CN109359550A - Manchu document seal extraction and removal method based on deep learning technology - Google Patents
Manchu document seal extraction and removal method based on deep learning technology Download PDFInfo
- Publication number
- CN109359550A CN109359550A CN201811100870.0A CN201811100870A CN109359550A CN 109359550 A CN109359550 A CN 109359550A CN 201811100870 A CN201811100870 A CN 201811100870A CN 109359550 A CN109359550 A CN 109359550A
- Authority
- CN
- China
- Prior art keywords
- seal
- network
- document
- manchu
- language
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims description 14
- 238000013135 deep learning Methods 0.000 title claims description 13
- 238000000605 extraction Methods 0.000 title claims description 13
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 3
- 238000012360 testing method Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000005070 sampling Methods 0.000 description 16
- 238000011160 research Methods 0.000 description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 5
- 238000002360 preparation method Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004321 preservation Methods 0.000 description 2
- 230000003252 repetitive effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000003115 biocidal effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 210000004709 eyebrow Anatomy 0.000 description 1
- 210000000720 eyelash Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G06V30/41—Analysis of document content
- G06V30/413—Classification of content, e.g. text, photographs or tables
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/04—Context-preserving transformations, e.g. by using an importance map
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of language of the Manchus document seal Abstraction and minimizing technology based on depth learning technology belongs to ethnic group's document image detection identification field.Technical essential is as follows: pre-processing to language of the Manchus file and picture；Training generates network with the confrontation for extracting seal is saved；Training generates network with the confrontation for saving removal seal.Beneficial effect is: the language of the Manchus document seal Abstraction of the present invention based on depth learning technology and minimizing technology can extract seal region on language of the Manchus document to the maximum extent, thus the word under removing seal；Seal can also be removed, to restore the language of the Manchus word segment under seal to the maximum extent.
Description
Technical field
The invention belongs to ethnic group document image detection identification fields more particularly to a kind of based on deep learning technology
Language of the Manchus document seal Abstraction and minimizing technology.
Background technique
Most of archives in Manchu is all single part, only existing copy or dilute, if for a long time, high-frequency use will necessarily be to archives
Original part causes centainly to damage, in order to enable precious archives in Manchu persistence to go down, to the electronization of language of the Manchus ancient books archives
It is a kind of trend, i.e. archives in Manchu can be preserved in a manner of image, therefore to the research of archives in Manchu image and be utilized urgent
In the eyebrows and eyelashes.In language of the Manchus research process, because understanding that the talent of the language of the Manchus is deficient, the research to the language of the Manchus is caused to stagnate mostly, because
This continues to study language of the Manchus document using the method that computer is combined with deep learning, is conducive to enhance the exploitation to archives in Manchu
With utilization.Seal in one side language of the Manchus document can reflect the important information such as author or ownership of document, language of the Manchus ancient books
In seal be also identify the archives value and analysis and research archives in Manchu in content important evidence.Therefore, from the language of the Manchus
In file and picture extract seal relevant information, for language of the Manchus document research and analyse with utilize it is necessary；On the other hand it prints
Chapter, which exists, invests the literal phenomenon of the language of the Manchus, hinder to identify entire document content, and to the segmentation of document text row, Character segmentation
For research field etc., seal is noise, thus by language of the Manchus document seal remove and retain seal under language of the Manchus character it is non-
It is often significant.
The prior art passes through artificial single image mostly and handles, such as is removed using softwares such as some PS, not only time-consuming numerous
It is trivial and ineffective.
Summary of the invention
In order to solve above-mentioned problems of the prior art, the present invention proposes a kind of language of the Manchus based on deep learning technology
Document seal Abstraction and minimizing technology, this method can extract seal region on language of the Manchus document to the maximum extent, to remove
Word under seal；Seal can also be removed, to restore the language of the Manchus word segment under seal to the maximum extent.
Technical solution is as follows:
A kind of language of the Manchus document seal Abstraction and minimizing technology based on deep learning technology, steps are as follows:
S1, language of the Manchus file and picture is pre-processed；
S2, training and the confrontation generation network for saving extraction seal；
S3, training and the confrontation generation network for saving removal seal.
