CN110210410A - A kind of Handwritten Digit Recognition method based on characteristics of image - Google Patents
A kind of Handwritten Digit Recognition method based on characteristics of image Download PDFInfo
- Publication number
- CN110210410A CN110210410A CN201910481876.5A CN201910481876A CN110210410A CN 110210410 A CN110210410 A CN 110210410A CN 201910481876 A CN201910481876 A CN 201910481876A CN 110210410 A CN110210410 A CN 110210410A
- Authority
- CN
- China
- Prior art keywords
- handwritten
- area
- image
- convolution
- region
- 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 abstract description 42
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims abstract description 9
- 238000012800 visualization Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 24
- 210000002569 neuron Anatomy 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- 230000035945 sensitivity Effects 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 238000005316 response function Methods 0.000 claims description 8
- 230000008676 import Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000007480 spreading Effects 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 210000005036 nerve Anatomy 0.000 claims 1
- 238000012360 testing method Methods 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010252 digital analysis Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/382—Preprocessing; Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The Handwritten Digit Recognition method based on characteristics of image that the invention discloses a kind of, comprising the following steps: (1) image preprocessing；(2) it is based on pretreated handwritten numeral image data set, network model, input area, convolution region, pond region and the output area of planned network is constructed, and be trained, finally trains the network model for Handwritten Digit Recognition；(3) visualization system interface is constructed, shows the process for importing handwritten numeral image, pretreatment and identification number.The present invention is based on original images to carry out denoising and the operation of de-redundancy feature, and the structure and parameter of the network model based on digital picture feature construction for identification has faster recognition speed, higher accuracy and preferable recognition effect；Show the process of analysis and identification number using graphic interface, its analysis and identification process are shown with visualization interface, effectively meet the requirement currently to Handwritten Digit Recognition.
Description
Technical Field
The invention relates to an analysis and recognition method of image information, in particular to a handwritten number recognition method based on image characteristics.
Background
Handwritten digit recognition technology is one of the main areas of pattern recognition research. In recent years, with the development of handwriting recognition technology, recognition of numbers in documents has reached a good recognition rate, and many have been commercially used. Due to the economic development of the current, the financial market is increasingly developed, the bill business is developed quickly, for example, bills such as personal certificates, checks, invoices, incoming bills and the like need to process a large amount of information, if the information is input by manpower, a large amount of manpower and material resources are wasted undoubtedly, and the problems of high cost, low efficiency and the like are caused. Therefore, handwritten digit recognition appears to be very necessary. The existing handwritten number recognition methods are mainly two, namely a number recognition method based on a proximity algorithm and a number recognition method based on a support vector machine.
The digital identification method based on the proximity algorithm mainly has an excellent classification effect on text classification, and comprises tasks of information retrieval, information filtering and the like. However, the proximity algorithm still has disadvantages, which are mainly represented by that when the size of a high-dimensional text vector sample is large, the time and space complexity of the algorithm is high, and when a new sample to be classified comes, the distance (or similarity) between the new sample and all training samples needs to be calculated every time, which greatly reduces the efficiency of the algorithm.
The support vector machine-based number recognition method is a supervised learning model, and is commonly used for pattern recognition, classification, and regression analysis. Compared with the traditional model, the support vector machine has obviously higher accuracy in solving the small sample data, but the support vector machine still has the defects that the difficulty of feature extraction is high, and the performance of the system is directly influenced by the quality of the feature extraction.
In practical application, the two methods have certain defects, which are often expressed as poor anti-noise capability of the model, and some redundant features are easy to extract, and when the number of handwritten digital pixels is low or the image is blurred, the recognition accuracy is low. Therefore, the current identification technology of the handwritten numbers is difficult to meet the requirements of the current identification of the handwritten numbers.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the defects in the prior art, the invention aims to provide a handwritten number recognition method based on image characteristics, which can realize high-accuracy recognition of some handwritten numbers with low pixels and low definition, further display the analysis and recognition processes of the handwritten numbers by a visual interface, and effectively meet the current requirements on handwritten number recognition.
