CN108062575A - A kind of high similarity graph picture identification and sorting technique - Google Patents
A kind of high similarity graph picture identification and sorting technique Download PDFInfo
- Publication number
- CN108062575A CN108062575A CN201810006119.8A CN201810006119A CN108062575A CN 108062575 A CN108062575 A CN 108062575A CN 201810006119 A CN201810006119 A CN 201810006119A CN 108062575 A CN108062575 A CN 108062575A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- msubsup
- image
- wavelet transformation
- 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
- 238000000034 method Methods 0.000 title claims abstract description 21
- 230000009466 transformation Effects 0.000 claims abstract description 25
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 23
- 230000015556 catabolic process Effects 0.000 claims abstract description 5
- 239000000284 extract Substances 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims abstract description 5
- 238000005457 optimization Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000000638 solvent extraction Methods 0.000 claims abstract description 4
- 238000013528 artificial neural network Methods 0.000 claims abstract description 3
- 238000007781 pre-processing Methods 0.000 claims abstract 2
- 238000000605 extraction Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 230000035945 sensitivity Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims 1
- 238000010606 normalization Methods 0.000 claims 1
- 238000013145 classification model Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000007935 neutral effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to technical field of image processing, especially a kind of high similarity graph picture identification and sorting technique.Method of the present invention is the main object extracted first by partitioning algorithm in image, realizes that image subject is suitably separated with background；It prevents image background from being interfered to main object classification, improves the performance of image Accurate classification；Picture breakdown is different frequency domain subgraphs using wavelet transformation by data preprocessing phase, and extracts each frequency domain Sub-Image Feature；Then using the output of wavelet transformation as the input of convolutional neural networks, optimization is updated network weight parameter by repeatedly training, obtains the highest neural network parameter of classification accuracy, obtains optimal classification model.The combination that the present invention passes through wavelet transformation and convolutional neural networks；It can realize accurate identification, the classification to image；It can be used for the processing of image.
Description
Technical field
The present invention relates to technical field of image processing, especially a kind of high similarity graph picture identification and sorting technique.
Background technology
Renninger L.W et al. think that image simple texture analysis can be searched out the clue of scene classification, and
And simply texture analysis can simplify image recognition processes and calculate.The texture of Benmokhtar et al. blending images and
Color characteristic realizes football scene classification；But this method is classified merely with global characteristics, and separating capacity is limited.LBP is
The local binary pattern of image is calculated by the local binary pattern to " consistent " in image, and to each of image
Kind pattern count carries out statistics with histogram；But this method is general to coloured image classifying quality.People are to remote sensing image in recent years
On textural characteristics using gray level co-occurrence matrixes algorithm, which is between the pixel with certain spatial relation
Joint Distribution, the joint histogram of two pixel grey scales pair can be regarded as, belong to second-order statistics.Gray level co-occurrence matrixes describe ash
Spatial dependence is spent, illustrates a kind of spatial relationship of the pixel grey scale under texture pattern, but this method is not grabbed completely
The firmly characteristics of image of image local gray scale；Therefore it is not ideal enough for image local special characteristic extraction effect.Though wavelet transformation
Image local textural characteristics can be effectively extracted, but it is general to image general classification performance；There is the defects of the following aspects：
(1) it is not notable enough to image local Minutiae extraction；(2) application range is narrow, general to the classifying quality of coloured image；(3)
The general characteristic classification of image is failed to reach satisfactory degree.
The content of the invention
Present invention solves the technical problem that it is to provide a kind of high similarity for merging wavelet transformation and convolutional neural networks
Image identifies and sorting technique.
The present invention solve above-mentioned technical problem technical solution be：
The method is the main object extracted first by partitioning algorithm in image, realizes that image subject is fitted with background
Work as separation；It prevents image background from being interfered to main object classification, improves the performance of image Accurate classification；Data prediction rank
Picture breakdown is different frequency domain subgraphs by Duan Liyong wavelet transformations, and extracts each frequency domain Sub-Image Feature；Then by wavelet transformation
Input of the output as convolutional neural networks, optimization is updated network weight parameter by repeatedly training, acquisition is classified
The highest neural network parameter of accuracy rate, obtains optimal classification model.
Each frequency domain Sub-Image Feature of extraction is to calculate the energy parameter of each frequency domain subgraph as texture descriptor；To not
Generic, different resolution image carries out preliminary feature extraction and determines optimal textural characteristics diversity factor threshold values.
The method will input convolutional neural networks after each subgraph energy feature normalized.
