CN110276881A - A kind of banknote serial number recognition methods based on convolution loop neural network - Google Patents
A kind of banknote serial number recognition methods based on convolution loop neural network Download PDFInfo
- Publication number
- CN110276881A CN110276881A CN201910390398.7A CN201910390398A CN110276881A CN 110276881 A CN110276881 A CN 110276881A CN 201910390398 A CN201910390398 A CN 201910390398A CN 110276881 A CN110276881 A CN 110276881A
- Authority
- CN
- China
- Prior art keywords
- layers
- image
- banknote
- layer
- serial number
- 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
- 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
- G07—CHECKING-DEVICES
- G07D—HANDLING OF COINS OR VALUABLE PAPERS, e.g. TESTING, SORTING BY DENOMINATIONS, COUNTING, DISPENSING, CHANGING OR DEPOSITING
- G07D7/00—Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency
- G07D7/004—Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency using digital security elements, e.g. information coded on a magnetic thread or strip
- G07D7/0047—Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency using digital security elements, e.g. information coded on a magnetic thread or strip using checkcodes, e.g. coded numbers derived from serial number and denomination
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07D—HANDLING OF COINS OR VALUABLE PAPERS, e.g. TESTING, SORTING BY DENOMINATIONS, COUNTING, DISPENSING, CHANGING OR DEPOSITING
- G07D7/00—Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency
- G07D7/20—Testing patterns thereon
- G07D7/2008—Testing patterns thereon using pre-processing, e.g. de-blurring, averaging, normalisation or rotation
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07D—HANDLING OF COINS OR VALUABLE PAPERS, e.g. TESTING, SORTING BY DENOMINATIONS, COUNTING, DISPENSING, CHANGING OR DEPOSITING
- G07D7/00—Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency
- G07D7/20—Testing patterns thereon
- G07D7/2016—Testing patterns thereon using feature extraction, e.g. segmentation, edge detection or Hough-transformation
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Security & Cryptography (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of banknote serial number recognition methods based on convolution loop neural network, first acquisition banknote image, and pre-process to banknote image；The serial number area in pretreated banknote image is positioned by two-step method；The serial number area area image random division that positioning is obtained is sample areas image and target area image；Create convolution loop neural network；Finally sample areas image input convolution loop neural network is trained；Target area image is identified using the convolution loop neural network that training finishes, obtains the sequence number of corresponding bank note.The method of the present invention carries out end-to-end identification to the serial number area area image in banknote image using convolution loop neural network, obtains the sequence number of corresponding bank note；It is not necessary that explicit Character segmentation link is added in identification process, the higher monocase cutting of difficulty in traditional algorithm and monocase identification operation are avoided, the efficiency and accuracy rate of sequence number identification are improved.
Description
Technical field
The present invention relates to optical character recognition technology fields, are based on convolution loop neural network more particularly, to one kind
Banknote serial number recognition methods.
Background technique
Sequence number is that the unique identities of rmb paper currency prove information, passes through the identification of sequence number and looking into for fast database
It askes, may be implemented for national currency management.Currently used banknote serial number, which knows method for distinguishing, template matching method and nerve
Network algorithm.
Wherein the method based on template matching is to carry out feature extraction to character behind drawing template establishment library, then to each mould
Plate library is matched, but the calculation amount of template matching method is very big, and accuracy of identification is not high；And neural network algorithm has parallel
Property, study property, rapidity the advantages that, it also requires largely training template just can guarantee the accuracy of identification.
In addition in traditional sequence number recognizer, identification process is divided into two steps: Character segmentation and character recognition.But
It is higher for the Character segmentation difficulty of sequence number, for monocase cutting therein and correction process complex steps, increase knowledge
Other difficulty.
Summary of the invention
The present invention be solve existing banknote serial number identification process it is cumbersome, the problem of accuracy of identification deficiency provides one
Banknote serial number recognition methods of the kind based on convolution loop neural network.