The technical scheme is as follows: a handwritten digit recognition method based on image characteristics comprises the following steps:
(1) carrying out preprocessing operation on the image;
(2) constructing a network model, designing an input area, a convolution area, a pooling area and an output area of the network, and training the preprocessed handwritten digital image data set to obtain the network model for handwritten digital recognition;
(3) and constructing a visual system interface, and displaying the processes of importing the handwritten digital image, preprocessing and recognizing the number. The visualization system comprises an image import module, a preprocessing module and a handwritten number recognition module; the image import module is used for acquiring a handwritten digital image, the preprocessing module preprocesses and displays the handwritten digital image, and the handwritten digital recognition module recognizes handwritten digits and outputs the digits and recognized duration by using a trained handwritten digit recognition network model.
Further, the step (1) specifically includes the following steps:
(1.1) denoising the imported original handwritten digital image;
(1.2) reshaping the image and graying the image, i.e.
Y＝0.3R+0.59G+0.11B
Where Y is the processed value and R, G and B are the three components of the image chromaticity, respectively;
(1.3) carrying out the value normalization of the pixel matrix, and scaling the pixel value range from 0-255 to 0-1.
Further, the step (2) specifically includes the following steps:
(2.1) determining an input area and an output area of the prediction network model, wherein the preprocessed image data is m2Each pixel point is converted into an m multiplied by m matrix to be used as the input of the network; networkThe output of (1) is 10 numbers in total, and each number corresponds to a target vector, so that the output area of the network is set to be 10 network nodes;
(2.2) determining a middle region; the middle area comprises a convolution area with two layers of enhanced image characteristics and two corresponding down-sampling areas; 32 feature maps are obtained from the input area to the first convolution area using 32 convolution filters, and after down-sampling, the second convolution area contains 64 feature maps, each connected to each map of the previous layer. The network structure of an optimal middle area can also be determined by trial and error by gradually increasing or decreasing the number of filters and comparing the training results.
And (2.3) using a linear rectification function as a response function, and using an error gradient reduction method as a training algorithm to adjust the weight and the threshold of each region, so that the training error mean square value is minimized and approaches to the target output value of the network model.
The step (2.3) comprises:
(2.3.1) Forward propagation phase, i.e. one sample (X) of the datasetp,Yp) Inputting the data into the network model, and finally calculating corresponding actual output O through region-by-region transformationp(ii) a The error is measured at this stage using a squared error cost function:
assuming the class is class C and there are N training samples, the error is ENComprises the following steps:
wherein,the k-th dimension representing the target output of the nth sample,a k-dimension representing the actual output of the nth sample; error E of nth samplenExpressed as:
(2.3.2) phase of back propagation, i.e. actual output of OpCorresponding target output YpAnd (4) reversely spreading the errors according to the principle of minimizing the errors and adjusting corresponding weights.
In this stage:
sensitivity of the basis defining the neuron is:
where E denotes the error between the actual output value and the expected output value in the forward propagation, u denotes the input, b denotes the basis of the neuron, and the sensitivity of the i-th layer is expressed as:
wherein,the operator represents the multiplication of each element of the matrix, the T operator represents the transposition of the matrix, and f is a response function;
and (4) adjusting weight parameters of a convolution area and a downsampling area in the network model by using a gradient descent method.
Specifically, the weight parameter of the downsampling area is adjusted by the following formula:
wherein,represents the weight of the ith neuron in the jth downsampling area, the down function represents the downsampling function,representing the bias of the ith neuron in the jth downsampling region,represents the weight of the ith-1 neuron in the jth downsampling region,represents the base of the ith neuron in the jth downsampling region, and f is the activation function of the layer.
And adjusting the weight parameter of the convolution region by adopting the following formula:
wherein,representing the sensitivity of the ith neuron in the jth convolution region,represents the bias of the (i + 1) th neuron of the jth convolution region,representing the ith neuron of the jth convolution regionThe input of the input data is carried out,and (3) the output of the (i + 1) th neuron of the jth convolution region is shown, the up function is an up sampling function, and f is an activation function of the layer.