Further, specific step is as follows for pretreatment in step S1:
S1.1, operation is normalized in image；
S1.2, the confrontation for being prepared as confrontation generation network and removal seal that seal is extracted in training generate the data of network.
Further, specific step is as follows in step S2 and/or step S3:
S2.1, building have the generator G1 of U-net structure；
S2.2, in generator G1, if input picture is obtained by convolution several times, leakyReLU operation, BN layer operation
Dry network layer；
The last one network layer is by up-sampling UpSampling2D operation, convolution, Dropout in S2.3, step S2.2
U1 layers are obtained after layer operation, BN layer operation, connection penultimate network layer；
S2.4, u1 layers by up-sampling UpSampling2D operation, convolution, Dropout layer operation, BN layers of behaviour several times
Make, connection network layer operation obtains network output image；
S2.5, building have the arbiter D1 of the convolutional neural networks of two classification, and the network that generator G1 is generated exports
Image and true picture are fed together arbiter D1；
S2.6, target loss function is defined:
Wherein: x is image array to be processed, and y is its supervision image array, z be meet Gaussian Profile and x, y it is big
Small identical random matrix, each pixel value range of z matrix are [0,255], and D1 is determining device, and it is adjustable that G1, which is to generate λ device,
Hyper parameter, value range are [0,1].
S2.7, trained confrontation generation network model is saved.
Further, further comprising the steps of:
S4, confrontation generation network model is verified and is tested.
Further, with the data composition verifying collection verifying extraction seal for the confrontation generation network for extracting seal to antibiosis
At network model, and trim network parameter obtains the confrontation generation network model of final extraction seal, saves new model and is used for
New sample is tested；The confrontation of the data composition verifying collection verifying removal seal of network is generated with the confrontation of removal seal
Network model is generated, and trim network parameter obtains the confrontation generation network model of final removal seal, saves new model and uses
It is tested in new sample.
The beneficial effects of the present invention are:
Language of the Manchus document seal Abstraction of the present invention based on deep learning technology and minimizing technology can be to greatest extent
Ground extracts seal region on language of the Manchus document, thus the word under removing seal；Seal can also be removed, thus extensive to the maximum extent
Duplicate the language of the Manchus word segment under chapter.
Detailed description of the invention
Fig. 1 is the generation schematic network structure in the present invention with U-Net structure extraction seal region；
Fig. 2 is the generation schematic network structure with U-Net structure removal seal region in the present invention；
Fig. 3 is 2 flow chart of the embodiment of the present invention.
Specific embodiment
1-3 does further the language of the Manchus document seal Abstraction based on deep learning technology with minimizing technology with reference to the accompanying drawing
Explanation.
Embodiment 1
The method for generating the language of the Manchus document seal Abstraction and removal of network is fought based on depth, is made of four parts, respectively
Network, training are generated with the confrontation for saving extraction seal region for the pretreatment of language of the Manchus file and picture, training and save removal seal
Confrontation generate the result of network, the non-test sample of test.
Step 1: language of the Manchus file and picture pretreatment
1.1 normalization
The picture that scanning is collected is pre-processed, picture size is uniformly normalized to operation, is unified for 2048*
2992.The size for normalizing size is adjustable, should go a size appropriate according to all acquisition image sizes, avoid figure
Piece distortion is serious.
1.2 data preparation
Data1: the data that network prepares, the supervision of lid seal image and it are generated for the confrontation that seal region is extracted in training
Image, that is, seal area image is 1 group of input, totally 100 groups of data.
Data2: for training removal seal Document retrieval confrontation generate network prepare data, lid seal image and it
Supervision image be the file and picture of seal be 1 group of input, totally 100 groups of data.
Step 2: training generates network with the confrontation for extracting seal region is saved
The confrontation that seal region is extracted in 2.1 buildings generates network N et1
Confrontation generates network and is made of two sub-networks, i.e. generator G1 and arbiter D1.