To realize the above goal of the invention, and the technological means used is:
A kind of banknote serial number recognition methods based on convolution loop neural network, comprising the following steps:
S1. banknote image is obtained, and banknote image is pre-processed；
S2. the serial number area in pretreated banknote image is positioned by two-step method；The two-step method packet
Include: the sequence number region in banknote image is carried out coarse segmentation according to priori knowledge by the first step；Second step passes through sciagraphy
Precise positioning is carried out to the serial number area after coarse segmentation；
S3. serial number area area image random division positioning obtained is sample areas image and target area image；
S4. convolution loop neural network is created；
S5. sample areas image input convolution loop neural network is trained；The convolution finished using training
Recognition with Recurrent Neural Network identifies target area image, obtains the sequence number of corresponding bank note.
In above scheme, the sequence number in banknote image is positioned, obtains serial number area area image, then use convolution
Recognition with Recurrent Neural Network carries out end-to-end identification to target area image, obtains the sequence number of corresponding bank note；Character recognition is turned
Sequence Learning problem is turned to, it is not necessary that explicit Character segmentation link is added in identification process, avoids the higher monocase of difficulty
Cutting and monocase identification operation.
Preferably, the step S1 is specifically included:
The banknote image is subjected to gray processing；
The median filtering for the use of kernel being 3 × 3 carries out noise reduction process to the banknote image of gray processing；
Banknote image after noise reduction is subjected to the binarization operation that threshold value is 150；
The reversing situation of banknote image is judged according to the distribution situation of two-value banknote image the right and left pixel value, then is passed through
The tone of bank note bottom-left quadrant judges the front and back sides situation of banknote image.In this preferred embodiment, so that the method for the present invention can
To be suitable for the bank note of different resolution or different background color.
Preferably, the step S2 is specifically included:
Intercept respectively pretreated banknote image horizontal direction 1/4 rectangular area and pretreated banknote image
1/3 rectangular area of vertical direction；
Obtained rectangular area is subjected to floor projection, wherein including sequence number and denomination number, institute in the rectangular area
It states sequence number and denomination number and generates two relatively independent wave crests respectively, the corresponding peak amplitude of sequence number is higher than pair of denomination number
Sequence number and denomination number are separated by two relatively independent wave crests, obtain accurately serial number area by the peak amplitude answered
Image.
Preferably, convolution loop neural network described in step S4 includes convolutional layer, circulation layer and transcription layer, wherein rolling up
Lamination uses CNN network, for extracting characteristic sequence from the area image of input；Circulation layer uses RNN network, for predicting
The label distribution of the characteristic sequence extracted from convolutional layer；It transcribes layer and uses CTC model, the label for will be obtained from circulation layer point
Cloth is converted into the sequence number of corresponding bank note by duplicate removal integrated operation.
Preferably, the CNN network includes sequentially connected six convolutional coding structures, and each convolutional coding structure includes successively
The convolutional layer of connection, BN layers, ReLU layers and dropout layers of activation primitive.
Preferably, six convolutional coding structures are followed successively by the first convolutional layer, the first BN layers, the first activation primitive from top to bottom
ReLU layers, the first dropout layers, the second convolutional layer, the 2nd BN layers, ReLU layers of the second activation primitive, the 2nd dropout layers,
Three convolutional layers, the 3rd BN layers, ReLU layers of third activation primitive, the 3rd dropout layers, Volume Four lamination, the 4th BN layers, the 4th swash
Function ReLU layers, the 4th dropout layers, the 5th convolutional layer, the 5th BN layers, ReLU layers of the 5th activation primitive, the 5th dropout living
Layer, the 6th convolutional layer, the 6th BN layers, ReLU layers of the 6th activation primitive, the 6th dropout layers；Wherein the first activation primitive ReLU
After layer, ReLU layers of the second activation primitive, after ReLU layers of third activation primitive, the also company of respectively corresponding after ReLU layers of the 5th activation primitive
It is connected to the first maximum pond layer, the second maximum pond layer, third maximum pond layer, the 5th maximum pond layer.In this preferred embodiment
In, by adding BN layers in each convolutional coding structure, accelerate the convergence rate of model, and alleviating deep layer to a certain degree
In network the problem of gradient disperse, so that CNN network of the invention is trained to be more easier and stablize；By in each convolutional coding structure
In add dropout layers, prevent over-fitting, improve the generalization ability of CNN network model.
