CN110347851A - Image search method and system based on convolutional neural networks - Google Patents
Image search method and system based on convolutional neural networks Download PDFInfo
- Publication number
- CN110347851A CN110347851A CN201910463906.XA CN201910463906A CN110347851A CN 110347851 A CN110347851 A CN 110347851A CN 201910463906 A CN201910463906 A CN 201910463906A CN 110347851 A CN110347851 A CN 110347851A
- Authority
- CN
- China
- Prior art keywords
- image
- data base
- image data
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Library & Information Science (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides image search methods and system based on convolutional neural networks, comprising: establishes image data base, batch carries out image preprocessing；Utilize Inception-v2 image convolution neural network model of the image data base training based on GoogLeNet；Image to be detected is acquired, and it is pre-processed, characteristic vector pickup is carried out to the image image to be retrieved in image data base with image convolution neural network model；Euclidean distance is carried out to each of characteristics of image and image data base to be retrieved feature vector and calculates similarity, it sorts to database images according to the ascending sequence of Euclidean distance, corresponding top n image feedback is completed into image retrieval procedure to user in order.The invention avoids the factors such as image scaling, rotation, translation to influence caused by search result, and the model training time is greatly improved, and efficiently solve causes character representation not comprehensive because figure texture is complicated, so as to cause the not high problem of retrieval precision.
Description
Technical field
The present invention relates to computer visual image searching fields, and in particular to the image retrieval side based on convolutional neural networks
Method and system.
Background technique
In recent years, universal with smart machine with the development of internet, outburst is presented in the picture number stored on network
Formula increases, and has caused image data overload problem, how effectively to have retrieved the interested figure of user from extensive large nuber of images
Research hotspot as becoming present image processing with analysis field research.Since traditional text based retrieval manually marks cost
It is high, there are user's subjectivity, image retrieval technologies are developing progressively by text based retrieval as content-based retrieval, are based on
The image retrieval process of content is to calculate inquiry by extracting the images bottom visual signatures such as color of image, texture, shape
Image returns to image identical or the most similar with query image at a distance from test image feature.But low-level image feature and height
There are semantic gap between layer semanteme, the retrieval that cannot accurately reflect user is intended to.As social information's technology high-speed develops
And the arrival of big data era, not only amount of images increases, but also picture material also becomes complicated multiplicity, traditional based on interior
The retrieval technique of appearance has been unable to satisfy the demand of user, and search result is easy to be influenced by image scaling, translation, rotation, scheme
The low-level features of picture are relative complex, and when search fee and accuracy rate is lower.
With the fast development of computer hardware and the exponential increase of data volume, depth learning technology especially convolution mind
It can learn the distribution to data directly from image data through network, in image classification, image segmentation, target detection, face
Success is achieved in the visual tasks such as identification, accuracy rate has surmounted the mankind.Utilize the powerful figure of convolutional neural networks
As ability in feature extraction, convolutional neural networks model is established, to improve image retrieval performance.
Summary of the invention
The technical problem to be solved in the present invention is that the skill not high, computationally intensive for above-mentioned current image retrieval precision
Art problem provides image search method and system based on convolutional neural networks and solves above-mentioned technological deficiency.
Image search method based on convolutional neural networks, comprising:
Step 1 establishes image data base, and batch carries out image preprocessing；
Step 2, Inception-v2 image volume of the image data base training based on GoogLeNet established using step 1
Product neural network model；
Step 3, acquisition image to be detected, and it is pre-processed, the image convolution neural network mould established with step 2
Type in the image data base established in step 1 image and the image to be retrieved carry out characteristic vector pickup respectively；
Step 4, to each of the characteristics of image of image to be detected and image data base feature vector carry out it is European away from
It from similarity is calculated, sorts to database images according to the ascending sequence of Euclidean distance, in order by corresponding preceding n figure
As feeding back to user, image retrieval procedure is completed.
Further, step 1 specifically includes:
Step 1.1 collects image, prepares the image data base for retrieval, carries out manual classification to the image in library, and
It is renamed according to place classification；
All images are uniformly zoomed to fixed size by step 1.2；
Image in image data base is normalized and zero averaging step 1.3, to eliminate to whole image data
The mean value of the image data.
Further, specific using the database training convolutional neural networks model established in step 1 described in step 2
Method is: image data is produced after by Inception-v2 neural network model using the full articulamentum of oneself creation
Raw output, the softmax layer of Inception-v2 model and full articulamentum are removed, plus newly complete behind pool_3 layers
The number of articulamentum, number of network node 2048, output node is the other number of data set picture category, is migrated to model
It needs to define training parameter when study.
Further, the specific method of step 3 is: the figure established with the convolutional neural networks model step 1 that step 2 is established
As the image zooming-out feature vector in database, construction feature library, to the image feature vector to be retrieved acquired in step 1, note
The feature vector of image to be retrieved is X, and the feature vector of the image in image data base is Yi, i=1,2 ..., and N, N are from figure
As the number for the feature vector extracted in database.
Further, the specific method of step 4 is: the feature vector extracted according to step 3 calculates the spy of image to be retrieved
The Euclidean distance between the vector X and feature vector Yi of the image in image data base is levied, the image in image data base is pressed
Euclidean distance ascending arrangement after sequence, extracts a few width images corresponding with minimum euclidean distance, as to be retrieved
The highest preceding n image of similarity.
Image search method system based on convolutional neural networks, comprising: processor and storage equipment；The processor adds
It carries and executes the instruction in the storage equipment and data for realizing any one image retrieval based on convolutional neural networks
Method.
Compared with prior art, the invention has the advantages that:
1, the invention proposes the image search methods based on convolutional neural networks, according to the self study of convolutional neural networks
Ability avoids the influence caused by search result of the factors such as image scaling, rotation, translation, adaptable.
2, existing network model is finely adjusted using lesser dedicated data set, improves the accuracy of model；Benefit
Accelerated with GPU, the model training time is greatly improved.
3, using convolutional neural networks model extraction characteristics of image, efficiently solving leads to mark sheet because figure texture is complicated
Show not comprehensively, so as to cause the not high problem of retrieval precision.
Detailed description of the invention
With reference to the accompanying drawing and specific embodiment elaborates to the present invention, in attached drawing:
Fig. 1 is that the present invention is based on the image search method flow charts of convolutional neural networks；
Fig. 2 is that the present invention is based on the convolutional neural networks structure charts of the image search method of convolutional neural networks；
Fig. 3 is that the present invention is based on the correspondence retrieval precision curve graphs of the image search method of convolutional neural networks.
Specific embodiment
In order to understand the technical features, objects and effects of the invention more easily, now compares attached drawing and this hair is described in detail
Bright specific embodiment.
Image search method based on convolutional neural networks, as shown in Figure 1, comprising:
Step 1 establishes image data base, and batch carries out image preprocessing, specifically includes:
Step 1.1 collects image, prepares the image data base for retrieval, carries out manual classification to the image in library, and
It is renamed according to place classification；
Image in all image data bases is divided into training set and test set part by step 1.2, and training set is used to train
Retrieval model, test set are used to the quality of testing model.
All images are uniformly zoomed to fixed size by step 1.3, in the structure of entire convolutional neural networks, convolution
Layer and pond layer do not require picture size, but it is fixed size that full articulamentum, which just needs picture, because of full connection
The weight matrix of layer is the matrix of fixed size after training.
Image in image data base is normalized and zero averaging step 1.4, to eliminate to whole image data
The mean value of the image data.
Step 2, as shown in Fig. 2, the image data base training established using step 1 is based on GoogLeNet's
Inception-v2 image convolution neural network model；
GoogLeNet is the champion of the match of ImageNet in 2014, and model has 22 layers, the core of Inception-v2 model
It is batch normalization and convolution nuclear subsitution, the data distribution that batch normalization solves intrinsic nerve member changed is asked
Topic, so that each layer of output, which all standardizes, has arrived N (0,1), while allowing relatively high learning rate；Replace part dropout
Function, the convolution nuclear subsitution of Inception-v2 be using two 3x3 convolution kernel replace a 5x5 convolution kernel, parameter
It reduces but receptive field is identical, network burden can be mitigated.
It is using the database training convolutional neural networks model specific method established in step 1 described in step 2: will
Image data generates output using the full articulamentum of oneself creation after by Inception-v2 neural network model, because
For the feature of transfer learning training, the neural network weight of the full articulamentum oneself added is only trained in the training of network, can't
Influence the parameter of Inception-v2.The softmax layer of Inception-v2 model and full articulamentum are removed, at pool_3 layers
Behind plus new full articulamentum, number of network node 2048, the number of output node is the other number of data set picture category,
It needs to define training parameter when carrying out transfer learning to model.
It is right in this way to make neural network have the function of nonlinear operation it is necessary to introduce nonlinear activation primitive
Approaching for Any Nonlinear Function can be realized by neural network model, so that neural network model be made to can extend to
In non-linear volume model.Activation primitive selects ReLU, can preferably solve the problems, such as that neural network gradient disappears.
The purpose of model training is desirable to loss function to be reduced as far as possible, but is not intended to allow convolution in true application
Mould neural network is completely fitted the behavior of training data, and is desirable to convolutional neural networks training and obtains the inherence of learning data
Rule.L2 regularization is that L2 norm penalty term is added in restriction on the parameters, and formula is as follows:
Moving average model is the method that model can be allowed more healthy and stronger, is calculated in training network using gradient decline
When method, in most cases network can be made to have more outstanding table in test data set using moving average model
It is existing.In the training process of neural network, system needs to safeguard a shadow variable, and calculation formula is shown below:
Vs=dr*Vs+(1-dr) * V,
VsFor the shadow variable of system maintenance, drFor attenuation rate, V is variable to be updated.From formula as can be seen that drIt determines
Model modification speed, drCloser to 1, model is more stable.
The network of transfer learning selects cross entropy cost function as the loss function of the model.
Common cost function selection is secondary cost function, but the shortcomings that secondary cost function is initialization
When weight and biasing have a long way to go with final weight and biasing, the speed of the update of parameter will be difficult to improve.Intersect
The formula of entropy cost function:
Wherein: n is the sum of training data, and x is all training inputs, and y is corresponding mark output, and a is neural network
Output.Cross entropy cost function does not have the problem of pace of learning downslide, this characteristic makes its performance better than secondary cost
Function.
In the instruction of neural network, weight and the speed for biasing every single-step iteration update are directly controlled by learning rate, are being learned
The setting of habit rate it is bigger when, then weight and biasing are likely to occur in the case where swinging back and forth between optimal value, if
Parameter is too small, although can guarantee convergence, the efficiency of training be will be greatly reduced, so learning rate can neither be too big, also not
It can be too small.In order to solve this problem, I trains network using the learning rate of decaying, selects the learning rate pair of exponential decay
Network is trained.
The performance of neural network can be improved with the increase of frequency of training for theoretically, but if in limited number
According to upper unlimited training neural network, a possibility that over-fitting of neural network can be improved, thus general in unknown data
Magnificent is less able.
Step 3, acquisition image to be detected, and it is pre-processed, the image convolution neural network mould established with step 2
Type carries out characteristic vector pickup to the image image to be retrieved in the image data base established in step 1, and specific method is:
With step 2 establish convolutional neural networks model step 1 establish image data base in image zooming-out feature to
Amount, construction feature library remember that the feature vector of image to be retrieved is X, figure to the image feature vector to be retrieved acquired in step 1
As the image in database feature vector be Yi, i=1,2 ..., N, N are the feature vector extracted from image data base
Number.
Step 4 carries out Euclidean distance meter to each of characteristics of image and image data base to be retrieved feature vector
Similarity is calculated, is sorted to database images according to the ascending sequence of Euclidean distance, it is in order that corresponding top n image is anti-
Feed user, completes image retrieval procedure, specific method is:
According to the feature vector that step 3 is extracted, the image in the feature vector, X and image data base of image to be retrieved is calculated
Feature vector Yi between Euclidean distance, by the image in image data base press the ascending arrangement of Euclidean distance, after sequence,
Extract a few width images corresponding with minimum euclidean distance, the highest top n image of the similarity as to be retrieved.
The present invention is based on the correspondence retrieval precision curve graph of the image search method of convolutional neural networks is as shown in Figure 3.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form, all of these belong to the protection of the present invention.
Claims (6)
1. the image search method based on convolutional neural networks characterized by comprising
Step 1 establishes image data base, and batch carries out image preprocessing；
The Inception-v2 image convolution mind of step 2, the image data base training established using step 1 based on GoogLeNet
Through network model；
Step 3, acquisition image to be detected, and it is pre-processed, the image convolution neural network model pair established with step 2
The image in image data base and the image to be retrieved established in step 1 carry out characteristic vector pickup respectively；
Step 4 carries out Euclidean distance meter to each of the characteristics of image of image to be detected and image data base feature vector
Similarity is calculated, is sorted to database images according to the ascending sequence of Euclidean distance, it is in order that corresponding preceding n image is anti-
Feed user, completes image retrieval procedure.
2. the image search method according to claim 1 based on convolutional neural networks, which is characterized in that step 1 is specific
Include:
Step 1.1 collects image, prepares the image data base for retrieval, to the image progress manual classification in library, and according to
Place classification is renamed；
All images are uniformly zoomed to fixed size by step 1.2；
Image in image data base is normalized and zero averaging step 1.3, to eliminate the figure to whole image data
As the mean value of data.
3. the image search method according to claim 1 based on convolutional neural networks, which is characterized in that institute in step 2
What is stated is using the database training convolutional neural networks model specific method established in step 1: image data is being passed through
Output is generated using the full articulamentum of oneself creation after Inception-v2 neural network model, by Inception-v2 mould
The softmax layer and full articulamentum of type remove, and plus new full articulamentum behind pool_3 layers, number of network node is
2048, the number of output node is the other number of data set picture category, needs to define when carrying out transfer learning to model
Training parameter.
4. the image search method according to claim 1 based on convolutional neural networks, which is characterized in that the tool of step 3
Body method is: with step 2 establish convolutional neural networks model step 1 establish image data base in image zooming-out feature to
Amount, construction feature library remember that the feature vector of image to be retrieved is X, figure to the image feature vector to be retrieved acquired in step 1
As the image in database feature vector be Yi, i=1,2 ..., N, N are the feature vector extracted from image data base
Number.
5. the image search method according to claim 1 based on convolutional neural networks, which is characterized in that the tool of step 4
Body method is: the feature vector extracted according to step 3 calculates the figure in the feature vector, X and image data base of image to be retrieved
Image in image data base is pressed the ascending arrangement of Euclidean distance, sequence by the Euclidean distance between the feature vector Yi of picture
Afterwards, a few width images corresponding with minimum euclidean distance, the highest preceding n image of the similarity as to be retrieved are extracted.
6. the image search method system based on convolutional neural networks characterized by comprising processor and storage equipment；Institute
Processor is stated to load and execute the instruction in the storage equipment and data for realizing any one described in Claims 1 to 5
Image search method of the kind based on convolutional neural networks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910463906.XA CN110347851A (en) | 2019-05-30 | 2019-05-30 | Image search method and system based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910463906.XA CN110347851A (en) | 2019-05-30 | 2019-05-30 | Image search method and system based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110347851A true CN110347851A (en) | 2019-10-18 |
Family
ID=68174477
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910463906.XA Pending CN110347851A (en) | 2019-05-30 | 2019-05-30 | Image search method and system based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110347851A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111694977A (en) * | 2020-06-12 | 2020-09-22 | 桂林电子科技大学 | Vehicle image retrieval method based on data enhancement |
| CN111709292A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Compressor vibration fault detection method based on recursive graph and deep convolutional network |
| CN111797921A (en) * | 2020-07-02 | 2020-10-20 | 中国工商银行股份有限公司 | Image data comparison method and device |
| CN112507158A (en) * | 2020-12-18 | 2021-03-16 | 北京百度网讯科技有限公司 | Image processing method and device |
| CN113780304A (en) * | 2021-08-09 | 2021-12-10 | 国网安徽省电力有限公司检修分公司 | Transformer substation equipment image retrieval method and system based on neural network |
| CN114429638A (en) * | 2022-04-06 | 2022-05-03 | 四川省大数据中心 | Construction drawing examination management system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256246A (en) * | 2017-06-06 | 2017-10-17 | 西安工程大学 | PRINTED FABRIC image search method based on convolutional neural networks |
| CN108681708A (en) * | 2018-05-16 | 2018-10-19 | 福州精益精科技有限责任公司 | A kind of vena metacarpea image-recognizing method, device and storage medium based on Inception neural network models |
| CN109190666A (en) * | 2018-07-30 | 2019-01-11 | 北京信息科技大学 | Flowers image classification method based on improved deep neural network |
| WO2019074316A1 (en) * | 2017-10-12 | 2019-04-18 | 주식회사 웰시스템코리아 | Convolutional artificial neural network-based recognition system in which registration, search, and reproduction of image and video are divided between and performed by mobile device and server |
-
2019
- 2019-05-30 CN CN201910463906.XA patent/CN110347851A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256246A (en) * | 2017-06-06 | 2017-10-17 | 西安工程大学 | PRINTED FABRIC image search method based on convolutional neural networks |
| WO2019074316A1 (en) * | 2017-10-12 | 2019-04-18 | 주식회사 웰시스템코리아 | Convolutional artificial neural network-based recognition system in which registration, search, and reproduction of image and video are divided between and performed by mobile device and server |
| CN108681708A (en) * | 2018-05-16 | 2018-10-19 | 福州精益精科技有限责任公司 | A kind of vena metacarpea image-recognizing method, device and storage medium based on Inception neural network models |
| CN109190666A (en) * | 2018-07-30 | 2019-01-11 | 北京信息科技大学 | Flowers image classification method based on improved deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| 刘吉: "深度学习在医学图像识别中的应用研究", 《青岛大学学报(自然科学版)》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111709292A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Compressor vibration fault detection method based on recursive graph and deep convolutional network |
| CN111709292B (en) * | 2020-05-18 | 2023-05-26 | 杭州电子科技大学 | Compressor vibration fault detection method based on recursion diagram and deep convolution network |
| CN111694977A (en) * | 2020-06-12 | 2020-09-22 | 桂林电子科技大学 | Vehicle image retrieval method based on data enhancement |
| CN111797921A (en) * | 2020-07-02 | 2020-10-20 | 中国工商银行股份有限公司 | Image data comparison method and device |
| CN112507158A (en) * | 2020-12-18 | 2021-03-16 | 北京百度网讯科技有限公司 | Image processing method and device |
| CN113780304A (en) * | 2021-08-09 | 2021-12-10 | 国网安徽省电力有限公司检修分公司 | Transformer substation equipment image retrieval method and system based on neural network |
| CN113780304B (en) * | 2021-08-09 | 2023-12-05 | 国网安徽省电力有限公司超高压分公司 | Substation equipment image retrieval method and system based on neural network |
| CN114429638A (en) * | 2022-04-06 | 2022-05-03 | 四川省大数据中心 | Construction drawing examination management system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110347851A (en) | Image search method and system based on convolutional neural networks | |
| Zhao et al. | ApLeaf: An efficient android-based plant leaf identification system | |
| CN108038122B (en) | Trademark image retrieval method | |
| CN106021406B (en) | A kind of online mask method of iterative image of data-driven | |
| CN105574063A (en) | Image retrieval method based on visual saliency | |
| CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
| CN106021603A (en) | Garment image retrieval method based on segmentation and feature matching | |
| CN108229503A (en) | A kind of feature extracting method for clothes photo | |
| CN105931241B (en) | A kind of automatic marking method of natural scene image | |
| CN108710894A (en) | A kind of Active Learning mask method and device based on cluster representative point | |
| CN110609917B (en) | Image retrieval method and system based on convolutional neural network and significance detection | |
| CN107239777A (en) | A kind of tableware detection and recognition methods based on various visual angles graph model | |
| CN109583498A (en) | A kind of fashion compatibility prediction technique based on low-rank regularization feature enhancing characterization | |
| CN109213853A (en) | A kind of Chinese community's question and answer cross-module state search method based on CCA algorithm | |
| CN113705579A (en) | Automatic image annotation method driven by visual saliency | |
| CN107918761A (en) | A kind of single sample face recognition method based on multiple manifold kernel discriminant analysis | |
| CN109509191A (en) | A kind of saliency object detection method and system | |
| CN107016371A (en) | UAV Landing Geomorphological Classification method based on improved depth confidence network | |
| CN106844785A (en) | Saliency segmentation-based content-based image retrieval method | |
| Fang et al. | Identification of apple leaf diseases based on convolutional neural network | |
| CN110298392A (en) | A kind of semisupervised classification method that label constraint learns from the more hypergraphs of weight | |
| CN106874942A (en) | A kind of object module fast construction method semantic based on regular expression | |
| CN103399863B (en) | Image search method based on the poor characteristic bag of edge direction | |
| Qi et al. | Exploring illumination robust descriptors for human epithelial type 2 cell classification | |
| Rangkuti et al. | Batik image retrieval based on similarity of shape and texture characteristics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191018 |
|
| RJ01 | Rejection of invention patent application after publication |