Generator G1: using the encoder network with U-net structure, constructing generation encoder as shown in Figure 1,
In, d0 is input picture, and d0 operates to obtain d1 by convolution (64 4*4 cores, step-length 2), leakyReLU；D1 passes through convolution
(128 4*4 cores, step-length 2), leakyReLU, BN layer operation obtain d2；Passing through convolution, (256 4*4 cores, step-length are d2
2), leakyReLU, BN layer operation obtain d3；D3 is by convolution (512 4*4 cores, step-length 2), leakyReLU, BN layers of behaviour
Obtain d4；D4 is obtaining d5 by convolution (512 4*4 cores, step-length 2), leakyReLU, BN layer operation；D5 is through pulleying
Product (512 4*4 cores, step-length 2), leakyReLU, BN layer operation obtain d6；D6 is passing through convolution (512 4*4 cores, step-length
For 2), leakyReLU, BN layer operation obtain d7；D7 is by up-sampling UpSampling2D (size=2), convolution (512*1*
1), Dropout layers, BN layers, connection d6 layers after obtain u1；U1 is by up-sampling UpSampling2D (size=2), convolution
(512*1*1), Dropout layers, BN layers, connection d5 layers after obtain u2；U2 process up-sampling UpSampling2D (size=2),
Convolution (512*1*1), Dropout layers, BN layers, connection d4 layers after obtain u3；U3 is by up-sampling UpSampling2D (size=
2), convolution (256*1*1), Dropout layers, BN layers, connection d3 layers after obtain u4；U4 is by up-sampling UpSampling2D
(size=2), convolution (128*1*1), Dropout layers, BN layers, connection d2 layers after obtain u5；U5 is by up-sampling
UpSampling2D (size=2), convolution (128*1*1), Dropout layers, BN layers, connection d1 layers after obtain u6；U6 is by upper
It is that network exports picture that u7 is obtained after sampling UpSampling2D (size=2), convolution (3*4*4, step-length 1).
The convolutional neural networks of arbiter D1: one two classification: the picture that generator generates is sent together with true picture
Enter arbiter D1, the structure of D1 is that (128*4*4, step-length are convolutional layer (64*4*4, step-length 2) → LeakyReLU → convolutional layer
2) → LeakyReLU → BN layers of (momentum of → LeakyReLU → BN layers of (momentum 0.8) → convolutional layer (256*4*4, step-length 2)
For 0.8) → convolutional layer (512*4*4, step-length 2) → LeakyReLU → BN layers of (momentum 0.8) → convolutional layer (1*4*4, step
It is a length of 1).
The network that 2.2 training are built
Define target loss function:
Using the generator G1 and arbiter D1 constructed in 2.1, using Adam stochastic gradient descent optimizer, above formula is
Objective function repetitive exercise.Setting the number of iterations epoch=10000 times.
Above formula is divided to two, and first item generational loss, Section 2 is reconstruct loss.Wherein x is image array to be processed, y
For its supervision image array, z be meet Gaussian Profile and the identical random matrix of x, y size, each pixel value of z matrix
Range [0,255].X is the language of the Manchus document with seal of acquisition, and in seal Abstraction task, y is that no document text only has
The supervision image of seal；In seal removal task, y is the supervision image of the not only document text of seal.It is generated in G1
Device, G1It is the image that generates after input G1 that (x, z), which is with x, z,；D1 is arbiter, D1(x, y) is to be inputted in arbiter D1 with x, y
Loss, D1(x,G1(x, z)) it is by the G of x and generation1The loss of (x, z) feeding arbiter D1.Last is reconstruct loss,
As supervision image y and the L1 distance for generating image, wherein λ is adjustable hyper parameter, and value range is [0,1], and E is to ask about it
The expectation of footmark.
2.3, which save the trained confrontation for extracting seal region, generates network model
Threshold value is set, current trained network model is saved when loss is less than threshold value, finally can be reserved for multiple
Which effect of the model of preservation will be compared because current loss is the loss of training data on test set for trained model
Fruit is better.
Step 3: training generates network with the confrontation for saving removal seal
3.1 is consistent with the network structure in second step, but independent mutually with second network.Independent input, it is independent
Training, is independently arranged all parameters.