Has the advantages that: compared with the prior art, the invention has the following remarkable progress: 1. the invention carries out the operation of removing dryness and redundancy characteristics based on the original image, has faster identification speed and better identification accuracy: 2. the structure and parameters of the network model for identification are constructed based on the digital image characteristics, so that the method has high accuracy and good identification effect; 3. the invention adopts a graphical interface to display the processes of analyzing and identifying numbers.
Drawings
FIG. 1 is a flowchart of the operation of a handwritten digit recognition method based on image features in the present invention;
FIG. 2 is a flow chart of the operation of the present invention for building a network model for handwritten digit recognition;
FIG. 3 is a schematic diagram of the interface of the handwritten digit recognition system established in the present invention.
Detailed Description
The following detailed description of the embodiments of the present invention will be provided with reference to the drawings and examples, so as to fully understand how to implement the technical solution of the present invention and achieve the technical effects. It should be noted that, as long as there is no conflict, the embodiments and the features of the embodiments of the present invention may be combined with each other, and the technical solutions formed are within the scope of the present invention.
Additionally, the steps illustrated in the flow charts of the figures may be performed in a computer system such as a set of computer-executable instructions and, although a logical order is illustrated in the flow charts, in some cases, the steps illustrated or described may be performed in an order different than here.
The invention carries out preprocessing operations such as graying, denoising, normalization and the like on the image based on the original handwritten digital image, thereby being convenient for extracting the key characteristics of the image more quickly when establishing a recognition network model. A large amount of handwritten digital image data is used as a training set, a network model for digital recognition is trained, and a handwritten digital analysis recognition system based on the model is established, so that the problems of low accuracy rate of handwritten digital recognition and the like are solved.
As shown in fig. 1, a method for recognizing handwritten numbers based on image features includes the following steps:
(1) image pre-processing
De-noising the imported original handwritten digital image, reshaping the image to 28 x 28 pixel size, and graying the image, i.e. de-noising the image
Y＝0.3R+0.59G+0.11B
Where Y is the processed value and R, G and B are the three components of the image chromaticity, respectively. And finally, carrying out numerical value normalization on the pixel matrix, and scaling the pixel value range from 0-255 to 0-1. After the series of image preprocessing, the characteristics of the handwritten numbers are more highlighted, and the subsequent characteristic analysis is facilitated.
(2) Establishment of hand-written digit recognition network model
The module constructs a network model based on a preprocessed handwritten digital image data set, designs an input area, a convolution area, a pooling area and an output area of a network, trains and finally trains the network model for handwritten digital recognition.
Specifically, the construction method of the network model for identifying the handwritten numbers comprises the following steps:
step 1, determining an input area and an output area of the prediction network model, wherein the preprocessed image data are 784 pixel points, and converting the pixel points into a matrix of 28 multiplied by 28 to be used as the input of the network. The output of the network is 10 numbers of 0-9, and each number corresponds to one target vector, so that the output area of the network is set to be 10 network nodes.
And 2, determining a middle area which is an important bridge connecting the input area and the output area of the network and determines the structure and the performance of the whole network. The middle region includes two layers of convolution regions of enhanced image features and corresponding two downsampled regions. 32 feature maps are obtained from the input area to the first convolution area using 32 convolution filters, and after down-sampling, the second convolution area contains 64 feature maps, each connected to each map of the previous layer. During the training process, the training results can be compared through repeated experiments of gradually increasing or decreasing the number of filters, so as to determine the network structure of an optimal middle area.
And 3, determining a response function and a training algorithm, using a linear rectification function as the response function, exerting good differentiability and nonlinear mapping capability of the response function, and using a method of error gradient reduction as the training algorithm to adjust the weight and threshold of each region, so as to minimize the mean square value of the training error and finally approach the target output value of the network model.