Preferably, the volume of first convolutional layer, the second convolutional layer, third convolutional layer, Volume Four lamination, the 5th convolutional layer
Product core size is 3x3, and the convolution kernel size of the 6th convolutional layer is 2x1, the described first maximum pond layer and the second maximum pond
The size of layer is 2x2, and the size of third maximum pond layer and the 5th maximum pond layer is 2x1.In this preferred embodiment, the
Three maximum pond layers and the 5th maximum pond layer use the matrix window of 2x1 size to replace common 2x2 square window, guarantee
CNN network facilitates the knowledge of character to obtain longer characteristic sequence with lateral length when extracting characteristic sequence
Not；Matrix window can produce rectangle receptive field simultaneously, help to identify narrow character, improve the essence that characteristic sequence extracts
Degree.
Preferably, the RNN network includes sequentially connected two Bi-directional LSTM elementary layer, two Bi-
Directional LSTM elementary layer includes 256 units.In this preferred embodiment, Bi-directional LSTM is used
Instead of traditional LSTM, is extracted by forward process and backward procedure extraction considers past and following feature simultaneously, it can
Simultaneously using the history and Future Data of some input in time series data, the accuracy rate of sequence number identification is improved.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
The method of the present invention obtains serial number area area image, then use by positioning to the sequence number in banknote image
Convolution loop neural network carries out end-to-end identification to target area image, obtains the sequence number of corresponding bank note；Pass through convolution
Recognition with Recurrent Neural Network is converted into Sequence Learning problem by the combination of CNN network and LSTM network, by character recognition, without identifying
Explicit Character segmentation link is added in the process, avoids the higher monocase cutting of difficulty in traditional algorithm and monocase identification behaviour
Make, the identification process for solving existing banknote serial number is cumbersome, the problem of accuracy of identification deficiency, improves banknote serial number identification
Efficiency and accuracy rate.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent actual product
Size；
To those skilled in the art, it is to be understood that certain known features and its explanation, which may be omitted, in attached drawing
's.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
A kind of banknote serial number recognition methods based on convolution loop neural network, comprising the following steps:
S1. banknote image is obtained, and banknote image is pre-processed；
The step specifically includes:
The banknote image is subjected to gray processing；
Noise reduction process is carried out using banknote image of the median filtering that kernel is 3x3 to gray processing；
Banknote image after noise reduction is subjected to the binarization operation that threshold value is 150；
The reversing situation of banknote image is judged according to the distribution situation of two-value banknote image the right and left pixel value, then is passed through
The tone of bank note bottom-left quadrant judges the front and back sides situation of banknote image.
S2. the serial number area in pretreated banknote image is positioned by two-step method；The two-step method packet
Include: the sequence number region in banknote image is carried out coarse segmentation according to priori knowledge by the first step；Second step passes through sciagraphy
Precise positioning is carried out to the serial number area after coarse segmentation；
The step specifically includes:
Intercept respectively pretreated banknote image horizontal direction 1/4 rectangular area and pretreated banknote image
1/3 rectangular area of vertical direction；
Obtained rectangular area is subjected to floor projection, wherein including sequence number and denomination number, institute in the rectangular area
It states sequence number and denomination number and generates two relatively independent wave crests respectively, the corresponding peak amplitude of sequence number is higher than pair of denomination number
Sequence number and denomination number are separated by two relatively independent wave crests, obtain accurately serial number area by the peak amplitude answered
Image.
S3. serial number area area image random division positioning obtained is sample areas image and target area image；
S4. convolution loop neural network is created；
The convolution loop neural network includes convolutional layer, circulation layer and transcription layer, and wherein convolutional layer uses CNN net
Network, for extracting characteristic sequence from the area image of input；Circulation layer uses RNN network, extracts for predicting from convolutional layer
Characteristic sequence label distribution；It transcribes layer and uses CTC model, go to reform for passing through the label obtained from circulation layer distribution
Closing operation is converted into the sequence number of corresponding bank note；Wherein CTC model, that is, CTC loss function, by the guidance of CTC loss function,
It realizes that character position is approximate with category soft to be aligned.
Wherein the CNN network includes sequentially connected six convolutional coding structures, six convolutional coding structures from top to bottom according to
It is secondary be the first convolutional layer, the first BN layers, ReLU layers of the first activation primitive, the first dropout layers, the second convolutional layer, the 2nd BN layers,
Second ReLU layers of activation primitive, the 2nd dropout layers, third convolutional layer, the 3rd BN layers, ReLU layers of third activation primitive, third
Dropout layers, Volume Four lamination, the 4th BN layers, ReLU layers of the 4th activation primitive, the 4th dropout layers, the 5th convolutional layer,
Five BN layers, ReLU layers of the 5th activation primitive, the 5th dropout layers, the 6th convolutional layer, the 6th BN layers, the 6th activation primitive ReLU
Layer, the 6th dropout layers；Wherein after ReLU layers of the first activation primitive, ReLU layers of the second activation primitive, third activation primitive ReLU
The first maximum pond layer, the second maximum pond layer, third have also been connected respectively after layer, after ReLU layers of the 5th activation primitive most
Great Chiization layer, the 5th maximum pond layer.
Wherein, the convolution of first convolutional layer, the second convolutional layer, third convolutional layer, Volume Four lamination, the 5th convolutional layer
Core size is 3x3, and the convolution kernel size of the 6th convolutional layer is 2x1, the described first maximum pond layer and the second maximum pond layer
Size be 2x2, the size of third maximum pond layer and the 5th maximum pond layer is 2x1.Third is most in the present embodiment 1
Great Chiization layer and the 5th maximum pond layer use the matrix window of 2x1 size to replace common 2x2 square window, guarantee CNN
Network facilitates the identification of character to obtain longer characteristic sequence with lateral length when extracting characteristic sequence；
Wherein, the RNN network includes sequentially connected two Bi-directional LSTM elementary layer, two Bi-
Directional LSTM elementary layer includes 256 units.
S5. sample areas image input convolution loop neural network is trained；The convolution finished using training
Recognition with Recurrent Neural Network identifies target area image, obtains the sequence number of corresponding bank note.The present embodiment 1 obtains 1832 altogether
Banknote image obtains 1500 sample areas images and 332 target area images after step S1~S3；Respectively by this
The convolution loop neural network that 1500 sample areas image input step S4 are established is trained, and the volume finished using training
Product Recognition with Recurrent Neural Network identifies 332 target area images.First by target area image normalized before identification
For the size of 32 × W, characteristic pattern then is extracted by six convolutional coding structures in CNN network, characteristic pattern is pressed column cutting, it is each
512 dimensional features of column, the Bi-directional LSTM for being input to two layers of each Unit 256 classify, and lose letter by CTC
Several guidances is realized that character position is approximate with category and soft is aligned.Then the convolution loop neural network for recycling training to finish
Target area image is identified, the sequence number of corresponding bank note is obtained.The effect finally identified is as shown in table 1 below, thus may be used
See, high using sequence number recognition accuracy of the method for the present invention to banknote image, the network model in the method for the present invention has
Good generalization ability and robustness.
| Number | Letter | Number+letter | Whole | |
| Positive exact figures | 2656 | 662 | 3318 | 330 |
| Error number | 0 | 2 | 2 | 2 |
| Sum | 2656 | 664 | 3320 | 332 |
| Accuracy rate | 100% | 99.70% | 99.94% | 99.40% |
Table 1
The terms describing the positional relationship in the drawings are only for illustration, should not be understood as the limitation to this patent；
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (8)
1. a kind of banknote serial number recognition methods based on convolution loop neural network, which comprises the following steps:
S1. banknote image is obtained, and banknote image is pre-processed；
S2. the serial number area in pretreated banknote image is positioned by two-step method；The two-step method includes:
Sequence number region in banknote image is carried out coarse segmentation according to priori knowledge by one step；Second step is by sciagraphy to rough segmentation
Serial number area after cutting carries out precise positioning；
S3. serial number area area image random division positioning obtained is sample areas image and target area image；
S4. convolution loop neural network is created；
S5. sample areas image input convolution loop neural network is trained；The convolution loop finished using training
Neural network identifies target area image, obtains the sequence number of corresponding bank note.
2. banknote serial number recognition methods according to claim 1, which is characterized in that the step S1 is specifically included:
The banknote image is subjected to gray processing；
Noise reduction process is carried out using banknote image of the median filtering that kernel is 3x3 to gray processing；
Banknote image after noise reduction is subjected to the binarization operation that threshold value is 150；
The reversing situation of banknote image is judged according to the distribution situation of two-value banknote image the right and left pixel value, then passes through bank note
The tone of bottom-left quadrant judges the front and back sides situation of banknote image.
3. banknote serial number recognition methods according to claim 1, which is characterized in that the step S2 is specifically included:
1/4 rectangular area and the pretreated banknote image for intercepting pretreated banknote image horizontal direction respectively are vertical
1/3 rectangular area in direction；
Obtained rectangular area is subjected to floor projection, wherein including sequence number and denomination number, the sequence in the rectangular area
Row number and denomination number generate two relatively independent wave crests respectively, and the corresponding peak amplitude of sequence number is higher than the corresponding of denomination number
Sequence number and denomination number are separated by two relatively independent wave crests, obtain accurately serial number area area image by peak amplitude.
4. described in any item banknote serial number recognition methods according to claim 1~3, which is characterized in that described in step S4
Convolution loop neural network include convolutional layer, circulation layer and transcription layer, wherein convolutional layer uses CNN network, for from input
Area image in extract characteristic sequence；Circulation layer uses RNN network, the mark of the characteristic sequence for predicting to extract from convolutional layer
Label distribution；It transcribes layer and uses CTC model, for the label obtained from circulation layer distribution to be converted into pair by duplicate removal integrated operation
Answer the sequence number of bank note.
5. banknote serial number recognition methods according to claim 4, which is characterized in that the CNN network includes successively connecting
Six convolutional coding structures connect, each convolutional coding structure include sequentially connected convolutional layer, BN layers, ReLU layers of activation primitive and
Dropout layers.
6. banknote serial number recognition methods according to claim 5, which is characterized in that six convolutional coding structures are from upper past
Under be followed successively by the first convolutional layer, the first BN layers, ReLU layers of the first activation primitive, the first dropout layers, the second convolutional layer, second
BN layers, ReLU layers of the second activation primitive, the 2nd dropout layers, third convolutional layer, the 3rd BN layers, ReLU layers of third activation primitive,
3rd dropout layers, Volume Four lamination, the 4th BN layers, ReLU layers of the 4th activation primitive, the 4th dropout layers, the 5th convolution
Layer, the 5th BN layers, ReLU layers of the 5th activation primitive, the 5th dropout layers, the 6th convolutional layer, the 6th BN layers, the 6th activation primitive
ReLU layers, the 6th dropout layers；Wherein after ReLU layers of the first activation primitive, ReLU layers of the second activation primitive, third activation primitive
The first maximum pond layer, the second maximum pond layer, the have also been connected respectively after ReLU layers, after ReLU layers of the 5th activation primitive
Three maximum pond layers, the 5th maximum pond layer.
7. banknote serial number recognition methods according to claim 6, which is characterized in that first convolutional layer, volume Two
Lamination, third convolutional layer, Volume Four lamination, the 5th convolutional layer convolution kernel size be 3x3, the convolution kernel of the 6th convolutional layer is big
Small is 2x1, and the size of the first maximum pond layer and the second maximum pond layer is 2x2, third maximum pond layer and the
The size of five maximum pond layers is 2x1.