The confrontation of 3.2 building removal seals generates network N et2
Construct the target image difference that network structure is identical as 2.2, only exports.
Generator G2: where d0 is input picture, and d0 is grasped by convolution (64 4*4 cores, step-length 2), leakyReLU
Obtain d1；D1 obtains d2 by convolution (128 4*4 cores, step-length 2), leakyReLU, BN layer operation；D2 is passing through convolution
(256 4*4 cores, step-length 2), leakyReLU, BN layer operation obtain d3；Passing through convolution, (512 4*4 cores, step-length are d3
2), leakyReLU, BN layer operation obtain d4；D4 is by convolution (512 4*4 cores, step-length 2), leakyReLU, BN layers of behaviour
Obtain d5；D5 is obtaining d6 by convolution (512 4*4 cores, step-length 2), leakyReLU, BN layer operation；D6 is through pulleying
Product (512 4*4 cores, step-length 2), leakyReLU, BN layer operation obtain d7；D7 is by up-sampling UpSampling2D (size
=2), convolution (512*1*1), Dropout layers, BN layers, connection d6 layers after obtain u1；U1 is by up-sampling UpSampling2D
(size=2), convolution (512*1*1), Dropout layers, BN layers, connection d5 layers after obtain u2；U2 is by up-sampling
UpSampling2D (size=2), convolution (512*1*1), Dropout layers, BN layers, connection d4 layers after obtain u3；U3 is by upper
Sampling UpSampling2D (size=2), convolution (256*1*1), Dropout layer, BN layers, connect d3 layers after obtain u4；U4 warp
Cross up-sampling UpSampling2D (size=2), convolution (128*1*1), Dropout layers, BN layers, connection d2 layers after obtain u5；
U5 is obtained after up-sampling UpSampling2D (size=2), convolution (128*1*1), Dropout layers, BN layers, d1 layers of connection
u6；It is network output figure that u6 obtains u7 after up-sampling UpSampling2D (size=2), convolution (3*4*4, step-length 1)
Piece.
The convolutional neural networks of arbiter D2: one two classification: the picture that generator generates is sent together with true picture
Enter arbiter D1, the structure of D1 is that (128*4*4, step-length are convolutional layer (64*4*4, step-length 2) → LeakyReLU → convolutional layer
2) → LeakyReLU → BN layers of (momentum of → LeakyReLU → BN layers of (momentum 0.8) → convolutional layer (256*4*4, step-length 2)
For 0.8) → convolutional layer (512*4*4, step-length 2) → LeakyReLU → BN layers of (momentum 0.8) → convolutional layer (1*4*4, step
It is a length of 1).
The network that 3.2 training are built
Using the generator G2 and arbiter D2 constructed in 3.1, using Adam stochastic gradient descent optimizer, in 2.2
Loss formula be objective function repetitive exercise.Setting the number of iterations epoch=10000 times.
3.3 confrontation for saving trained removal seal generate network model
Threshold value is set, current trained network model Model2 is saved when loss is less than threshold value.
Step 4: verification and testing
The network model saved in 2.3 is verified with the verifying collection in Data1, and trim network parameter obtains final extraction
The network model Model1 in language of the Manchus seal region, saving model1 can be to new test sample.
The network model saved in 3.3 is verified with the verifying collection in Data2, and trim network parameter obtains final extraction
The network model Model2 in seal region is removed, saving model2 can be to new test sample.
Embodiment 2
1. data preparation and pretreatment
(1) collection of language of the Manchus file and picture can be the modes such as to scan, take pictures to obtain from language of the Manchus ancient books document accordingly
Language of the Manchus file and picture.
(2) image preprocessing, normalized image size.
(3) data preparation of seal extracted region network: band seal file and picture and target image only seal administrative division map
As right.
(4) data preparation of seal removal network: band seal file and picture and target image only file and picture pair.
2. building confrontation generates network
It realizes that seal Abstraction building confrontation generates network N et1, is made of generator G1 (as shown in Figure 1), arbiter D1.
It realizes that seal removal building confrontation generates network N et2, is made of generator G2 (as shown in Figure 2), arbiter D2.