The wavelet transformation is as follows：
Formula (1) is first depending on by image discretization；
Wherein a is scaling function, and b is location parameter；
Then, wavelet transformation carries out fast decoupled, formula such as (2)-(5)；Wherein h and g is respectively the low pass of wavelet transformation
Wave filter and high-pass filter；
Aj(k, l)=∑k1∑k2h(k1-2k)h(k2-2l)fj-1(k1,k2) (2)
The convolutional neural networks convolutional layer update is as follows：
In each layer of convolutional layer of convolutional neural networks, the characteristic pattern of preceding layer and a needs are by learning what is obtained
Convolution kernel carries out convolutional calculation, new characteristic pattern is obtained, using formula (6)：
Wherein Mj represents the input feature vector figure combination of selection, and Ki, j are i-th kind of characteristic pattern of input and the jth kind spy of output
The convolution kernel used in the connection between figure is levied, bj is the biasing corresponding to jth kind characteristic pattern, and f is activation primitive；
Using formula (7), the sensitivity of convolutional neural networks neuron is calculated；Wherein up () represents up-sampling operation；
Kinematic error cost function
In down-sampling layer, new characteristic pattern is obtained by formula (9)；
Carry out Calculation of Sensitivity
It is final to calculate gradient to obtain preferably network weight parameter；
Picture breakdown is different frequency domain subgraphs by wavelet transformation by the present invention, then calculates the energy of each frequency domain subgraph
Parameter is as texture descriptor, and to different classes of, different resolution image carries out preliminary feature extraction and determines that optimal texture is special
Diversity factor threshold values is levied, convolutional neural networks will be inputted after each subgraph energy feature normalized, passes through repeatedly training adjustment net
Network parameter and the classification accuracy for verifying weights reality finally obtain the highest network model of classification accuracy.Become by small echo
It changes and the combination of convolutional neural networks；It can realize accurate identification, the classification to image.
Description of the drawings
The present invention is further described below in conjunction with the accompanying drawings：
Fig. 1 is flow chart of the present invention.
Specific embodiment
As shown in Figure 1, the present invention extracts the main object in image by partitioning algorithm first, image subject and the back of the body are realized
Scape suitably separates, and prevents image background from being interfered to main object classification, improves the performance of image Accurate classification；Data are located in advance
The reason stage carries out picture breakdown using wavelet transformation and extracts each frequency domain Sub-Image Feature；Then using the output of wavelet transformation as god
Input through network is updated network weight parameter optimization by repeatedly training, obtains the highest nerve of classification accuracy
Network parameter obtains optimal classification model.
The core of the present invention mainly includes two parts, is respectively local pretreatment stage of the wavelet transformation to image data,
And the final process of global classification is carried out to image in convolutional neural networks.In wavelet transformation, due to image data category
Two-dimensional digital signal, then first by image discretization, foundation formula (1).
Wherein a is scaling function, and b is location parameter.
Wavelet transformation carries out fast decoupled, formula such as (2)-(5).Wherein h and g is respectively the low-pass filter of wavelet transformation
And high-pass filter.
Aj(k, l)=∑k1∑k2h(k1-2k)h(k2-2l)fj-1(k1,k2) (2)
Second technological core is the right value update of convolutional neural networks, the most termination of this some effects image data classification
Fruit.Higher classification accuracy in order to obtain, this technology have carried out many experiments, convolutional layer and pond of the emphasis in neutral net
Change the right value update part of layer.
In each layer of convolutional layer of convolutional neural networks, the characteristic pattern of preceding layer and a needs are by learning what is obtained
Convolution kernel carries out convolutional calculation, new characteristic pattern is can obtain, such as formula (6).
Wherein MjRepresent the input feature vector figure combination of selection, Ki,jIt is i-th kind of characteristic pattern of input and the jth kind spy of output
Levy the convolution kernel used in the connection between figure, bjIt is the biasing corresponding to jth kind characteristic pattern, f is activation primitive.
The sensitivity of convolutional neural networks neuron is calculated, such as formula (7).Wherein up () represents up-sampling operation.
Error cost function
And in down-sampling layer, new characteristic pattern can be obtained by formula (9).
Calculation of Sensitivity
It is final to calculate gradient to obtain preferably network weight parameter.
After right value update is carried out to the convolutional layer of neutral net and sample level, also need to set in detail excellent to other layers
Change parameter, each layer parameter of convolutional neural networks such as table 2 that this technology uses.
Present networks include altogether 5 convolutional layers, 3 pond layers, 3 full articulamentums, 1 softmax classification layer, 2 LRN
Layer and two stochastical sampling layers find this parameter hypograph classification accuracy height by multiple test.
Each layer parameter of 1. convolutional neural networks of table
Each meaning of parameters of the present invention is as shown in the table：
2. parameter concrete meaning of table
Claims (5)
1. a kind of high similarity graph picture identification and sorting technique, it is characterised in that：The method is to pass through partitioning algorithm first
The main object in image is extracted, realizes that image subject is suitably separated with background；Prevent image background from being made to main object classification
Into interference, the performance of raising image Accurate classification；Picture breakdown is different frequencies using wavelet transformation by data preprocessing phase
Domain subgraph, and extract each frequency domain Sub-Image Feature；Then using the output of wavelet transformation as the input of convolutional neural networks, by more
Secondary training is updated network weight parameter optimization, obtains the highest neural network parameter of classification accuracy, is most preferably divided
Class model.
2. according to the method described in claim 1, it is characterized in that：Each frequency domain Sub-Image Feature of the extraction is to calculate each frequency domain
The energy parameter of subgraph is as texture descriptor；Preliminary feature extraction is carried out to different classes of, different resolution image and is determined
Optimal textural characteristics diversity factor threshold values.
3. according to the method described in claim 2, it is characterized in that：The method is after wavelet transformation by each subgraph energy
Convolutional neural networks are inputted after feature normalization processing.
4. according to the method described in claim 3, it is characterized in that：After image data input network, the pond of convolutional neural networks
Change the right value update of layer and convolutional layer, obtain preferably sorting parameter.
According to claim 2, the wavelet transformation is as follows：
Formula (1) is first depending on by image discretization；
<mrow>
<msub>
<mi>&psi;</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mi>a</mi>
<mn>0</mn>
<mrow>
<mo>-</mo>
<mfrac>
<mi>j</mi>
<mn>2</mn>
</mfrac>
</mrow>
</msubsup>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>a</mi>
<mn>0</mn>
<mrow>
<mo>-</mo>
<mi>j</mi>
</mrow>
</msubsup>
<mi>x</mi>
<mo>-</mo>
<msub>
<mi>kb</mi>
<mn>0</mn>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein a is scaling function, and b is location parameter；ψj,kIt is the output of wavelet transformation discretization；
Then, wavelet transformation carries out fast decoupled, formula such as (2)-(5)；Wherein h and g is respectively the low-pass filtering of wavelet transformation
Device and high-pass filter；
Aj(k, l)=∑k1∑k2h(k1-2k)h(k2-2l)fj-1(k1,k2) (2)
<mrow>
<msubsup>
<mi>D</mi>
<mi>J</mi>
<mn>1</mn>
</msubsup>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>,</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>1</mn>
</mrow>
</msub>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>2</mn>
</mrow>
</msub>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>-</mo>
<mn>2</mn>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mi>h</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>2</mn>
<mo>-</mo>
<mn>2</mn>
<mi>l</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>f</mi>
<mrow>
<mi>j</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>,</mo>
<mi>k</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>D</mi>
<mi>j</mi>
<mn>2</mn>
</msubsup>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>,</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>1</mn>
</mrow>
</msub>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>2</mn>
</mrow>
</msub>
<mi>h</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>-</mo>
<mi>k</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>2</mn>
<mo>-</mo>
<mn>2</mn>
<mi>l</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>f</mi>
<mrow>
<mi>j</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>,</mo>
<mi>k</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>D</mi>
<mi>j</mi>
<mn>3</mn>
</msubsup>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>,</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>1</mn>
</mrow>
</msub>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mn>2</mn>
</mrow>
</msub>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>-</mo>
<mi>k</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>2</mn>
<mo>-</mo>
<mn>2</mn>
<mi>l</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>f</mi>
<mrow>
<mi>j</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mn>1</mn>
<mo>,</mo>
<mi>k</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
5. according to the method described in claim 4, it is characterized in that：
The convolutional neural networks convolutional layer update is as follows：
In each layer of convolutional layer of convolutional neural networks, the characteristic pattern of preceding layer and a convolution needed by learning to obtain
Core carries out convolutional calculation, new characteristic pattern is obtained, using formula (6)；
<mrow>
<msubsup>
<mi>x</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>=</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>&Element;</mo>
<mi>M</mi>
<mi>j</mi>
</mrow>
</msub>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mrow>
<mi>l</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>*</mo>
<msubsup>
<mi>k</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>l</mi>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>b</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein Mj represents the input feature vector figure combination of selection, and Ki, j are i-th kind of characteristic pattern of input and the jth kind characteristic pattern of output
Between connection used in convolution kernel, bj is the biasing corresponding to jth kind characteristic pattern, and f is activation primitive；
Using formula (7), the sensitivity of convolutional neural networks neuron is calculated.Wherein up () represents up-sampling operation；
Error cost function
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>E</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>K</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mi>l</mi>
</msubsup>
</mrow>
</mfrac>
<mo>=</mo>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
</mrow>
</msub>
<msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&delta;</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
</mrow>
</msub>
<msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>P</mi>
<mi>i</mi>
<mrow>
<mi>l</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
</mrow>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
</mrow>
In down-sampling layer, new characteristic pattern is obtained by formula (9)；
<mrow>
<msubsup>
<mi>x</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>=</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&beta;</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mi>d</mi>
<mi>o</mi>
<mi>w</mi>
<mi>n</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>j</mi>
<mrow>
<mi>l</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mi>b</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>9</mn>
<mo>)</mo>
</mrow>
</mrow>
Carry out Calculation of Sensitivity
It is final to calculate gradient to obtain preferably network weight parameter；
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>E</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<msub>
<mi>b</mi>
<mi>j</mi>
</msub>
</mrow>
</mfrac>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
</mrow>
</msub>
<msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&delta;</mi>
<mi>j</mi>
<mi>l</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
</mrow>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>11</mn>
<mo>)</mo>
</mrow>
</mrow>
The high similar image identification for combining wavelet transformation and convolutional neural networks overcomes image data particulate with sorting technique
The shortcomings that texture feature extraction is insufficient is spent, the generalization ability to image data of classifying is improved, significantly improves image overall point
Class effect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810006119.8A CN108062575A (en) | 2018-01-03 | 2018-01-03 | A kind of high similarity graph picture identification and sorting technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810006119.8A CN108062575A (en) | 2018-01-03 | 2018-01-03 | A kind of high similarity graph picture identification and sorting technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108062575A true CN108062575A (en) | 2018-05-22 |
Family
ID=62141119
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810006119.8A Pending CN108062575A (en) | 2018-01-03 | 2018-01-03 | A kind of high similarity graph picture identification and sorting technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108062575A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034256A (en) * | 2018-08-02 | 2018-12-18 | 燕山大学 | A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features |
| CN109145723A (en) * | 2018-07-09 | 2019-01-04 | 长江大学 | A kind of seal recognition methods, system, terminal installation and storage medium |
| CN111079845A (en) * | 2019-12-20 | 2020-04-28 | 西安交通大学 | Classification and identification method for coal gangue and coal in coal |
| CN111382764A (en) * | 2018-12-29 | 2020-07-07 | 北大方正集团有限公司 | Neural network model establishing method and device and computer readable storage medium |
| TWI703961B (en) * | 2018-07-06 | 2020-09-11 | 台達電子工業股份有限公司 | Oral image analysis system and method |
| CN111767927A (en) * | 2020-06-03 | 2020-10-13 | 上海交通大学 | Lightweight license plate recognition method and system based on full convolution network |
| CN113362096A (en) * | 2020-03-04 | 2021-09-07 | 驰众信息技术（上海）有限公司 | Frame advertisement image matching method based on deep learning |
| CN114067392A (en) * | 2021-11-06 | 2022-02-18 | 北京九州安华信息安全技术有限公司 | Image identification method and device based on Riesz wavelet transformation |
| CN114169430A (en) * | 2021-12-06 | 2022-03-11 | 深圳依时货拉拉科技有限公司 | Method and device for identifying copied image, storage medium and computer equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971106A (en) * | 2014-05-27 | 2014-08-06 | 深圳市赛为智能股份有限公司 | Multi-view human facial image gender identification method and device |
| CN104751186A (en) * | 2015-04-10 | 2015-07-01 | 山东师范大学 | Iris image quality classification method based on BP (back propagation) network and wavelet transformation |
| CN107239751A (en) * | 2017-05-22 | 2017-10-10 | 西安电子科技大学 | High Resolution SAR image classification method based on the full convolutional network of non-down sampling contourlet |
| CN107368791A (en) * | 2017-06-29 | 2017-11-21 | 广东欧珀移动通信有限公司 | Living iris detection method and Related product |
| CN107463948A (en) * | 2017-07-13 | 2017-12-12 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels multiple features fusion network |
-
2018
- 2018-01-03 CN CN201810006119.8A patent/CN108062575A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971106A (en) * | 2014-05-27 | 2014-08-06 | 深圳市赛为智能股份有限公司 | Multi-view human facial image gender identification method and device |
| CN104751186A (en) * | 2015-04-10 | 2015-07-01 | 山东师范大学 | Iris image quality classification method based on BP (back propagation) network and wavelet transformation |
| CN107239751A (en) * | 2017-05-22 | 2017-10-10 | 西安电子科技大学 | High Resolution SAR image classification method based on the full convolutional network of non-down sampling contourlet |
| CN107368791A (en) * | 2017-06-29 | 2017-11-21 | 广东欧珀移动通信有限公司 | Living iris detection method and Related product |
| CN107463948A (en) * | 2017-07-13 | 2017-12-12 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels multiple features fusion network |
Non-Patent Citations (1)
| Title |
|---|
| 高强 等: "基于卷积神经网络的绝缘子故障识别算法研究", 《电测与仪表》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI703961B (en) * | 2018-07-06 | 2020-09-11 | 台達電子工業股份有限公司 | Oral image analysis system and method |
| CN109145723A (en) * | 2018-07-09 | 2019-01-04 | 长江大学 | A kind of seal recognition methods, system, terminal installation and storage medium |
| CN109034256A (en) * | 2018-08-02 | 2018-12-18 | 燕山大学 | A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features |
| CN111382764A (en) * | 2018-12-29 | 2020-07-07 | 北大方正集团有限公司 | Neural network model establishing method and device and computer readable storage medium |
| CN111382764B (en) * | 2018-12-29 | 2024-02-13 | 新方正控股发展有限责任公司 | Neural network model building method and device for face recognition or gesture recognition and computer readable storage medium |
| CN111079845A (en) * | 2019-12-20 | 2020-04-28 | 西安交通大学 | Classification and identification method for coal gangue and coal in coal |
| CN113362096A (en) * | 2020-03-04 | 2021-09-07 | 驰众信息技术（上海）有限公司 | Frame advertisement image matching method based on deep learning |
| CN111767927A (en) * | 2020-06-03 | 2020-10-13 | 上海交通大学 | Lightweight license plate recognition method and system based on full convolution network |
| CN114067392A (en) * | 2021-11-06 | 2022-02-18 | 北京九州安华信息安全技术有限公司 | Image identification method and device based on Riesz wavelet transformation |
| CN114169430A (en) * | 2021-12-06 | 2022-03-11 | 深圳依时货拉拉科技有限公司 | Method and device for identifying copied image, storage medium and computer equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108062575A (en) | A kind of high similarity graph picture identification and sorting technique | |
| Mei et al. | Hyperspectral image classification using group-aware hierarchical transformer | |
| CN108460342A (en) | Hyperspectral image classification method based on convolution net and Recognition with Recurrent Neural Network | |
| CN103366180A (en) | Cell image segmentation method based on automatic feature learning | |
| Liu et al. | The classification and denoising of image noise based on deep neural networks | |
| CN103927531A (en) | Human face recognition method based on local binary value and PSO BP neural network | |
| CN110647977B (en) | Method for optimizing Tiny-YOLO network for detecting ship target on satellite | |
| CN111080678A (en) | Multi-temporal SAR image change detection method based on deep learning | |
| CN114155371A (en) | Semantic segmentation method based on channel attention and pyramid convolution fusion | |
| CN112950780A (en) | Intelligent network map generation method and system based on remote sensing image | |
| Li et al. | Automatic Recognition of Rock Images Based on Convolutional Neural Network and Discrete Cosine Transform. | |
| Gong et al. | Multispectral and SAR image fusion for multi-scale decomposition based on least squares optimization rolling guidance filtering | |
| CN114937202A (en) | Double-current Swin transform remote sensing scene classification method | |
| Liu et al. | High-resolution remote sensing image information extraction and target recognition based on multiple information fusion | |
| Ichimura | Spatial frequency loss for learning convolutional autoencoders | |
| CN111275732B (en) | Foreground object image segmentation method based on depth convolution neural network | |
| CN112733736A (en) | Class imbalance hyperspectral image classification method based on enhanced oversampling | |
| Rezvani et al. | Single image denoising via a new lightweight learning-based model | |
| CN112818840B (en) | Unmanned aerial vehicle online detection system and method | |
| CN116594061B (en) | Seismic data denoising method based on multi-scale U-shaped attention network | |
| Yuan | Remote sensing image classification methods based on CNN: Challenge and trends | |
| CN117953310A (en) | Remote sensing multi-mode image classification method based on continuous scale feature network | |
| CN112418262A (en) | Vehicle re-identification method, client and system | |
| Ma et al. | An improved 3D-2D convolutional neural network based on feature optimization for hyperspectral image classification | |
| Premk et al. | Automatic latent fingerprint segmentation using convolutional neural networks |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180522 |