8. banknote serial number recognition methods according to claim 4, which is characterized in that the RNN network includes successively connecting
The two Bi-directional LSTM elementary layers connect, two Bi-directional LSTM elementary layers include 256 lists
Member.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910390398.7A CN110276881A (en) | 2019-05-10 | 2019-05-10 | A kind of banknote serial number recognition methods based on convolution loop neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910390398.7A CN110276881A (en) | 2019-05-10 | 2019-05-10 | A kind of banknote serial number recognition methods based on convolution loop neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110276881A true CN110276881A (en) | 2019-09-24 |
Family
ID=67959042
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910390398.7A Pending CN110276881A (en) | 2019-05-10 | 2019-05-10 | A kind of banknote serial number recognition methods based on convolution loop neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110276881A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689658A (en) * | 2019-10-08 | 2020-01-14 | 北京邮电大学 | Taxi bill identification method and system based on deep learning |
| CN110766026A (en) * | 2019-10-11 | 2020-02-07 | 上海上湖信息技术有限公司 | Method and device for rapidly identifying card number and model training method and device |
| CN111275046A (en) * | 2020-01-10 | 2020-06-12 | 中科鼎富（北京）科技发展有限公司 | Character image recognition method and device, electronic equipment and storage medium |
| CN111414844A (en) * | 2020-03-17 | 2020-07-14 | 北京航天自动控制研究所 | Container number identification method based on convolution cyclic neural network |
| CN111583502A (en) * | 2020-05-08 | 2020-08-25 | 辽宁科技大学 | Renminbi (RMB) crown word number multi-label identification method based on deep convolutional neural network |
| CN111723815A (en) * | 2020-06-23 | 2020-09-29 | 中国工商银行股份有限公司 | Model training method, image processing method, device, computer system, and 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 |
| CN112101343A (en) * | 2020-08-17 | 2020-12-18 | 广东工业大学 | License plate character segmentation and recognition method |
| CN112508023A (en) * | 2020-10-27 | 2021-03-16 | 重庆大学 | Deep learning-based end-to-end identification method for code-spraying characters of parts |
| CN112990208A (en) * | 2019-12-12 | 2021-06-18 | 搜狗（杭州）智能科技有限公司 | Text recognition method and device |
| CN117789214A (en) * | 2024-02-27 | 2024-03-29 | 常熟理工学院 | Chip surface character recognition method, system and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08251405A (en) * | 1995-03-13 | 1996-09-27 | Minolta Co Ltd | Device for discriminating attribute of image area |
| US5729623A (en) * | 1993-10-18 | 1998-03-17 | Glory Kogyo Kabushiki Kaisha | Pattern recognition apparatus and method of optimizing mask for pattern recognition according to genetic algorithm |
| CN102800148A (en) * | 2012-07-10 | 2012-11-28 | 中山大学 | RMB sequence number identification method |
| CN106875546A (en) * | 2017-02-10 | 2017-06-20 | 大连海事大学 | A kind of recognition methods of VAT invoice |
| CN108776779A (en) * | 2018-05-25 | 2018-11-09 | 西安电子科技大学 | SAR Target Recognition of Sequential Images methods based on convolution loop network |
| CN108896564A (en) * | 2018-07-16 | 2018-11-27 | 山东理工大学 | A kind of two-sided damage testing method of corn seed based on machine vision |
| CN109117848A (en) * | 2018-09-07 | 2019-01-01 | 泰康保险集团股份有限公司 | A kind of line of text character identifying method, device, medium and electronic equipment |
| CN109345685A (en) * | 2018-11-09 | 2019-02-15 | 威海华菱光电股份有限公司 | Paper product discrimination method and device and storage medium |
-
2019
- 2019-05-10 CN CN201910390398.7A patent/CN110276881A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5729623A (en) * | 1993-10-18 | 1998-03-17 | Glory Kogyo Kabushiki Kaisha | Pattern recognition apparatus and method of optimizing mask for pattern recognition according to genetic algorithm |
| JPH08251405A (en) * | 1995-03-13 | 1996-09-27 | Minolta Co Ltd | Device for discriminating attribute of image area |
| CN102800148A (en) * | 2012-07-10 | 2012-11-28 | 中山大学 | RMB sequence number identification method |
| CN106875546A (en) * | 2017-02-10 | 2017-06-20 | 大连海事大学 | A kind of recognition methods of VAT invoice |
| CN108776779A (en) * | 2018-05-25 | 2018-11-09 | 西安电子科技大学 | SAR Target Recognition of Sequential Images methods based on convolution loop network |
| CN108896564A (en) * | 2018-07-16 | 2018-11-27 | 山东理工大学 | A kind of two-sided damage testing method of corn seed based on machine vision |
| CN109117848A (en) * | 2018-09-07 | 2019-01-01 | 泰康保险集团股份有限公司 | A kind of line of text character identifying method, device, medium and electronic equipment |
| CN109345685A (en) * | 2018-11-09 | 2019-02-15 | 威海华菱光电股份有限公司 | Paper product discrimination method and device and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 陈绿平: "人民币图像序列号识别及其在假币串并系统上的应用", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689658A (en) * | 2019-10-08 | 2020-01-14 | 北京邮电大学 | Taxi bill identification method and system based on deep learning |
| CN110766026A (en) * | 2019-10-11 | 2020-02-07 | 上海上湖信息技术有限公司 | Method and device for rapidly identifying card number and model training method and device |
| CN112990208B (en) * | 2019-12-12 | 2024-04-30 | 北京搜狗科技发展有限公司 | Text recognition method and device |
| CN112990208A (en) * | 2019-12-12 | 2021-06-18 | 搜狗（杭州）智能科技有限公司 | Text recognition method and device |
| CN111275046A (en) * | 2020-01-10 | 2020-06-12 | 中科鼎富（北京）科技发展有限公司 | Character image recognition method and device, electronic equipment and storage medium |
| CN111275046B (en) * | 2020-01-10 | 2024-04-16 | 鼎富智能科技有限公司 | Character image recognition method and device, electronic equipment and storage medium |
| CN111414844A (en) * | 2020-03-17 | 2020-07-14 | 北京航天自动控制研究所 | Container number identification method based on convolution cyclic neural network |
| CN111414844B (en) * | 2020-03-17 | 2023-08-29 | 北京航天自动控制研究所 | Container number identification method based on convolutional neural network |
| CN111583502A (en) * | 2020-05-08 | 2020-08-25 | 辽宁科技大学 | Renminbi (RMB) crown word number multi-label identification method based on deep convolutional neural network |
| CN111723815A (en) * | 2020-06-23 | 2020-09-29 | 中国工商银行股份有限公司 | Model training method, image processing method, device, computer system, and medium |
| CN111723815B (en) * | 2020-06-23 | 2023-06-30 | 中国工商银行股份有限公司 | Model training method, image processing device, computer system and 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 |
| CN112101343A (en) * | 2020-08-17 | 2020-12-18 | 广东工业大学 | License plate character segmentation and recognition method |
| CN112508023A (en) * | 2020-10-27 | 2021-03-16 | 重庆大学 | Deep learning-based end-to-end identification method for code-spraying characters of parts |
| CN117789214B (en) * | 2024-02-27 | 2024-04-26 | 常熟理工学院 | Chip surface character recognition method, system and storage medium |
| CN117789214A (en) * | 2024-02-27 | 2024-03-29 | 常熟理工学院 | Chip surface character recognition method, system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110276881A (en) | A kind of banknote serial number recognition methods based on convolution loop neural network | |
| CN112818862B (en) | Face tampering detection method and system based on multi-source clues and mixed attention | |
| Kussul et al. | Improved method of handwritten digit recognition tested on MNIST database | |
| Liu et al. | Offline signature verification using a region based deep metric learning network | |
| CN106372581A (en) | Method for constructing and training human face identification feature extraction network | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| CN110348319A (en) | A kind of face method for anti-counterfeit merged based on face depth information and edge image | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN103761507B (en) | Local multi-value pattern face recognition method based on Weber law | |
| CN107423727B (en) | Face complex expression recognition methods based on neural network | |
| CN108764041A (en) | The face identification method of facial image is blocked for lower part | |
| CN109522883A (en) | A kind of method for detecting human face, system, device and storage medium | |
| CN109190566A (en) | A kind of fusion local code and CNN model finger vein identification method | |
| CN107491729B (en) | Handwritten digit recognition method based on cosine similarity activated convolutional neural network | |
| CN106529395B (en) | Signature image identification method based on depth confidence network and k mean cluster | |
| CN109753950A (en) | Dynamic human face expression recognition method | |
| CN108681735A (en) | Optical character recognition method based on convolutional neural networks deep learning model | |
| CN109614866A (en) | Method for detecting human face based on cascade deep convolutional neural networks | |
| CN107220598A (en) | Iris Texture Classification based on deep learning feature and Fisher Vector encoding models | |
| CN106203373A (en) | A kind of human face in-vivo detection method based on deep vision word bag model | |
| Ebrahimpour et al. | Decision templates with gradient based features for Farsi handwritten word recognition | |
| CN109614927A (en) | Micro- Expression Recognition based on front and back frame difference and Feature Dimension Reduction | |
| CN111860266B (en) | Disguised face recognition method based on depth features | |
| Dillak et al. | Face recognition using 3D GLCM and Elman Levenberg recurrent neural network | |
| CN106228163B (en) | A kind of poor ternary sequential image feature in part based on feature selecting describes method |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190924 |