Generating building for network of confrontation can be built by Open Source Platforms such as TensorFlow, Keras.
3. parameter is arranged
Epochs: for total iteration wheel number
Batch_size=1 is that sample size is arranged in each round iteration
The selection of optimizer: Adam:adaptive moment estimation, adaptive moments estimation.Square in probability theory
Be meant that: if a stochastic variable X obeys some distribution, the first moment of X is E (X), that is, sample mean, the two of X
Rank square is exactly E (X^2), that is, the average value of sample square.Adam algorithm is according to loss function to the gradient of each parameter
Single order moments estimation and second order moments estimation dynamic adjustment are directed to the learning rate of each parameter.Adam is also based on gradient decline
Method, but the Learning Step of iterative parameter has a determining range every time, will not cause because of very big gradient very big
Learning Step, the value of parameter is more stable.
4. start that preservation model is trained to start to train with backpropagation optimizer, in Net1, first to train arbiter D1,
Arbiter D1 is used to differentiate that input picture is that true picture or generator generate, and then generates the print extracted in training generator
The ability of chapter picture, then generator and arbiter confrontation learn, to all achieve the effect that.Finally save trained life
It grows up to be a useful person model.
5. verification and testing
It goes to verify by the picture for not appearing in training set, precision is continuously improved according to the parameter that precision adjusts model.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art within the technical scope of the present disclosure, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (5)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811100870.0A CN109359550B (en) | 2018-09-20 | 2018-09-20 | Manchu document seal extraction and removal method based on deep learning technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811100870.0A CN109359550B (en) | 2018-09-20 | 2018-09-20 | Manchu document seal extraction and removal method based on deep learning technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109359550A true CN109359550A (en) | 2019-02-19 |
| CN109359550B CN109359550B (en) | 2021-06-22 |
Family
ID=65351009
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811100870.0A Active CN109359550B (en) | 2018-09-20 | 2018-09-20 | Manchu document seal extraction and removal method based on deep learning technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359550B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163194A (en) * | 2019-05-08 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN110516201A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, device, electronic device and storage medium |
| CN110516577A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN112183538A (en) * | 2020-11-30 | 2021-01-05 | 华南师范大学 | A Manchu language recognition method and system |
| CN112801911A (en) * | 2021-02-08 | 2021-05-14 | 苏州长嘴鱼软件有限公司 | Method and device for removing Chinese character noise in natural image and storage medium |
| CN112950458A (en) * | 2021-03-19 | 2021-06-11 | 润联软件系统(深圳)有限公司 | Image seal removing method and device based on countermeasure generation network and related equipment |
| CN113065407A (en) * | 2021-03-09 | 2021-07-02 | 国网河北省电力有限公司 | A Seal Erasing Method for Financial Notes Based on Attention Mechanism and Generative Adversarial Network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787510A (en) * | 2016-02-26 | 2016-07-20 | 华东理工大学 | System and method for realizing subway scene classification based on deep learning |
| US20170061249A1 (en) * | 2015-08-26 | 2017-03-02 | Digitalglobe, Inc. | Broad area geospatial object detection using autogenerated deep learning models |
| CN107220506A (en) * | 2017-06-05 | 2017-09-29 | 东华大学 | Breast cancer risk assessment analysis system based on deep convolutional neural network |
| CN108470196A (en) * | 2018-02-01 | 2018-08-31 | 华南理工大学 | A method of handwritten numeral is generated based on depth convolution confrontation network model |
-
2018
- 2018-09-20 CN CN201811100870.0A patent/CN109359550B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170061249A1 (en) * | 2015-08-26 | 2017-03-02 | Digitalglobe, Inc. | Broad area geospatial object detection using autogenerated deep learning models |
| CN105787510A (en) * | 2016-02-26 | 2016-07-20 | 华东理工大学 | System and method for realizing subway scene classification based on deep learning |
| CN107220506A (en) * | 2017-06-05 | 2017-09-29 | 东华大学 | Breast cancer risk assessment analysis system based on deep convolutional neural network |