After the network construction is finished, the network can be trained by using the preprocessed image data, and the network training method mainly comprises two stages: the first phase is the forward propagation phase, i.e. one sample (X) of the data setp,Yp) Inputting the data into the network model, and finally calculating corresponding actual output O through region-by-region transformationp(ii) a The second phase is the backward propagation phase, i.e. the actual output OpCorresponding target output YpAnd (4) reversely spreading the errors according to the principle of minimizing the errors and adjusting corresponding weights.
Referring to fig. 2, specifically, the training method is as follows:
step 1, using a square error cost function to measure errors in a forward propagation stage, setting the category as C, and if N training samples are in total, determining an error ENComprises the following steps:
wherein,the k-th dimension representing the target output of the nth sample,representing the k-th dimension of the actual output of the nth sample. Error E of the nth sample since the errors of all training samples are simply added togethernCan be expressed as:
step 2, the back propagation phase introduces a concept, i.e. the sensitivity of the basis, which represents the back propagated error, which is the rate of change of the error to the basis b, expressed as:
where E denotes error, u denotes input, b denotes the basis of the neuron, and thus the sensitivity of the i-th layer can be expressed as:
wherein,the operator represents the multiplication of each element of the matrix, the T operator represents the transposition of the matrix, f is a response function, and then the weight can be updated by applying a delta rule.
Step 3, updating the weight of the downsampling area, wherein the down function represents a downsampling function, summing pixels in the n multiplied by n field of the input feature map, each feature map corresponds to different biases β and the base b of a neuron, f is an activation function, x is the weight, and the calculation formula is as follows:
and 4, updating the weight of the convolution area, performing convolution calculation on the input characteristic diagram and a convolution kernel which can be trained, adding a bias term, and finally obtaining an output characteristic diagram through an activation function, wherein the input characteristic diagram can have different combination modes. The sensitivity delta of the base of each neuron on the convolution region i is first determinediThe corresponding sensitivity map in the lower sampling area is up-sampled, the size of the sensitivity map after up-sampling is consistent with the size of the characteristic diagram of the convolution area i, j is the index of the lower sampling area, the up function is the up-sampling function, and the calculation formula is expressed as follows:
(3) construction of a visualization System
The visualization system comprises three logically independent functional modules: the device comprises an image import module, a preprocessing module and a handwritten number recognition module. In the image import module, a digital picture written by a PC end by using the drawing board can be selected to be read, and the image import module can also be connected with a mobile device to transmit a shot handwritten digital picture. In the preprocessing module, the imported original handwritten digital image is subjected to denoising processing, then the image is shaped and reformed into 28 × 28 pixels, and then the image is subjected to graying and normalization processing and displayed. In the hand-written digit recognition module, the trained hand-written digit recognition network model is used for recognizing hand-written digits and outputting the digits and the recognition duration. A schematic diagram of the graphical software is shown in fig. 3.
Claims (10)
1. A handwritten digit recognition method based on image features is characterized by comprising the following steps:
(1) carrying out preprocessing operation on the image;
(2) constructing a network model, designing an input area, a convolution area, a pooling area and an output area of the network, and training the preprocessed handwritten digital image data set to obtain the network model for handwritten digital recognition;
(3) and constructing a visual system interface, and displaying the processes of importing the handwritten digital image, preprocessing and recognizing the number.
2. The method according to claim 1, wherein the step (1) comprises the following steps:
(1.1) denoising the imported original handwritten digital image;
(1.2) reshaping the image and graying the image, i.e.
Y＝0.3R+0.59G+0.11B
Where Y is the processed value and R, G and B are the three components of the image chromaticity, respectively;
(1.3) carrying out the value normalization of the pixel matrix, and scaling the pixel value range from 0-255 to 0-1.
3. The method for recognizing handwritten numbers based on image features as claimed in claim 1, wherein the step (2) specifically comprises the following steps:
(2.1) determining an input area and an output area of the prediction network model, wherein the preprocessed image data is m2Each pixel point is converted into an m multiplied by m matrix to be used as the input of the network; the output of the network is 10 numbers which are 0-9, and each number corresponds to one target vector, so that the output area of the network is set to be 10 network nodes;
(2.2) determining a middle region;
and (2.3) using a linear rectification function as a response function, and using an error gradient reduction method as a training algorithm to adjust the weight and the threshold of each region, so that the training error mean square value is minimized and approaches to the target output value of the network model.
4. The method of claim 3, wherein the handwritten digit recognition method comprises: in the step (2.2), the middle area comprises a convolution area of two layers of enhanced image features and two corresponding down-sampling areas; 32 feature maps are obtained from the input area to the first convolution area using 32 convolution filters, and after down-sampling, the second convolution area contains 64 feature maps, each connected to each map of the previous layer.
5. The method of claim 3, wherein the step (2.2) further comprises: and (4) determining an optimal network structure of the middle area by repeatedly testing the number of filters gradually increased or decreased and comparing training results.
6. The method of claim 3, wherein the step (2.3) comprises:
forward propagation phase, i.e. one sample (X) of the data setp,Yp) Inputting the data into the network model, and finally calculating corresponding actual output O through region-by-region transformationp；
Phase of back propagation, i.e. outputting the actual output OpCorresponding target output YpAnd (4) reversely spreading the errors according to the principle of minimizing the errors and adjusting corresponding weights.
7. The method of claim 6, wherein the forward propagation stage measures the error using a squared error cost function:
assuming the class is class C and there are N training samples, the error is ENComprises the following steps:
wherein,the k-th dimension representing the target output of the nth sample,a k-dimension representing the actual output of the nth sample; error E of nth samplenExpressed as:
8. the method of claim 6, wherein in the back propagation phase:
sensitivity of the basis defining the neuron is:
where E denotes the error between the actual output value and the expected output value in the forward propagation, u denotes the input, b denotes the basis of the neuron, and the sensitivity of the i-th layer is expressed as:
wherein,the operator represents the multiplication of each element of the matrix, the T operator represents the transposition of the matrix, and f is a response function;
and (4) adjusting weight parameters of a convolution area and a downsampling area in the network model by using a gradient descent method.
9. The method of claim 8, wherein the handwritten digit recognition method comprises:
and adjusting the weight parameter of the downsampling area by adopting the following formula:
wherein,represents the weight of the ith neuron in the jth downsampling area, the down function represents the downsampling function,representing the bias of the ith neuron in the jth downsampling region,represents the weight of the ith-1 neuron in the jth downsampling region,representing the base of the ith neuron of the jth downsampling area, and f is the activation function of the layer;
and adjusting the weight parameter of the convolution region by adopting the following formula:
wherein,representing the sensitivity of the ith neuron in the jth convolution region,represents the bias of the (i + 1) th neuron of the jth convolution region,representing the ith nerve of the jth convolution regionThe input of the element is carried out,and (3) the output of the (i + 1) th neuron of the jth convolution region is shown, the up function is an up sampling function, and f is an activation function of the layer.
10. The method of recognizing handwritten figures based on image features of claim 1, wherein: the visualization system comprises an image import module, a preprocessing module and a handwritten number recognition module; the image import module is used for acquiring a handwritten digital image, the preprocessing module preprocesses and displays the handwritten digital image, and the handwritten digital recognition module recognizes handwritten digits and outputs the digits and recognized duration by using a trained handwritten digit recognition network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910481876.5A CN110210410B (en) | 2019-06-04 | 2019-06-04 | Handwritten number recognition method based on image characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910481876.5A CN110210410B (en) | 2019-06-04 | 2019-06-04 | Handwritten number recognition method based on image characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110210410A true CN110210410A (en) | 2019-09-06 |
| CN110210410B CN110210410B (en) | 2022-09-23 |
Family
ID=67790739
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910481876.5A Active CN110210410B (en) | 2019-06-04 | 2019-06-04 | Handwritten number recognition method based on image characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210410B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117788629A (en) * | 2024-02-28 | 2024-03-29 | 南京邮电大学 | Image generation method, device and storage medium with style personalization |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599941A (en) * | 2016-12-12 | 2017-04-26 | 西安电子科技大学 | Method for identifying handwritten numbers based on convolutional neural network and support vector machine |
| CN107153810A (en) * | 2016-03-04 | 2017-09-12 | 中国矿业大学 | A kind of Handwritten Numeral Recognition Method and system based on deep learning |
| CN107491729A (en) * | 2017-07-12 | 2017-12-19 | 天津大学 | The Handwritten Digit Recognition method of convolutional neural networks based on cosine similarity activation |
| CN108921007A (en) * | 2018-05-08 | 2018-11-30 | 河海大学常州校区 | A kind of Handwritten Numeral Recognition Method based on SqueezeNet |
-
2019
- 2019-06-04 CN CN201910481876.5A patent/CN110210410B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107153810A (en) * | 2016-03-04 | 2017-09-12 | 中国矿业大学 | A kind of Handwritten Numeral Recognition Method and system based on deep learning |
| CN106599941A (en) * | 2016-12-12 | 2017-04-26 | 西安电子科技大学 | Method for identifying handwritten numbers based on convolutional neural network and support vector machine |
| CN107491729A (en) * | 2017-07-12 | 2017-12-19 | 天津大学 | The Handwritten Digit Recognition method of convolutional neural networks based on cosine similarity activation |
| CN108921007A (en) * | 2018-05-08 | 2018-11-30 | 河海大学常州校区 | A kind of Handwritten Numeral Recognition Method based on SqueezeNet |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117788629A (en) * | 2024-02-28 | 2024-03-29 | 南京邮电大学 | Image generation method, device and storage medium with style personalization |
| CN117788629B (en) * | 2024-02-28 | 2024-05-10 | 南京邮电大学 | Image generation method, device and storage medium with style personalization |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110210410B (en) | 2022-09-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | A deep learning approach to patch-based image inpainting forensics | |
| CN111191736B (en) | Hyperspectral image classification method based on depth feature cross fusion | |
| RU2661750C1 (en) | Symbols recognition with the use of artificial intelligence | |
| CN111753828B (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
| CN108427921A (en) | A kind of face identification method based on convolutional neural networks | |
| US20190294921A1 (en) | Field identification in an image using artificial intelligence | |
| CN111325190B (en) | Expression recognition method and device, computer equipment and readable storage medium | |
| CN113239954A (en) | Attention mechanism-based image semantic segmentation feature fusion method | |
| CN113936195B (en) | Sensitive image recognition model training method and device and electronic equipment | |
| Rao et al. | Exploring deep learning techniques for kannada handwritten character recognition: A boon for digitization | |
| CN114220178A (en) | Signature identification system and method based on channel attention mechanism | |
| CN110599502A (en) | Skin lesion segmentation method based on deep learning | |
| Song et al. | Occluded offline handwritten Chinese character inpainting via generative adversarial network and self-attention mechanism | |
| CN114677377A (en) | Display screen defect detection method, training method, device, equipment and medium | |
| Li et al. | Adaptive mask sampling and manifold to Euclidean subspace learning with distance covariance representation for hyperspectral image classification | |
| CN110414516B (en) | Single Chinese character recognition method based on deep learning | |
| CN115205681A (en) | Lane line segmentation method, lane line segmentation device, and storage medium storing lane line segmentation program | |
| CN110210410B (en) | Handwritten number recognition method based on image characteristics | |
| CN116956214B (en) | Multi-mode fine granularity paper classification method and system based on regularized ensemble learning | |
| CN116630313A (en) | Fluorescence imaging detection system and method thereof | |
| Goud et al. | Text localization and recognition from natural scene images using ai | |
| Zanwar et al. | Hybrid optimization and effectual classification for high recognitions in OCR systems | |
| Xiang et al. | The study of Chinese calligraphy font style based on edge-guided filter and convolutional neural network | |
| CN114913345A (en) | Simplified image feature extraction method based on SIFT algorithm of FPGA | |
| Nguyen et al. | Classification of Road Pavement Defects Based on Convolution Neural Network in Keras |
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 |