A kind of scene character recognition method based on continuous convolution activation
CN108764233B (en) * 2018-05-08 2021-10-15 天津师范大学 Scene character recognition method based on continuous convolution activation

Also Published As

Publication number Publication date
CN106709494B (en) 2019-12-24

Similar Documents

Publication Publication Date Title
CN107256246B (en) printed fabric image retrieval method based on convolutional neural network
CN110532900B (en) Facial expression recognition method based on U-Net and LS-CNN
Yan et al. The fastest deformable part model for object detection
Hussain et al. Feature sets and dimensionality reduction for visual object detection
Guo et al. Pixel-wise classification method for high resolution remote sensing imagery using deep neural networks
Wang et al. Leaf recognition based on PCNN
Lin et al. Masked face detection via a modified LeNet
CN105095902B (en) Picture feature extracting method and device
CN110399821B (en) Customer satisfaction acquisition method based on facial expression recognition
CN106056082B (en) A kind of video actions recognition methods based on sparse low-rank coding
CN104951791B (en) data classification method and device
CN105139041A (en) Method and device for recognizing languages based on image
Shi et al. Fisher vector for scene character recognition: A comprehensive evaluation
Li et al. Lemon‐YOLO: An efficient object detection method for lemons in the natural environment
CN105117740A (en) Font identification method and device
CN110335206B (en) Intelligent filter method, device and computer readable storage medium
Qi et al. A multiscale deeply described correlatons-based model for land-use scene classification
CN112766229A (en) Human face point cloud image intelligent identification system and method based on attention mechanism
CN104794455A (en) Dongba hieroglyphic recognizing method
CN113435254A (en) Sentinel second image-based farmland deep learning extraction method
Liu et al. Discriminant sparse coding for image classification
CN116758609A (en) Lightweight face recognition method based on feature model improvement
CN106709494A (en) Coupled spatial learning-based scene character recognition method
CN105868711A (en) Method for identifying human body behaviors based on sparse and low rank
CN111242183A (en) Image identification and classification method and device based on attention mechanism

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20221130

Address after: 300392 Room 603, Building 1, No. 1, Huixue Road, Xuefu Industrial Zone, Xiqing District, Tianjin

Patentee after: Zhongfang Information Technology (Tianjin) Co.,Ltd.

Address before: 300387 Tianjin city Xiqing District West Binshui Road No. 393

Patentee before: TIANJIN NORMAL University