| CN108470196A (en) * | 2018-02-01 | 2018-08-31 | 华南理工大学 | A method of handwritten numeral is generated based on depth convolution confrontation network model |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163194A (en) * | 2019-05-08 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN110516201A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, device, electronic device and storage medium |
| CN110516577A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN110516577B (en) * | 2019-08-20 | 2022-07-12 | Oppo广东移动通信有限公司 | Image processing method, device, electronic device and storage medium |
| CN110516201B (en) * | 2019-08-20 | 2023-03-28 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN112183538A (en) * | 2020-11-30 | 2021-01-05 | 华南师范大学 | A Manchu language recognition method and system |
| CN112183538B (en) * | 2020-11-30 | 2021-03-02 | 华南师范大学 | Manchu recognition method and system |
| CN112801911A (en) * | 2021-02-08 | 2021-05-14 | 苏州长嘴鱼软件有限公司 | Method and device for removing Chinese character noise in natural image and storage medium |
| CN112801911B (en) * | 2021-02-08 | 2024-03-26 | 苏州长嘴鱼软件有限公司 | Method and device for removing text noise in natural image and storage medium |
| CN113065407A (en) * | 2021-03-09 | 2021-07-02 | 国网河北省电力有限公司 | A Seal Erasing Method for Financial Notes Based on Attention Mechanism and Generative Adversarial Network |
| CN113065407B (en) * | 2021-03-09 | 2022-07-12 | 国网河北省电力有限公司 | Financial bill seal erasing method based on attention mechanism and generation countermeasure network |
| CN112950458A (en) * | 2021-03-19 | 2021-06-11 | 润联软件系统(深圳)有限公司 | Image seal removing method and device based on countermeasure generation network and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109359550B (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109359550A (en) | Manchu document seal extraction and removal method based on deep learning technology | |
| CN107766324B (en) | Text consistency analysis method based on deep neural network | |
| CN108827605B (en) | An automatic extraction method of mechanical fault features based on improved sparse filtering | |
| CN110443293B (en) | Zero sample image classification method for generating confrontation network text reconstruction based on double discrimination | |
| CN111898447B (en) | Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method | |
| CN109116203A (en) | Power equipment partial discharges fault diagnostic method based on convolutional neural networks | |
| CN114067368B (en) | Classification and identification method of power grid hazard bird species based on deep convolution features | |
| CN103268607B (en) | A kind of common object detection method under weak supervision condition | |
| CN107451760B (en) | Rolling bearing fault diagnosis method based on time window slip limited Boltzmann machine | |
| CN109886021A (en) | A Malicious Code Detection Method Based on API Global Word Vector and Hierarchical Recurrent Neural Network | |
| CN104700100A (en) | Feature extraction method for high spatial resolution remote sensing big data | |
| CN113627401A (en) | EMG gesture recognition method with feature pyramid network fused with dual attention mechanism | |
| CN104239897A (en) | Visual feature representing method based on autoencoder word bag | |
| CN100461217C (en) | A Method of Image Texture Segmentation Based on Complexity Measure | |
| CN113076878B (en) | Constitution Identification Method Based on Convolutional Network Structure of Attention Mechanism | |
| CN109492625A (en) | A kind of human face identification work-attendance checking method based on width study | |
| CN111275108A (en) | A method for sample expansion of partial discharge data based on generative adversarial network | |
| CN103646255A (en) | Face detection method based on Gabor characteristics and extreme learning machine | |
| KR102635610B1 (en) | Seismic data generation method and device using deep learning | |
| CN108509601A (en) | A kind of flavour of food products assessment method based on big data analysis | |
| CN112765908A (en) | Gas turbine engine combustion chamber stability analysis method based on image sequence analysis | |
| CN110609477A (en) | A system and method for determining transient stability of power systems based on deep learning | |
| CN116502113A (en) | Sand shale recognition method based on XGBoost algorithm | |
| CN103077408A (en) | Method for converting seabed sonar image into acoustic substrate classification based on wavelet neutral network | |
| CN109685071A (en) | Brain electricity classification method based on the study of common space pattern feature width |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |