CN103984959A - Data-driven and task-driven image classification method - Google Patents
Data-driven and task-driven image classification method Download PDFInfo
- Publication number
- CN103984959A CN103984959A CN201410224860.3A CN201410224860A CN103984959A CN 103984959 A CN103984959 A CN 103984959A CN 201410224860 A CN201410224860 A CN 201410224860A CN 103984959 A CN103984959 A CN 103984959A
- Authority
- CN
- China
- Prior art keywords
- training
- image
- convolutional neural
- neural networks
- data
- 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 47
- 238000012549 training Methods 0.000 claims abstract description 78
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 62
- 238000012360 testing method Methods 0.000 claims abstract description 30
- 238000004458 analytical method Methods 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 9
- 230000004044 response Effects 0.000 claims description 9
- 238000013461 design Methods 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a data-driven and task-driven image classification method. The data-driven and task-driven classification method comprises the steps that a convolutional neural network structure is designed according to the scale of data sets and image content; a convolutional neural network model is trained through the given classified data sets; feature expression is extracted from training set images through a trained convolution neural network; images to be tested are input into the trained convolutional neural network and are classified. The data-driven and task-driven image classification method is based on nonlinear convolution feature learning, and the model can be adapted to the data sets through a date driving mode, so that the specific data set can be better described; errors of K-nearest neighbors can be directly optimized through a task-driving mode, and therefore a better performance can be obtained with respect to a K-nearest neighbor task; efficient training can be conducted through a GPU in the training stage, and efficient K-nearest neighbor image classification can be achieved just through a CPU in the testing stage; in this way, the data-driven and task-driven image classification method is quite suitable for a large-scale image classification task, a retrieval task and the like.
Description
Technical field
The present invention relates to Image Classfication Technology field in computer vision, particularly a kind of image classification method based on data and task-driven.
Background technology
Images Classification is the most basic one of the studying a question of computer vision, and its problem that will solve is exactly disconnected certain type objects that wherein whether comprises of a given image automatic judging.Images Classification problem is core topic of vision research, and many other vision research all will rely on and relate to Images Classification problem, and as objects in images detects, follows the tracks of, image is cut apart, object classification in video, detection, tracking, behavioural analysis, gesture identification etc.
K nearest neighbor Images Classification is a kind of image classification method, refers to that what when to Images Classification, adopt is the mode of k nearest neighbor ballot, and in K nearest image, the maximum classification of occurrence number is predicted as the classification of this test sample book.Except can realizing simply, efficiently image is classified, k nearest neighbor classification also has a lot of other characteristics.Such as k nearest neighbor Images Classification can obtain and the immediate sample of test pattern, can be applied in the fields such as image retrieval, face retrieval, video frequency searching.
Because selection and the image feature representation of sorter in conventional art is two independently processes, and k nearest neighbor classification is a nonparametric model, its prediction depends critically upon the space distribution of data, also be image feature representation, this is not optimum with regard to causing image feature representation with regard to k nearest neighbor classification, and classification performance is impacted.
In recent years, the development of Images Classification field rapidly, has obtained a lot of important breakthrough aspect sorting technique.Current, word bag model is one of image feature representation main flow framework.Word bag model is described and is carried out statistical nature description by the low-level image feature to the image block of intensive extraction, obtains the global feature of image is expressed.Word bag model conventionally by low-level image feature describe, the step such as vision word generates, low-level image feature coding, feature converge, sorter training and test forms, before sorter training, what we can think the employing of word bag model is that unsupervised mode is expressed image, no matter be that the traditional low-level image features such as SIFT, HOG or word bag model middle level features expressed, all do not use the label information of image, thereby such feature representation classifies for k nearest neighbor printenv model such, not optimum conventionally.
Summary of the invention
In view of this, fundamental purpose of the present invention is to provide a kind of image classification method based on data and task-driven, to realize on large scale image data set Images Classification more fast and accurately.
In order to achieve the above object, the present invention is by the following technical solutions:
An image classification method based on data and task-driven, comprising:
Data set is prepared, according to data set scale and picture material design convolutional neural networks structure;
Model training, is used the training of given grouped data set pair convolutional neural networks model;
Use the convolutional neural networks after training to carry out feature representation extraction to training set image;
Convolutional neural networks by after test pattern input training, is used k nearest neighbor mode to classify to test pattern.
Further, described data set is prepared, and according to data set scale and picture material design convolutional neural networks structure, also comprises:
One or more at least are in the following manner realized data and are strengthened: the marginal portion of 1) going image surrounding from original image random cropping with produce make new advances have nuance sample image; 2) in original image pixels, add random Gaussian to produce the sample image making new advances.
Further, described data set is prepared, and according to data set scale and picture material design convolutional neural networks structure, also comprises:
Image pattern is zoomed to fixed measure, and the input using vector of the stretching one-tenth of pixel as convolutional neural networks.
Further, described model training, is used the training of given grouped data set pair convolutional neural networks model, specifically also comprises:
Use convolutional neural networks as essential characteristic transformation model;
Based on adjacent component analysis expectation error rate as loss function to the training of convolutional neural networks model;
Optimization method based on gradient carries out network training, and uses GPU to carry out computing.
Further, the convolutional neural networks after described use training extracts feature representation to training set image, comprising:
Convolutional neural networks by after all training image input training, takes out the response of the full articulamentum of last one deck as the feature representation of every training image.
Further, the feature representation of training set image is configured to KD-tree pre-stored.
Further, described by the convolutional neural networks after test pattern input training, use k nearest neighbor mode is classified to test pattern, comprising:
For given test pattern, this image scaling is big or small to convolutional neural networks mode input, then send into convolutional neural networks and carry out forward calculation, take out the response of the full articulamentum of last one deck as the feature representation of this test pattern, use this expression in the feature representation of training set image, to carry out k nearest neighbor retrieval, the maximum classification of occurrence number in K nearest training image of feature representation is predicted as to the classification of this test pattern.
Further, described based on adjacent component analysis expectation error rate as loss function to the training of convolutional neural networks model, specifically comprise:
Adopt adjacent component analysis NCA to estimate k nearest neighbor error in classification, given N is to training sample { (x
i, y
i) | i=1 ..., N}, wherein x
iimage pattern, y
iits corresponding label, for a sample x
i, another sample x
jwith x
ibelonging to other definition of probability of same class is
Wherein F () is by the eigentransformation function of described convolutional neural networks;
Concerning adjacent component analysis, sample x
ithe probability that belongs to classification c and correctly classified is
Expectation error rate is
Wherein, y
ijrepresent sample x
iwith sample x
jwhether belong to same classification, y
i=y
jtime, y
ij=1, otherwise y
ij=0.Expectation error rate in formula (3) is that of k nearest neighbor classification error rate is approximate, the loss function by this expectation error rate as the network optimization.
Further, the described optimization method based on gradient carries out network training and specifically comprises one of following mode: random Gradient Descent, method of conjugate gradient, quasi-Newton method, L-BFGS.
The above-mentioned image classification method based on data and task-driven provided by the invention, compared with prior art has the following advantages:
1), adopt based on linear Convolution feature learning, can be with the mode implementation model of data-driven the self-adaptation to data set, thereby specific data set is better described.
2), by directly the error of k nearest neighbor being optimized, in the mode of task-driven, convolutional neural networks is optimized, can in k nearest neighbor task, obtain better performance.
3), in the training stage, can adopt GPU to carry out efficient training, at test phase, only need use CPU just can realize efficient k nearest neighbor Images Classification, be highly suitable for the tasks such as large-scale Images Classification, retrieval.
Accompanying drawing explanation
Fig. 1 is image classification method model training and the test flow chart based on data and task-driven according to the embodiment of the present invention;
Fig. 2 adopts adjacent component analysis on handwritten numeral database MNIST, to train the sub-schematic diagram of convolutional neural networks ground floor convolution obtaining according to the embodiment of the present invention;
Fig. 3 is the sub-schematic diagram of convolutional neural networks ground floor convolution that adopts adjacent component analysis to train on CIFAR-10 database to obtain according to the embodiment of the present invention;
Fig. 4 be the method dimensionality reduction that the present invention of MNIST use of numerals proposed according to the embodiment of the present invention to the result of 2 dimensions and with the contrast of additive method, the data point of different colours represents different numerals.
Embodiment
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.
Thought main points of the present invention are: 1) based on linear Convolution feature learning, can be with the mode implementation model of data-driven the self-adaptation to data set, thereby specific data set is better described; 2) the present invention is by directly the error of k nearest neighbor being optimized, and in the mode of task-driven, convolutional neural networks is optimized, and can in k nearest neighbor task, obtain better performance; 3) in the training stage, can adopt GPU to carry out efficient training, at test phase, only need use CPU just can realize efficient k nearest neighbor Images Classification, be highly suitable for the tasks such as large-scale Images Classification, retrieval.
As shown in Figure 1, Fig. 1 the first half is the image classification method model training process flow diagram based on data and task-driven according to the embodiment of the present invention.The structure that has shown convolutional neural networks in figure, before the sub-size of network layer convolution be 5 * 5, along with reducing of trellis diagram size, the sub-size of convolution adopting is below 3 * 3, has finally connected and has comprised 128 and 64 neuronic full articulamentums.By training set sample order is sent into convolutional neural networks, and use adjacent component analysis loss function to carry out from back to front error propagation, calculate the gradient of every layer network parameter, and use stochastic gradient descent algorithm to upgrade network with this gradient, realize the network model study of data-driven and task-driven.Fig. 1 the latter half has been set forth the test process of the method.The convolutional neural networks that different classes of original image input trains, the response that last full articulamentum of network is obtained is as the feature representation of image, formed a new non-linear space, image different classes of in this space can make a distinction better.
Method of the present invention comprises the following steps:
S1, data set are prepared, according to data set scale and picture material design convolutional neural networks structure.
Convolutional neural networks comprises a large amount of model parameters, and parameter means that model is more complicated more, easier training in there is over-fitting situation, algorithm performance on training set is fine, on test set, performance is very poor.Balance training data scale and model complexity are to prevent over-fitting, obtain the important channel of optimum performance.On the one hand, data volume is larger, and model training is more not easy over-fitting, and performance is better.But data volume is normally limited, this just need to adopt, and certain way is artificial produces new data, and we realize in the following manner data and strengthen: the marginal portion of 1) going image surrounding from original image random cropping with produce make new advances have nuance sample image; 2) in original image pixels, add random Gaussian to produce the sample image making new advances.All image patterns are all scaled to fixed measure, and vector of the stretching one-tenth of all pixels is as the input of convolutional neural networks.On the other hand, the in the situation that of given training set scale, need the model complexity of corresponding control convolutional neural networks.The model complexity of convolutional neural networks is conventionally directly related with the structure of model, and the number of plies of network is more, and the nodes of every layer is more, and can training parameter just more, model be just more complicated.
S2, based on adjacent component analysis expectation error rate, as loss function, convolutional neural networks is trained, optimization method adopts random Gradient Descent, and uses GPU to carry out computing.
Convolutional neural networks at Images Classification, the field such as detect, cut apart and be widely used.In these application, convolutional neural networks is normally trained based on general object classification criterion, as logistic recurrence, Softmax recurrence, cross entropy etc.The network that adopts this general standard training to obtain, can directly predict and obtain result, but will be used for directly processing k nearest neighbor classification problem, is not optimum conventionally.Above-mentioned general classification is end to end, from image, directly predicted, and k nearest neighbor problem normally has a feature representation to every image, use this feature representation to carry out k nearest neighbor retrieval, by the classification of K nearest sample, determine the classification of test sample book.If we use the network of general sorting criterion training to carry out feature representation (can get last one deck response of network as feature) to image, be difficult to guarantee that this feature representation is applicable to nearest neighbour classification sight.For learning the better feature representation for k nearest neighbor problem, we come directly k nearest neighbor error to be optimized by new training criterion, thereby guarantee that the feature representation of study is optimum under k nearest neighbor problem.
Directly using k nearest neighbor error is irrational as objective function, because need objective function can lead continuously, thereby can adopt random gradient descending method to upgrade network.We adopt adjacent component analysis (Neighborhood Component Analysis, NCA) to carry out approximate description to k nearest neighbor error in classification.Given N is to training sample { (x
i, y
i) | i=1 ..., N}, wherein x
iimage pattern, y
iit is its corresponding label.For a sample x
i, another sample x
jwith x
ibelonging to other definition of probability of same class is
Wherein F () is the non-linear transform function of a high complexity, and input picture is for conversion into a proper vector, and we represent F () with convolutional neural networks here, and last one deck response of taking out neural network is as feature representation.From formula 1, can find out sample x
jwith x
ifall into the Euclidean distance between the feature representation that other probability of same class is inversely proportional to both.
In k nearest neighbor classification, the prediction classification of a test sample book is to occur maximum classifications by adding up in its nearest K sample.Concerning adjacent component analysis, sample x
ithe probability that belongs to classification c and correctly classified is
So expectation error rate can be defined as
Y wherein
ijrepresent sample x
iwith sample x
jwhether belong to same classification, y
i=y
jtime, y
ij=1, otherwise y
ij=0.Expectation error rate in formula 3 is that of k nearest neighbor classification error rate is approximate, in this patent, with this, expects that error rate is as the objective function of the network optimization.The adjacent component loss function of formula 3 definition can be led continuously, thereby can use easily the optimization method based on gradient to carry out network training, as random Gradient Descent, method of conjugate gradient, quasi-Newton method, L-BFGS etc.
S3, the convolutional neural networks that all training image inputs are trained, take out the response of the full articulamentum of last one deck as the feature representation of every training image.After training convolutional neural networks, we are using this network as eigentransformation function F (), for extracting feature from image.K nearest neighbor classification is nonparametric model, and itself does not have parameter, only each image of training set need to be inputted to convolutional neural networks, and take out the full articulamentum of last one deck as the feature representation of this sample.The feature representation of the training set having extracted can be pre-stored, at test phase, for k nearest neighbor, retrieves.
S4, the convolutional neural networks that test pattern input is trained, take out the response of the full articulamentum of last one deck as the feature representation of this image, and use this feature representation to use k nearest neighbor mode to classify in the character pair of training set image is expressed, the maximum classification of occurrence number in the immediate K of a feature representation image is predicted as to the classification of this test pattern.The feature representation that the feature representation of test pattern and training set image extract in advance compares, and generally need to all once compare with all training samples, and this time complexity is proportional to training set size.For very large training set, this obvious cost is very high, for carrying out faster k nearest neighbor, searches, and adopts conventional this data structure of KD-Tree in nearest _neighbor retrieval, by K dimension space is cut apart, accelerates the speed of neighbour's retrieval.
Fig. 2 adopts adjacent component analysis on handwritten numeral database MNIST, to train the sub-schematic diagram of convolutional neural networks ground floor convolution obtaining according to the embodiment of the present invention.Wherein, what left figure showed is the sample in MNIST database, and what right figure showed is convolutional neural networks ground floor convolution after training, can find out, what learn is digital stroke substantially.
Fig. 3 is the sub-schematic diagram of convolutional neural networks ground floor convolution that adopts adjacent component analysis to train on CIFAR-10 database to obtain according to the embodiment of the present invention.From right figure, find out, convolution of learning is some edges and simple bottom visual pattern.
Fig. 4 be the method dimensionality reduction that MNIST use of numerals this patent proposed according to the embodiment of the present invention to the result of 2 dimensions and with the contrast of additive method, the data point of different colours represents different numerals.Can find out, different digital feature representation space distributions has realized preferably and having clustered, and is suitable for next adopting k nearest neighbor mode classification to classify.
In a word, the present invention proposes a kind of new image classification method based on data and task-driven, use deep layer convolutional neural networks as feature representation model, and use stochastic gradient descent algorithm on GPU, to carry out model training.After convolutional neural networks has been trained, for extracting feature representation from image and using k nearest neighbor way of search to carry out Images Classification.Experiment shows that this invention compares with the image classification algorithms of main flow based on k nearest neighbor and have that feature representation has strong identification, model training is subject to data-driven, task-driven, test process is very efficient, is suitable for k nearest neighbor Images Classification and retrieval under large scale data.
Above-described specific embodiment; object of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the foregoing is only specific embodiments of the invention; be not limited to the present invention; within the spirit and principles in the present invention all, any modification of making, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.
Claims (9)
1. the image classification method based on data and task-driven, is characterized in that, comprising:
Data set is prepared, according to data set scale and picture material design convolutional neural networks structure;
Model training, is used the training of given grouped data set pair convolutional neural networks model;
Use the convolutional neural networks after training to carry out feature representation extraction to training set image;
Convolutional neural networks by after test pattern input training, is used k nearest neighbor mode to classify to test pattern.
2. the image classification method based on data and task-driven according to claim 1, is characterized in that, data set is prepared, and according to data set scale and picture material design convolutional neural networks structure, also comprises:
One or more at least are in the following manner realized data and are strengthened: the marginal portion of 1) going image surrounding from original image random cropping with produce make new advances have nuance sample image; 2) in original image pixels, add random Gaussian to produce the sample image making new advances.
3. the image classification method based on data and task-driven according to claim 1, is characterized in that, data set is prepared, and according to data set scale and picture material design convolutional neural networks structure, also comprises:
Image pattern is zoomed to fixed measure, and the input using vector of the stretching one-tenth of pixel as convolutional neural networks.
4. the image classification method based on data and task-driven according to claim 1, is characterized in that, model training is used the training of given grouped data set pair convolutional neural networks model, specifically also comprises:
Use convolutional neural networks as essential characteristic transformation model;
Based on adjacent component analysis expectation error rate as loss function to the training of convolutional neural networks model;
Optimization method based on gradient carries out network training, and uses GPU to carry out computing.
5. the image classification method based on data and task-driven according to claim 1, is characterized in that, uses the convolutional neural networks after training to extract feature representation to training set image, comprising:
Convolutional neural networks by after all training image input training, takes out the response of the full articulamentum of last one deck as the feature representation of every training image.
6. the image classification method based on data and task-driven according to claim 5, is characterized in that, the feature representation of training set image is configured to KD-tree pre-stored.
7. the image classification method based on data and task-driven according to claim 1, is characterized in that, the convolutional neural networks by after test pattern input training, is used k nearest neighbor mode to classify to test pattern, comprising:
For given test pattern, this image scaling is big or small to convolutional neural networks mode input, then send into convolutional neural networks and carry out forward calculation, take out the response of the full articulamentum of last one deck as the feature representation of this test pattern, use this expression in the feature representation of training set image, to carry out k nearest neighbor retrieval, the maximum classification of occurrence number in K nearest training image of feature representation is predicted as to the classification of this test pattern.
8. the image classification method based on data and task-driven according to claim 4, is characterized in that, based on adjacent component analysis expectation error rate as loss function to the training of convolutional neural networks model, specifically comprise:
Adopt adjacent component analysis NCA to estimate k nearest neighbor error in classification, given N is to training sample { (x
i, y
i) | i=1 ..., N}, wherein, x
iimage pattern, y
iits corresponding label, for a sample x
i, another sample x
jwith x
ibelonging to other definition of probability of same class is
Wherein, F () is the eigentransformation function of described convolutional neural networks;
Concerning adjacent component analysis, sample x
ithe probability that belongs to classification c and correctly classified is
Expectation error rate is
Wherein, y
ijrepresent sample x
iwith sample x
jwhether belong to same classification, y
i=y
jtime, y
ij=1, otherwise y
ij=0; Expectation error rate in formula (3) is that of k nearest neighbor classification error rate is approximate, the loss function by this expectation error rate as the network optimization.
9. the image classification method based on data and task-driven according to claim 4, it is characterized in that, the optimization method based on gradient carries out network training and specifically comprises one of following mode: random Gradient Descent, method of conjugate gradient, quasi-Newton method, L-BFGS.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410224860.3A CN103984959B (en) | 2014-05-26 | 2014-05-26 | A kind of image classification method based on data and task-driven |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410224860.3A CN103984959B (en) | 2014-05-26 | 2014-05-26 | A kind of image classification method based on data and task-driven |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103984959A true CN103984959A (en) | 2014-08-13 |
| CN103984959B CN103984959B (en) | 2017-07-21 |
Family
ID=51276921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410224860.3A Active CN103984959B (en) | 2014-05-26 | 2014-05-26 | A kind of image classification method based on data and task-driven |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103984959B (en) |
Cited By (43)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182735A (en) * | 2014-08-18 | 2014-12-03 | 厦门美图之家科技有限公司 | Training optimization pornographic picture or video detection method based on convolutional neural network |
| CN104217225A (en) * | 2014-09-02 | 2014-12-17 | 中国科学院自动化研究所 | A visual target detection and labeling method |
| CN104361363A (en) * | 2014-11-25 | 2015-02-18 | 中国科学院自动化研究所 | Deep deconvolution feature learning network, generating method thereof and image classifying method |
| CN104517122A (en) * | 2014-12-12 | 2015-04-15 | 浙江大学 | Image target recognition method based on optimized convolution architecture |
| CN104572965A (en) * | 2014-12-31 | 2015-04-29 | 南京理工大学 | Search-by-image system based on convolutional neural network |
| CN104700118A (en) * | 2015-03-18 | 2015-06-10 | 中国科学院自动化研究所 | Pulmonary nodule benignity and malignancy predicting method based on convolutional neural networks |
| CN104778448A (en) * | 2015-03-24 | 2015-07-15 | 孙建德 | Structure adaptive CNN (Convolutional Neural Network)-based face recognition method |
| CN104794501A (en) * | 2015-05-14 | 2015-07-22 | 清华大学 | Mode identification method and device |
| CN104991959A (en) * | 2015-07-21 | 2015-10-21 | 北京京东尚科信息技术有限公司 | Method and system for retrieving same or similar image based on content |
| CN105022836A (en) * | 2015-08-14 | 2015-11-04 | 中国科学技术大学 | Compact depth CNN characteristic indexing method based on SIFT insertion |
| CN105354565A (en) * | 2015-12-23 | 2016-02-24 | 北京市商汤科技开发有限公司 | Full convolution network based facial feature positioning and distinguishing method and system |
| CN105590099A (en) * | 2015-12-22 | 2016-05-18 | 中国石油大学(华东) | Multi-user behavior identification method based on improved convolutional neural network |
| WO2016090520A1 (en) * | 2014-12-10 | 2016-06-16 | Xiaogang Wang | A method and a system for image classification |
| CN105718960A (en) * | 2016-01-27 | 2016-06-29 | 北京工业大学 | Image ordering model based on convolutional neural network and spatial pyramid matching |
| CN105718890A (en) * | 2016-01-22 | 2016-06-29 | 北京大学 | Method for detecting specific videos based on convolution neural network |
| CN105894046A (en) * | 2016-06-16 | 2016-08-24 | 北京市商汤科技开发有限公司 | Convolutional neural network training and image processing method and system and computer equipment |
| CN106056529A (en) * | 2015-04-03 | 2016-10-26 | 阿里巴巴集团控股有限公司 | Method and equipment for training convolutional neural network used for image recognition |
| CN106203442A (en) * | 2016-06-28 | 2016-12-07 | 北京小白世纪网络科技有限公司 | A kind of copy image feature extracting method based on degree of depth study |
| CN106326931A (en) * | 2016-08-25 | 2017-01-11 | 南京信息工程大学 | Mammary gland molybdenum target image automatic classification method based on deep learning |
| CN106373112A (en) * | 2016-08-31 | 2017-02-01 | 北京比特大陆科技有限公司 | Image processing method, image processing device and electronic equipment |
| CN106485192A (en) * | 2015-09-02 | 2017-03-08 | 富士通株式会社 | Training method for the neutral net of image recognition and device |
| CN106663186A (en) * | 2014-07-28 | 2017-05-10 | 北京市商汤科技开发有限公司 | A method for face recognition and a system thereof |
| CN106650694A (en) * | 2016-12-30 | 2017-05-10 | 江苏四点灵机器人有限公司 | Human face recognition method taking convolutional neural network as feature extractor |
| WO2017084222A1 (en) * | 2015-11-22 | 2017-05-26 | 南方医科大学 | Convolutional neural network-based method for processing x-ray chest radiograph bone suppression |
| WO2017088553A1 (en) * | 2015-11-23 | 2017-06-01 | 广州视源电子科技股份有限公司 | Method and system for rapidly identifying and marking electronic component polarity direction |
| CN107145898A (en) * | 2017-04-14 | 2017-09-08 | 北京航星机器制造有限公司 | A kind of ray image sorting technique based on neutral net |
| CN107256384A (en) * | 2017-05-22 | 2017-10-17 | 汕头大学 | A kind of card recognition and method of counting based on image and signal transacting |
| CN107492090A (en) * | 2016-06-09 | 2017-12-19 | 西门子保健有限责任公司 | Analyzed according to generated data using the tumor phenotypes based on image of machine learning |
| CN107564070A (en) * | 2017-09-05 | 2018-01-09 | 国网浙江省电力公司湖州供电公司 | The ranging of large scene binocular and its bearing calibration in the monitoring of overhead power transmission channel image |
| CN107609586A (en) * | 2017-09-08 | 2018-01-19 | 深圳市唯特视科技有限公司 | A kind of visual characteristic learning method based on self-supervision |
| CN107871497A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | Audio recognition method and device |
| CN107958219A (en) * | 2017-12-06 | 2018-04-24 | 电子科技大学 | Image scene classification method based on multi-model and Analysis On Multi-scale Features |
| CN108010049A (en) * | 2017-11-09 | 2018-05-08 | 华南理工大学 | Split the method in human hand region in stop-motion animation using full convolutional neural networks |
| CN108235770A (en) * | 2017-12-29 | 2018-06-29 | 深圳前海达闼云端智能科技有限公司 | image identification method and cloud system |
| CN108288282A (en) * | 2017-12-26 | 2018-07-17 | 浙江工业大学 | A kind of adaptive features select method for tracking target based on convolutional neural networks |
| CN108304546A (en) * | 2018-01-31 | 2018-07-20 | 杭州电子科技大学 | A kind of medical image search method based on content similarity and Softmax graders |
| CN109584142A (en) * | 2018-12-05 | 2019-04-05 | 网易传媒科技（北京）有限公司 | Image Intensified System and method, training method, medium and electronic equipment |
| WO2019085793A1 (en) * | 2017-11-01 | 2019-05-09 | 腾讯科技（深圳）有限公司 | Image classification method, computer device and computer readable storage medium |
| CN109740611A (en) * | 2019-01-25 | 2019-05-10 | 中电健康云科技有限公司 | Tongue image analysis method and device |
| CN110110780A (en) * | 2019-04-30 | 2019-08-09 | 南开大学 | A kind of picture classification method based on confrontation neural network and magnanimity noise data |
| CN110689961A (en) * | 2019-09-03 | 2020-01-14 | 重庆大学 | Gastric cancer disease risk detection device based on big data analysis technology |
| CN111758105A (en) * | 2018-05-18 | 2020-10-09 | 谷歌有限责任公司 | Learning data enhancement strategy |
| US11157811B2 (en) | 2019-10-28 | 2021-10-26 | International Business Machines Corporation | Stub image generation for neural network training |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110222724A1 (en) * | 2010-03-15 | 2011-09-15 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
| CN103177264A (en) * | 2013-03-14 | 2013-06-26 | 中国科学院自动化研究所 | Image classification method based on visual dictionary global topological representations |
| CN103366180A (en) * | 2013-06-14 | 2013-10-23 | 山东大学 | Cell image segmentation method based on automatic feature learning |
| CN103544506A (en) * | 2013-10-12 | 2014-01-29 | Tcl集团股份有限公司 | Method and device for classifying images on basis of convolutional neural network |
-
2014
- 2014-05-26 CN CN201410224860.3A patent/CN103984959B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110222724A1 (en) * | 2010-03-15 | 2011-09-15 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
| CN103177264A (en) * | 2013-03-14 | 2013-06-26 | 中国科学院自动化研究所 | Image classification method based on visual dictionary global topological representations |
| CN103366180A (en) * | 2013-06-14 | 2013-10-23 | 山东大学 | Cell image segmentation method based on automatic feature learning |
| CN103544506A (en) * | 2013-10-12 | 2014-01-29 | Tcl集团股份有限公司 | Method and device for classifying images on basis of convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| ALEX KRIZHEVSKY: "ImageNet Classification with Deep Convolutional Neural Networks", 《ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS》 * |
| 高学等: "基于CNN和随机弹性形变的相似手写汉字识别", 《华南理工大学学报(自然科学版)》 * |
Cited By (64)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106663186B (en) * | 2014-07-28 | 2018-08-21 | 北京市商汤科技开发有限公司 | method and system for face recognition |
| CN106663186A (en) * | 2014-07-28 | 2017-05-10 | 北京市商汤科技开发有限公司 | A method for face recognition and a system thereof |
| CN104182735A (en) * | 2014-08-18 | 2014-12-03 | 厦门美图之家科技有限公司 | Training optimization pornographic picture or video detection method based on convolutional neural network |
| CN104217225A (en) * | 2014-09-02 | 2014-12-17 | 中国科学院自动化研究所 | A visual target detection and labeling method |
| CN104217225B (en) * | 2014-09-02 | 2018-04-24 | 中国科学院自动化研究所 | A kind of sensation target detection and mask method |
| CN104361363A (en) * | 2014-11-25 | 2015-02-18 | 中国科学院自动化研究所 | Deep deconvolution feature learning network, generating method thereof and image classifying method |
| CN104361363B (en) * | 2014-11-25 | 2018-01-16 | 中国科学院自动化研究所 | Depth deconvolution feature learning network, generation method and image classification method |
| WO2016090520A1 (en) * | 2014-12-10 | 2016-06-16 | Xiaogang Wang | A method and a system for image classification |
| CN104517122A (en) * | 2014-12-12 | 2015-04-15 | 浙江大学 | Image target recognition method based on optimized convolution architecture |
| CN104572965A (en) * | 2014-12-31 | 2015-04-29 | 南京理工大学 | Search-by-image system based on convolutional neural network |
| CN104700118A (en) * | 2015-03-18 | 2015-06-10 | 中国科学院自动化研究所 | Pulmonary nodule benignity and malignancy predicting method based on convolutional neural networks |
| CN104778448A (en) * | 2015-03-24 | 2015-07-15 | 孙建德 | Structure adaptive CNN (Convolutional Neural Network)-based face recognition method |
| CN104778448B (en) * | 2015-03-24 | 2017-12-15 | 孙建德 | A kind of face identification method based on structure adaptive convolutional neural networks |
| CN106056529A (en) * | 2015-04-03 | 2016-10-26 | 阿里巴巴集团控股有限公司 | Method and equipment for training convolutional neural network used for image recognition |
| CN106056529B (en) * | 2015-04-03 | 2020-06-02 | 阿里巴巴集团控股有限公司 | Method and equipment for training convolutional neural network for picture recognition |
| CN104794501A (en) * | 2015-05-14 | 2015-07-22 | 清华大学 | Mode identification method and device |
| CN104794501B (en) * | 2015-05-14 | 2021-01-05 | 清华大学 | Pattern recognition method and device |
| CN104991959A (en) * | 2015-07-21 | 2015-10-21 | 北京京东尚科信息技术有限公司 | Method and system for retrieving same or similar image based on content |
| CN104991959B (en) * | 2015-07-21 | 2019-11-05 | 北京京东尚科信息技术有限公司 | A kind of method and system of the same or similar image of information retrieval based on contents |
| CN105022836B (en) * | 2015-08-14 | 2018-07-03 | 中国科学技术大学 | Compact depth CNN aspect indexing methods based on SIFT insertions |
| CN105022836A (en) * | 2015-08-14 | 2015-11-04 | 中国科学技术大学 | Compact depth CNN characteristic indexing method based on SIFT insertion |
| CN106485192A (en) * | 2015-09-02 | 2017-03-08 | 富士通株式会社 | Training method for the neutral net of image recognition and device |
| CN106485192B (en) * | 2015-09-02 | 2019-12-06 | 富士通株式会社 | Training method and device of neural network for image recognition |
| WO2017084222A1 (en) * | 2015-11-22 | 2017-05-26 | 南方医科大学 | Convolutional neural network-based method for processing x-ray chest radiograph bone suppression |
| WO2017088553A1 (en) * | 2015-11-23 | 2017-06-01 | 广州视源电子科技股份有限公司 | Method and system for rapidly identifying and marking electronic component polarity direction |
| CN105590099A (en) * | 2015-12-22 | 2016-05-18 | 中国石油大学(华东) | Multi-user behavior identification method based on improved convolutional neural network |
| CN105590099B (en) * | 2015-12-22 | 2019-02-01 | 中国石油大学(华东) | A kind of more people's Activity recognition methods based on improvement convolutional neural networks |
| CN105354565A (en) * | 2015-12-23 | 2016-02-24 | 北京市商汤科技开发有限公司 | Full convolution network based facial feature positioning and distinguishing method and system |
| CN105718890A (en) * | 2016-01-22 | 2016-06-29 | 北京大学 | Method for detecting specific videos based on convolution neural network |
| CN105718960B (en) * | 2016-01-27 | 2019-01-04 | 北京工业大学 | Based on convolutional neural networks and the matched image order models of spatial pyramid |
| CN105718960A (en) * | 2016-01-27 | 2016-06-29 | 北京工业大学 | Image ordering model based on convolutional neural network and spatial pyramid matching |
| CN107492090A (en) * | 2016-06-09 | 2017-12-19 | 西门子保健有限责任公司 | Analyzed according to generated data using the tumor phenotypes based on image of machine learning |
| CN105894046A (en) * | 2016-06-16 | 2016-08-24 | 北京市商汤科技开发有限公司 | Convolutional neural network training and image processing method and system and computer equipment |
| CN105894046B (en) * | 2016-06-16 | 2019-07-02 | 北京市商汤科技开发有限公司 | Method and system, the computer equipment of convolutional neural networks training and image procossing |
| CN106203442A (en) * | 2016-06-28 | 2016-12-07 | 北京小白世纪网络科技有限公司 | A kind of copy image feature extracting method based on degree of depth study |
| CN106203442B (en) * | 2016-06-28 | 2019-04-05 | 北京小白世纪网络科技有限公司 | A kind of copy image feature extracting method based on deep learning |
| CN106326931A (en) * | 2016-08-25 | 2017-01-11 | 南京信息工程大学 | Mammary gland molybdenum target image automatic classification method based on deep learning |
| CN106373112B (en) * | 2016-08-31 | 2020-08-04 | 北京比特大陆科技有限公司 | Image processing method and device and electronic equipment |
| CN106373112A (en) * | 2016-08-31 | 2017-02-01 | 北京比特大陆科技有限公司 | Image processing method, image processing device and electronic equipment |
| CN107871497A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | Audio recognition method and device |
| CN106650694A (en) * | 2016-12-30 | 2017-05-10 | 江苏四点灵机器人有限公司 | Human face recognition method taking convolutional neural network as feature extractor |
| CN107145898A (en) * | 2017-04-14 | 2017-09-08 | 北京航星机器制造有限公司 | A kind of ray image sorting technique based on neutral net |
| CN107256384A (en) * | 2017-05-22 | 2017-10-17 | 汕头大学 | A kind of card recognition and method of counting based on image and signal transacting |
| CN107564070B (en) * | 2017-09-05 | 2020-06-12 | 国网浙江省电力公司湖州供电公司 | Large-scene binocular ranging and correcting method in overhead power transmission channel image monitoring |
| CN107564070A (en) * | 2017-09-05 | 2018-01-09 | 国网浙江省电力公司湖州供电公司 | The ranging of large scene binocular and its bearing calibration in the monitoring of overhead power transmission channel image |
| CN107609586A (en) * | 2017-09-08 | 2018-01-19 | 深圳市唯特视科技有限公司 | A kind of visual characteristic learning method based on self-supervision |
| WO2019085793A1 (en) * | 2017-11-01 | 2019-05-09 | 腾讯科技（深圳）有限公司 | Image classification method, computer device and computer readable storage medium |
| US11361192B2 (en) | 2017-11-01 | 2022-06-14 | Tencent Technology (Shenzhen) Company Limited | Image classification method, computer device, and computer-readable storage medium |
| CN108010049A (en) * | 2017-11-09 | 2018-05-08 | 华南理工大学 | Split the method in human hand region in stop-motion animation using full convolutional neural networks |
| CN107958219A (en) * | 2017-12-06 | 2018-04-24 | 电子科技大学 | Image scene classification method based on multi-model and Analysis On Multi-scale Features |
| CN108288282A (en) * | 2017-12-26 | 2018-07-17 | 浙江工业大学 | A kind of adaptive features select method for tracking target based on convolutional neural networks |
| CN108288282B (en) * | 2017-12-26 | 2022-04-08 | 浙江工业大学 | Adaptive feature selection target tracking method based on convolutional neural network |
| CN108235770B (en) * | 2017-12-29 | 2021-10-19 | 达闼机器人有限公司 | Image identification method and cloud system |
| CN108235770A (en) * | 2017-12-29 | 2018-06-29 | 深圳前海达闼云端智能科技有限公司 | image identification method and cloud system |
| WO2019127451A1 (en) * | 2017-12-29 | 2019-07-04 | 深圳前海达闼云端智能科技有限公司 | Image recognition method and cloud system |
| CN108304546A (en) * | 2018-01-31 | 2018-07-20 | 杭州电子科技大学 | A kind of medical image search method based on content similarity and Softmax graders |
| CN111758105A (en) * | 2018-05-18 | 2020-10-09 | 谷歌有限责任公司 | Learning data enhancement strategy |
| CN109584142A (en) * | 2018-12-05 | 2019-04-05 | 网易传媒科技（北京）有限公司 | Image Intensified System and method, training method, medium and electronic equipment |
| CN109740611A (en) * | 2019-01-25 | 2019-05-10 | 中电健康云科技有限公司 | Tongue image analysis method and device |
| CN110110780A (en) * | 2019-04-30 | 2019-08-09 | 南开大学 | A kind of picture classification method based on confrontation neural network and magnanimity noise data |
| CN110110780B (en) * | 2019-04-30 | 2023-04-07 | 南开大学 | Image classification method based on antagonistic neural network and massive noise data |
| CN110689961A (en) * | 2019-09-03 | 2020-01-14 | 重庆大学 | Gastric cancer disease risk detection device based on big data analysis technology |
| CN110689961B (en) * | 2019-09-03 | 2022-12-09 | 重庆大学 | Gastric cancer disease risk detection device based on big data analysis technology |
| US11157811B2 (en) | 2019-10-28 | 2021-10-26 | International Business Machines Corporation | Stub image generation for neural network training |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103984959B (en) | 2017-07-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103984959A (en) | Data-driven and task-driven image classification method | |
| CN109344736B (en) | Static image crowd counting method based on joint learning | |
| CN106447658B (en) | Conspicuousness object detection method based on global and local convolutional network | |
| CN109271522B (en) | Comment emotion classification method and system based on deep hybrid model transfer learning | |
| CN107943856A (en) | A kind of file classification method and system based on expansion marker samples | |
| CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
| CN106156777B (en) | Text picture detection method and device | |
| CN104966104A (en) | Three-dimensional convolutional neural network based video classifying method | |
| CN104573669A (en) | Image object detection method | |
| CN104217225A (en) | A visual target detection and labeling method | |
| CN105205448A (en) | Character recognition model training method based on deep learning and recognition method thereof | |
| CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
| CN103236068B (en) | A kind of topography matching process | |
| CN110569843B (en) | Intelligent detection and identification method for mine target | |
| CN104463101A (en) | Answer recognition method and system for textual test question | |
| CN103049763A (en) | Context-constraint-based target identification method | |
| CN110263174B (en) | Topic category analysis method based on focus attention | |
| Sajanraj et al. | Indian sign language numeral recognition using region of interest convolutional neural network | |
| CN106778687A (en) | Method for viewing points detecting based on local evaluation and global optimization | |
| CN104966105A (en) | Robust machine error retrieving method and system | |
| CN105279519A (en) | Remote sensing image water body extraction method and system based on cooperative training semi-supervised learning | |
| CN109685065A (en) | Printed page analysis method, the system of paper automatic content classification | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| CN104200233A (en) | Clothes classification and identification method based on Weber local descriptor | |
| CN113554100A (en) | Web service classification method for enhancing attention network of special composition picture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |