CN109460777A - Picture classification method, device and computer readable storage medium - Google Patents
Picture classification method, device and computer readable storage medium Download PDFInfo
- Publication number
- CN109460777A CN109460777A CN201811183703.7A CN201811183703A CN109460777A CN 109460777 A CN109460777 A CN 109460777A CN 201811183703 A CN201811183703 A CN 201811183703A CN 109460777 A CN109460777 A CN 109460777A
- Authority
- CN
- China
- Prior art keywords
- picture
- finite field
- feature space
- sorted
- feature
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application involves a kind of picture classification method, device, computer readable storage medium and computer equipments, by obtaining each picture to be sorted, each picture input feature vector to be sorted is extracted into model, obtain the feature vector of each picture to be sorted, obtain distribution situation of the feature vector in feature space of each picture to be sorted, according to the distribution situation, the finite field for carrying out classification to the feature space is divided, and obtains picture classification result.By the way that picture to be sorted is carried out characteristic vector pickup, obtain distribution situation of the feature vector in feature space of each picture to be sorted, the finite field that distribution situation based on feature space carries out classification divides, non-targeted picture is removed from Target Photo, reduce influence of the non-targeted picture to Target Photo nicety of grading, solve the picture in need for participating in classification be made of the different Target Photo of N kind and non-targeted picture, lead to the accuracy of picture classification and the low problem of precision.
Description
Technical field
This application involves Internet technical field, more particularly to a kind of picture classification method, device, computer-readable deposit
Storage media and computer equipment.
Background technique
With the continuous development of Internet technology, AI artificial intelligence is being widely applied in every field, computer view
Feel one of the important application as artificial intelligence, when classifying for needs to a large amount of pictures, plays very big help.
Currently, the mode classified to a large amount of pictures generallys use: the picture that needs are classified extracts correspondence
Feature vector, based on the common feature of each feature vector, the standard being previously set by certain is divided into spirogram piece respectively
Class in, show different degrees of feature between the class and class divided.By it is in need participate in classification picture be by N kind
Different Target Photos and non-targeted picture composition, after being classified using this classification method to picture, therefore, picture classification
Accuracy and precision it is low.
Summary of the invention
Based on this, it is necessary to for by the picture in need for participating in classification be by the different Target Photo of N kind and non-
Target Photo composition causes the accuracy of picture classification and the low problem of precision to provide a kind of picture classification method, device, calculating
Machine readable storage medium storing program for executing and computer equipment.
A kind of picture classification method characterized by comprising
Obtain each picture to be sorted；
Each picture input feature vector to be sorted is extracted into model, obtains the feature vector of each picture to be sorted；
Obtain distribution situation of the feature vector in feature space of each picture to be sorted；
According to the distribution situation, the finite field for carrying out classification to the feature space is divided, and obtains picture classification result.
It is described according to the distribution situation in one of the embodiments, the limited of classification is carried out to the feature space
Domain divides, and the step of obtaining picture classification result includes:
According to the feature vector of each picture to be sorted in the distribution situation of feature space, the feature space is carried out
Finite field divides, and obtains corresponding finite field of all categories；
According to each picture in determining each finite field, picture classification result is obtained.
In one of the embodiments, according to the feature vector of each picture to be sorted feature space distribution feelings
Condition carries out finite field division to the feature space, obtains corresponding finite field of all categories, comprising:
According to the feature vector of picture to be sorted in the distribution situation of feature space, determine of all categories in feature space
Heart position；
Using each center as the centre of sphere, pre-determined distance is radius, is divided to the feature space；
It is described according to each picture in determining each finite field, obtain picture classification result, comprising: according to be located at it is each corresponding
The intracorporal picture of ball obtains picture classification result of all categories.
In one of the embodiments, according to the feature vector of each picture to be sorted feature space distribution feelings
Condition carries out finite field division to the feature space, obtains corresponding finite field of all categories, comprising:
Obtain the band Density Estimator function and threshold value of feature space of all categories；
According to the band Density Estimator function, each picture to be sorted for calculating the feature space of all categories is corresponding
The density of point；
Finite field division is carried out to the feature space according to the density and the threshold value；
It is described according to each picture in determining each finite field, obtain picture classification result, comprising: be greater than according to density described
The picture of threshold value obtains the classification results of picture of all categories.
In one of the embodiments, according to the feature vector of each picture to be sorted feature space distribution feelings
Condition carries out finite field division to the feature space, obtains corresponding finite field of all categories, comprising:
By the feature vector of the picture to be sorted in the distribution of feature space, inputs single class that training obtains in advance and support
Vector machine disaggregated model；
Finite field of all categories is carried out to the feature space by the one-class support vector machines disaggregated model to divide, and is obtained
Finite field of all categories.
In one of the embodiments, according to the feature vector of each picture to be sorted feature space distribution feelings
Condition carries out finite field division to the feature space, obtains corresponding finite field of all categories, comprising:
Based on nearest neighbor algorithm, finite field of all categories is carried out to the feature space and is divided, finite field of all categories is obtained.
A kind of picture classifier in one of the embodiments, comprising:
Picture to be sorted obtains module, for obtaining each picture to be sorted；
Characteristic vector pickup module obtains each to be sorted for each picture input feature vector to be sorted to be extracted model
The feature vector of picture；
Feature space obtains module, for obtaining distribution feelings of the feature vector in feature space of each picture to be sorted
Condition；
Picture classification result determining module, for according to the distribution situation, carrying out classification to the feature space to have
Confinement divides, and obtains picture classification result.
The picture classification result determining module includes: in one of the embodiments,
Finite field division unit, for according to the feature vector of each picture to be sorted feature space distribution feelings
Condition carries out finite field division to the feature space, obtains corresponding finite field of all categories；
Picture classification result determination unit, for obtaining picture classification result according to each picture in determining each finite field.
A kind of computer readable storage medium is stored with computer program, when the computer program is executed by processor,
So that the processor is executed such as the step of the picture classification method.
A kind of computer equipment, including memory and processor, the memory are stored with computer program, the calculating
When machine program is executed by the processor, so that the processor is executed such as the step of the picture classification method.
Above-mentioned picture classification method, device, computer readable storage medium and computer equipment, it is each to be sorted by obtaining
Each picture input feature vector to be sorted is extracted model, obtains the feature vector of each picture to be sorted, obtained each described by picture
The feature vector of picture to be sorted carries out the feature space according to the distribution situation in the distribution situation of feature space
The finite field of classification divides, and obtains picture classification result.By the way that picture to be sorted is carried out characteristic vector pickup, obtain respectively wait divide
Distribution situation of the feature vector of class picture in feature space, the finite field stroke of the distribution situation progress classification based on feature space
Point, non-targeted picture is removed from Target Photo, influence of the non-targeted picture to Target Photo nicety of grading is reduced, solves
It is in need participate in classification picture be made of the different Target Photo of N kind and non-targeted picture, lead to the accurate of picture classification
Spend the problem low with precision.
Detailed description of the invention
Fig. 1 is the flow diagram of picture classification method in the application one embodiment；
Fig. 2 is to be mentioned in picture classification method based on Softmax Loss and Center Loss in the application one embodiment
The X-Y scheme that the feature vector taken is shown；
Fig. 3 is the feature vector based on t-SNE algorithm is shown in picture classification method in the application one embodiment two
Dimension figure；
Fig. 4 is the feature vector of the mode in the application one embodiment based on centre distance refusal in picture classification method
The X-Y scheme of display；
Fig. 5 is the space vector figure based on kernel function density in picture classification method in the application one embodiment；
Fig. 6 is the space vector figure based on nearest neighbor algorithm in picture classification method in the application one embodiment；
Fig. 7 is the schematic diagram of picture classifier in the application one embodiment；
Fig. 8 is the structural block diagram of computer equipment in the application one embodiment.
Specific embodiment
For the objects, technical solutions and advantages of the application are more clearly understood, with reference to the accompanying drawings and embodiments, to this
Application is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the application,
The protection scope of the application is not limited.
As shown in Figure 1, in one embodiment, providing a kind of picture classification method, specifically comprising the following steps S210
To step S240:
Step S210 obtains each picture to be sorted.
Wherein, picture to be sorted is the picture classified.The picture to be sorted can pass through various possible sides
Formula obtains, such as can be the picture of user's acquisition, saves into memory, when being classified, obtains from memory；
It is also possible to the picture that user acquires in real time, is uploaded in picture classifier, as picture to be sorted.
Each picture input feature vector to be sorted is extracted model, obtains the feature vector of each picture to be sorted by step S220.
Wherein, Feature Selection Model is the feature extractor based on depth convolutional neural networks.Pass through depth convolutional Neural
(convolutional neural networks are a kind of feedforward neural networks to network, its artificial neuron can respond in a part of coverage area
Surrounding cells have outstanding performance for large-scale image procossing, it includes convolutional layer (convolutional layer) and pond layer
(pooling layer), and depth convolutional neural networks are exactly the convolutional neural networks comprising multitiered network) by picture to be sorted
Carry out characteristic vector pickup.By Feature Selection Model, as shown in Fig. 2, picture to be sorted, which is abstracted as in higher dimensional space, to be had
The feature vector of " separating between convergence, class in class " characteristic, can be and set Softmax for the loss function of neural network
The cum rights of Loss and Center Loss and, using Softmax in the height of different classes of easy separability and Center Loss
Poly- property realizes the extraction of feature vector, obtains the feature vector of each picture to be sorted.
Step S240 obtains distribution situation of the feature vector in feature space of each picture to be sorted.Wherein, pass through feature
After each picture to be sorted is abstracted as the feature vector in feature space by extraction model, the feature vector of each picture to be sorted is obtained
In feature space distribution situation, the distribution situation of feature space can be the two-dimensional feature space obtained after dimension-reduction treatment
Distribution situation is also possible to the distribution situation of the feature space of other dimensions.For example, using t-SNE algorithm by feature extraction mould
Type is extracted on high dimensional feature vector dimensionality reduction to two-dimensional space and is shown, as shown in figure 3, being wherein non-targeted production compared with dark colour
Product, as the dotted line frame in Fig. 3 show an example of non-targeted product.Step S260, according to distribution situation, to feature space
The finite field for carrying out classification divides, and obtains picture classification result.
Wherein, according to the feature vector of each picture to be sorted feature space distribution situation, based on it is of all categories to point
Distribution of the class picture in feature space carries out finite field division to feature space, so that (positive sample refers to target to positive sample
Picture) it falls in finite field, negative sample (negative sample refers to non-targeted picture) is fallen in outside finite field, utilizes finite field set pair
The outer negative sample of finite field is excluded, and picture classification result is obtained.It can be by based on the center of all categories in feature space
Distance carries out finite field division, can also be by being based on one-class support vector machines (One Class Support Vector
Machine finite field division) is carried out, can also estimate to carry out finite field division by the distribution density based on kernel function, may be used also
By carrying out finite field division based on nearest neighbor algorithm (KNN).
Above-mentioned picture classification method is extracted each picture input feature vector to be sorted by obtaining each picture to be sorted
Model obtains the feature vector of each picture to be sorted, obtains point of the feature vector in feature space of each picture to be sorted
Cloth situation, according to the distribution situation, the finite field for carrying out classification to the feature space is divided, and obtains picture classification result.
By the way that picture to be sorted is carried out characteristic vector pickup, distribution feelings of the feature vector in feature space of each picture to be sorted are obtained
Condition, the finite field that the distribution situation based on feature space carries out classification divide, that is, being located at the picture in finite field is positive sample, position
Non-targeted picture is reduced to mesh so that non-targeted picture is removed from Target Photo in the picture outside finite field for negative sample
The influence for piece nicety of grading of marking on a map solves the problems, such as that the accuracy of picture classification and precision are low.
In one embodiment, according to the distribution situation, the finite field for carrying out classification to feature space is divided, and obtains figure
The step of piece classification results includes: the distribution situation according to the feature vector of each picture to be sorted in feature space, to feature sky
Between carry out finite field division, obtain corresponding finite field of all categories；According to each picture in determining each finite field, picture point is obtained
Class result.
Wherein, according to the feature vector of each picture to be sorted feature space distribution situation, based on it is of all categories to point
Distribution of the class picture in feature space carries out finite field division to each classification in feature space, obtains correspondence of all categories
Finite field so that positive sample of all categories is fallen in each finite field, negative sample of all categories is fallen in outside each finite field, benefit
Negative sample outside each finite field is excluded with finite field set, obtains picture classification result.It can be by based on of all categories
Centre distance in feature space carries out finite field division, can also be by being based on one-class support vector machines (One Class
Support Vector Machine) finite field division is carried out, it can also estimate to carry out by the distribution density based on kernel function
Finite field divides, can also be by carrying out finite field division based on nearest neighbor algorithm (KNN).
In one embodiment, according to the feature vector of each picture to be sorted feature space distribution situation, to feature
Space carries out finite field division, obtains corresponding finite field of all categories, comprising: according to the feature vector of picture to be sorted in feature
The distribution situation in space determines the center of all categories in feature space；Using each center as the centre of sphere, pre-determined distance is half
Diameter divides feature space；According to each picture in determining each finite field, picture classification result is obtained, comprising: according to position
In each corresponding intracorporal picture of ball, picture classification result of all categories is obtained.
Specifically, in the present embodiment, the mode based on centre distance refusal carries out not confinement and divides.Wherein, pre-determined distance
It can be user and picture sample picture extracted into feature vector using depth convolutional neural networks, the feature vector of higher-dimension
It is shown on X-Y scheme, observes distribution (as shown in Figure 4) of the feature on X-Y scheme, be calculated for distinguishing positive sample and bearing
The pre-determined distance of sample.According to picture in feature space distribution situation, be integrated into feature space with positive sample of all categories
Heart position is the centre of sphere, and pre-determined distance is that radius obtains higher-dimension sphere set, and each picture fallen in set (is fallen in each finite field
Each picture) be considered positive sample, fall in set outside each picture (each picture fallen in outside each finite field) be considered negative sample,
The negative sample that will be located in sphere set is rejected from the finite field of the category, and the positive sample in sphere set is such
The classification results of other picture.
In the present embodiment, finite field division is carried out based on the centre distance of all categories in feature space, set will be fallen in
Outer each picture is considered that negative sample is excluded, and improves the accuracy and precision of picture classification.
In one embodiment, according to the feature vector of each picture to be sorted feature space distribution situation, to feature
Space carries out finite field division, obtains corresponding finite field of all categories, comprising: obtain the band cuclear density of feature space of all categories
Estimation function and threshold value；According to band Density Estimator function, each picture to be sorted for calculating feature space of all categories is corresponding
The density of point；Finite field division is carried out to feature space according to density and threshold value；According to each picture in determining each finite field, obtain
To picture classification result, comprising: be greater than the picture of threshold value according to density, obtain the classification results of picture of all categories.
Wherein, the distribution according to positive sample in feature space obtains the band of the total space using kernel density estimation algorithm
Density Estimator function, and given threshold, as shown in figure 5, the density of point corresponding in feature space is greater than preset threshold
Each picture to be sorted is considered positive sample, and the density of point corresponding in feature space is less than to each picture to be sorted of preset threshold
It is considered negative sample；Obtain the classification results of picture of all categories.Wherein, threshold value method of determination can be with are as follows: utilizes depth convolution
Neural network to samples pictures extract feature vector, according to samples pictures feature space distribution；Estimated based on kernel function density
The band Density Estimator function of the total space based on each group variety is calculated in calculating method, from the band Density Estimator letter of the total space
Number obtain corresponding preset threshold of all categories, according to band Density Estimator function, calculate feature space of all categories respectively to point
The density of the corresponding point of class picture.
It is corresponding in feature space to the feature vector of a picture using kernel density estimation algorithm in the present embodiment
The density of point analyzed, each picture to be sorted that the density of point corresponding in feature space is less than preset threshold is considered
Negative sample is excluded, and the accuracy and precision of picture classification are improved.
In one embodiment, according to the feature vector of each picture to be sorted feature space distribution situation, to feature
Space carries out finite field division, obtains corresponding finite field of all categories, comprising: by the feature vector of picture to be sorted in feature sky
Between distribution, input the obtained one-class support vector machines disaggregated model of training in advance；By one-class support vector machines disaggregated model pair
Feature space carries out finite field of all categories and divides, and obtains finite field of all categories.
Wherein, support vector machines (SVM goes back support vector network) is supervised learning mould related to relevant learning algorithm
Type can analyze data, recognition mode, for classification and regression analysis.One group of training sample is given, each label is two
Class, a SVM training algorithm establish a model, distribute new example as a kind of or other classes, become non-probability two
First linear classification.Each classification is trained using training sample set in the distribution of feature space and obtains one-class support vector machines
Disaggregated model.One-class support vector machines disaggregated model according to the feature vector of the picture to be sorted of input feature space point
Cloth, the feature vector for treating category images carry out finite field of all categories in the distribution of feature space and divide, obtain of all categories
Finite field, so that positive sample (positive sample refers to Target Photo) is fallen in finite field, (negative sample refers to non-targeted negative sample
Picture) it falls in outside finite field, negative sample outside finite field is excluded using finite field set, obtains picture classification result.
In the present embodiment, by the obtained one-class support vector machines disaggregated model of training to treat the feature of category images to
The distribution measured in feature space carries out finite field division of all categories, to be carried out with finite field set to negative sample outside finite field
It excludes, improves the accuracy and precision of picture classification.
In one embodiment, according to the feature vector of each picture to be sorted feature space distribution situation, to feature
Space carries out finite field division, obtains corresponding finite field of all categories, comprising: is based on nearest neighbor algorithm, carries out to feature space each
The finite field of classification divides, and obtains finite field of all categories.
Wherein, by nearest neighbor algorithm, (so-called k nearest neighbor algorithm is a given training dataset, real to new input
Example finds the K example closest with the example in training data concentration, and the majority of this K example belongs to some class, just this
Exemplary classes are inputted into this class) finite field division (as shown in Figure 6) is carried out to the feature space of the picture, so that positive sample
This (positive sample refers to Target Photo) falls in finite field, and negative sample (negative sample refers to non-targeted picture) falls in finite field
Outside, negative sample outside finite field is excluded using finite field set, obtains picture classification result.
In the present embodiment, the feature vector for treating category images is carried out respectively in the distribution of feature space by nearest neighbor algorithm
The finite field of classification divides, to be excluded with finite field set to negative sample outside finite field, improves the standard of picture classification
Exactness and precision.
In one embodiment, a kind of picture classification method extracts institute's classification in need using depth convolutional neural networks
The high dimensional feature vector extracted is carried out dimensionality reduction using t-SNE algorithm, by high dimensional feature vector by the feature vector of picture
Dimensionality reduction shows to two-dimensional space, and on two-dimensional space figure, by two-dimensional space data determine finite field division mode and
Pre-determined distance that finite field divides, threshold value etc., wherein set Softmax for the loss function of depth convolutional neural networks
The cum rights of Loss and Center Loss and, using Softmax in the height of different classes of easy separability and Center Loss
Poly- property realizes the extraction of feature vector, and being abstracted as picture to be sorted in higher dimensional space has " separating between convergence, class in class " special
The feature vector of property.
Distribution based on each picture to be sorted in feature space carries out finite field division to feature space, so that positive sample
Originally it falls in finite field, negative sample is fallen in outside finite field, is excluded using finite field set to negative sample.Wipe-out mode is such as
Under:
Based on the mode of all categories in feature space centre distance: being integrated into the center of feature space with positive sample of all categories
For the centre of sphere, certain distance is that radius obtains higher-dimension sphere set, and the sample fallen in set is considered positive sample, and it is outer to fall in set
Be considered negative sample.
Mode based on one-class support vector machines (One Class Support Vector Machine): to each class
Not, one-class support vector machines are trained, such sample is divided in the distribution of feature space using training sample set.
The mode of distribution density estimation based on kernel function: according to positive sample in the distribution of feature space, kernel function is utilized
Density estimation algorithm obtains the band Density Estimator function of the total space, and given threshold.For any sample, in feature sky
Between in corresponding point density be greater than threshold value when be considered positive sample, be considered negative sample less than threshold value.
Mode based on nearest neighbor algorithm (KNN): in feature space using nearest neighbor algorithms such as KNN, space is divided, root
Negative sample is determined according to division result.
Above-mentioned picture classification method, based on the mode of all categories in feature space centre distance, one-class support vector machines
The mode of mode, the mode of the distribution density of kernel function estimation or nearest neighbor algorithm (KNN) falls in positive sample in finite field, bears
Sample is fallen in outside finite field, is excluded using finite field set to negative sample, and the accuracy and precision of picture classification are improved.
Solve it is in need participate in classification picture be made of the different Target Photo of N kind and non-targeted picture, lead to picture classification
Accuracy and the low problem of precision.
It should be understood that although each step in the flow chart of Fig. 1 is successively shown according to the instruction of arrow, this
A little steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly state otherwise herein, these steps
It executes there is no the limitation of stringent sequence, these steps can execute in other order.Moreover, at least part in Fig. 1
Step may include that perhaps these sub-steps of multiple stages or stage are executed in synchronization to multiple sub-steps
It completes, but can execute at different times, the execution sequence in these sub-steps or stage, which is also not necessarily, successively to be carried out,
But it can be executed in turn or alternately at least part of the sub-step or stage of other steps or other steps.
In one embodiment, as shown in fig. 7, a kind of picture classifier, comprising: picture to be sorted obtains module 310,
For obtaining each picture to be sorted；Characteristic vector pickup module 320, for each picture input feature vector to be sorted to be extracted model,
Obtain the feature vector of each picture to be sorted；Feature space obtains module 330, for obtaining the feature vector of each picture to be sorted
In the distribution situation of feature space；Picture classification result determining module 340, for being carried out to feature space according to distribution situation
The finite field of classification divides, and obtains picture classification result.
In one embodiment, the picture classification result determining module includes: finite field division unit, for according to respectively to
The feature vector of category images carries out finite field division to feature space, it is of all categories right to obtain in the distribution situation of feature space
The finite field answered；Picture classification result determination unit, for obtaining picture classification knot according to each picture in determining each finite field
Fruit.
Above-mentioned picture classifier, picture to be sorted obtain module 310 and are mentioned by obtaining each picture to be sorted, feature vector
Each picture input feature vector to be sorted is extracted model by modulus block 320, obtains the feature vector of each picture to be sorted, feature space obtains
Modulus block 330 obtains distribution situation of the feature vector in feature space of each picture to be sorted, picture classification result determining module
340 according to distribution situation, and the finite field for carrying out classification to the feature space divides, and obtains picture classification result.By will be to
Category images carries out characteristic vector pickup, obtains the feature vector of each picture to be sorted in the distribution situation of feature space, is based on
The finite field that the distribution situation of feature space carries out classification divides, and non-targeted picture is removed from Target Photo, reduces non-mesh
Mark on a map influence of the piece to Target Photo nicety of grading, solve the picture in need for the participating in classification target figure different by N kind
Piece and non-targeted picture composition, lead to the accuracy of picture classification and the low problem of precision.
In one embodiment, a kind of computer equipment, including memory and processor are provided, which is stored with
Computer program, which performs the steps of when executing computer program obtains each picture to be sorted；By each figure to be sorted
Piece input feature vector extracts model, obtains the feature vector of each picture to be sorted；The feature vector of each picture to be sorted is obtained in spy
Levy the distribution situation in space；According to distribution situation, the finite field for carrying out classification to feature space is divided, and obtains picture classification knot
Fruit.
In one embodiment, it also performs the steps of when processor executes computer program according to each picture to be sorted
Feature vector feature space distribution situation, to feature space carry out finite field division, obtain of all categories corresponding limited
Domain；According to each picture in determining each finite field, picture classification result is obtained.
In one embodiment, it also performs the steps of when processor executes computer program according to picture to be sorted
Feature vector determines the center of all categories in feature space in the distribution situation of feature space；Using each center as ball
The heart, pre-determined distance are radius, are divided to feature space；According to each corresponding intracorporal picture of ball is located at, obtain of all categories
Picture classification result.
In one embodiment, it is also performed the steps of when processor executes computer program and obtains all kinds of another characteristics
The band Density Estimator function and threshold value in space；According to band Density Estimator function, calculate feature space of all categories respectively to
The density of the corresponding point of category images；Finite field division is carried out to the feature space according to density and threshold value；It is big according to density
In the picture of threshold value, the classification results of picture of all categories are obtained.
In one embodiment, it also performs the steps of when processor executes computer program by the spy of picture to be sorted
Vector is levied in the distribution of feature space, inputs the one-class support vector machines disaggregated model that training obtains in advance；From single class support to
Amount machine disaggregated model carries out finite field of all categories to feature space and divides, and obtains finite field of all categories.
In one embodiment, it also performs the steps of when processor executes computer program based on nearest neighbor algorithm, to spy
Sign space carries out finite field of all categories and divides, and obtains finite field of all categories.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program performs the steps of when being executed by processor obtains each picture to be sorted；Each picture input feature vector to be sorted is extracted
Model obtains the feature vector of each picture to be sorted；Obtain distribution feelings of the feature vector in feature space of each picture to be sorted
Condition；According to distribution situation, the finite field for carrying out classification to feature space is divided, and obtains picture classification result.
In one embodiment, it is also performed the steps of when computer program is executed by processor according to each figure to be sorted
The feature vector of piece carries out finite field division to feature space in the distribution situation of feature space, obtains of all categories corresponding having
Confinement；According to each picture in determining each finite field, picture classification result is obtained.
In one embodiment, it also performs the steps of when computer program is executed by processor according to picture to be sorted
Feature vector in the distribution situation of feature space, determine the center of all categories in feature space；It is with each center
The centre of sphere, pre-determined distance are radius, are divided to feature space；According to each corresponding intracorporal picture of ball is located at, obtain all kinds of
Other picture classification result.
In one embodiment, it is also performed the steps of when computer program is executed by processor and obtains spy of all categories
Levy the band Density Estimator function and threshold value in space；According to band Density Estimator function, each of feature space of all categories is calculated
The density of the corresponding point of picture to be sorted；Finite field division is carried out to the feature space according to density and threshold value；According to density
Greater than the picture of threshold value, the classification results of picture of all categories are obtained.
In one embodiment, it is also performed the steps of when computer program is executed by processor by picture to be sorted
Feature vector inputs the one-class support vector machines disaggregated model that training obtains in advance in the distribution of feature space；It is supported by single class
Vector machine disaggregated model carries out finite field of all categories to feature space and divides, and obtains finite field of all categories.
In one embodiment, it is also performed the steps of when computer program is executed by processor based on nearest neighbor algorithm, it is right
Feature space carries out finite field of all categories and divides, and obtains finite field of all categories.
Fig. 8 shows the internal structure chart of computer equipment in one embodiment.As shown in figure 8, the computer equipment packet
Including the computer equipment includes processor, memory, network interface, input unit and the display screen connected by system bus.
Wherein, memory includes non-volatile memory medium and built-in storage.The non-volatile memory medium of the computer equipment stores
There is operating system, can also be stored with computer program, when which is executed by processor, may make processor realization figure
Piece classification method.Computer program can also be stored in the built-in storage, when which is executed by processor, may make
Processor executes picture classification method.The display screen of computer equipment can be liquid crystal display or electric ink display screen,
The input unit of computer equipment can be the touch layer covered on display screen, be also possible to be arranged on computer equipment shell
Key, trace ball or Trackpad can also be external keyboard, Trackpad or mouse etc..
It will be understood by those skilled in the art that structure shown in Fig. 8, only part relevant to application scheme is tied
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific computer equipment
It may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a non-volatile computer and can be read
In storage medium, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, provided herein
Each embodiment used in any reference to memory, storage, database or other media, may each comprise non-volatile
And/or volatile memory.Nonvolatile memory may include that read-only memory (ROM), programming ROM (PROM), electricity can be compiled
Journey ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include random access memory
(RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM
(SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
The limitation to claim therefore cannot be interpreted as.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of picture classification method characterized by comprising
Obtain each picture to be sorted；
Each picture input feature vector to be sorted is extracted into model, obtains the feature vector of each picture to be sorted；
Obtain distribution situation of the feature vector in feature space of each picture to be sorted；
According to the distribution situation, the finite field for carrying out classification to the feature space is divided, and obtains picture classification result.
2. picture classification method according to claim 1, which is characterized in that it is described according to the distribution situation, to described
The finite field that feature space carries out classification divides, and the step of obtaining picture classification result includes:
According to the feature vector of each picture to be sorted in the distribution situation of feature space, the feature space is carried out limited
Domain divides, and obtains corresponding finite field of all categories；
According to each picture in determining each finite field, picture classification result is obtained.
3. picture classification method according to claim 2, which is characterized in that according to the feature of each picture to be sorted to
The distribution situation in feature space is measured, finite field division is carried out to the feature space, obtains corresponding finite field of all categories, is wrapped
It includes:
According to the feature vector of picture to be sorted in the distribution situation of feature space, the centre bit of all categories in feature space is determined
It sets；
Using each center as the centre of sphere, pre-determined distance is radius, is divided to the feature space；
It is described according to each picture in determining each finite field, obtain picture classification result, comprising: according to be located at each corresponding sphere
Interior picture obtains picture classification result of all categories.
4. picture classification method according to claim 2, which is characterized in that according to the feature of each picture to be sorted to
The distribution situation in feature space is measured, finite field division is carried out to the feature space, obtains corresponding finite field of all categories, is wrapped
It includes:
Obtain the band Density Estimator function and threshold value of feature space of all categories；
According to the band Density Estimator function, the corresponding point of each picture to be sorted of the feature space of all categories is calculated
Density；
Finite field division is carried out to the feature space according to the density and the threshold value；
It is described according to each picture in determining each finite field, obtain picture classification result, comprising: according to density be greater than the threshold value
Picture, obtain the classification results of picture of all categories.
5. picture classification method according to claim 2, which is characterized in that according to the feature of each picture to be sorted to
The distribution situation in feature space is measured, finite field division is carried out to the feature space, obtains corresponding finite field of all categories, is wrapped
It includes:
By the feature vector of the picture to be sorted in the distribution of feature space, single class supporting vector that training obtains in advance is inputted
Machine disaggregated model；
Finite field of all categories is carried out to the feature space by the one-class support vector machines disaggregated model to divide, and is obtained all kinds of
Other finite field.
6. picture classification method according to claim 2, which is characterized in that according to the feature of each picture to be sorted to
The distribution situation in feature space is measured, finite field division is carried out to the feature space, obtains corresponding finite field of all categories, is wrapped
It includes:
Based on nearest neighbor algorithm, finite field of all categories is carried out to the feature space and is divided, finite field of all categories is obtained.
7. a kind of picture classifier characterized by comprising
Picture to be sorted obtains module, for obtaining each picture to be sorted；
Characteristic vector pickup module obtains each picture to be sorted for each picture input feature vector to be sorted to be extracted model
Feature vector；
Feature space obtains module, for obtaining distribution situation of the feature vector in feature space of each picture to be sorted；
Picture classification result determining module, for carrying out the finite field of classification to the feature space according to the distribution situation
It divides, obtains picture classification result.
8. picture classifier according to claim 2, which is characterized in that the picture classification result determining module packet
It includes:
Finite field division unit, for according to the feature vector of each picture to be sorted feature space distribution situation, it is right
The feature space carries out finite field division, obtains corresponding finite field of all categories；
Picture classification result determination unit, for obtaining picture classification result according to each picture in determining each finite field.
9. a kind of computer readable storage medium, be stored with computer program makes when the computer program is executed by processor
The processor is obtained to execute as described in any one of claims 1 to 6 the step of picture classification method.
10. a kind of computer equipment, including memory and processor, the memory is stored with computer program, the calculating
When machine program is executed by the processor, so that the processor executes the picture classification as described in any one of claims 1 to 6
The step of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811183703.7A CN109460777B (en) | 2018-10-11 | 2018-10-11 | Picture classification method and device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811183703.7A CN109460777B (en) | 2018-10-11 | 2018-10-11 | Picture classification method and device and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109460777A true CN109460777A (en) | 2019-03-12 |
| CN109460777B CN109460777B (en) | 2021-09-07 |
Family
ID=65607500
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811183703.7A Active CN109460777B (en) | 2018-10-11 | 2018-10-11 | Picture classification method and device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460777B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222724A (en) * | 2019-05-15 | 2019-09-10 | 平安科技（深圳）有限公司 | A kind of picture example detection method, apparatus, computer equipment and storage medium |
| CN111242230A (en) * | 2020-01-17 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Image processing method and image classification model training method based on artificial intelligence |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| CN103996047A (en) * | 2014-03-04 | 2014-08-20 | 西安电子科技大学 | Hyperspectral image classification method based on compression spectrum clustering integration |
| CN105488509A (en) * | 2015-11-19 | 2016-04-13 | Tcl集团股份有限公司 | Image clustering method and system based on local chromatic features |
| CN106650660A (en) * | 2016-12-19 | 2017-05-10 | 深圳市华尊科技股份有限公司 | Vehicle type recognition method and terminal |
| US9904866B1 (en) * | 2012-06-21 | 2018-02-27 | Amazon Technologies, Inc. | Architectures for object recognition |
| CN108446729A (en) * | 2018-03-13 | 2018-08-24 | 天津工业大学 | Egg embryo classification method based on convolutional neural networks |
| CN108509862A (en) * | 2018-03-09 | 2018-09-07 | 华南理工大学 | Anti- angle and the fast human face recognition for blocking interference |
-
2018
- 2018-10-11 CN CN201811183703.7A patent/CN109460777B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| US9904866B1 (en) * | 2012-06-21 | 2018-02-27 | Amazon Technologies, Inc. | Architectures for object recognition |
| CN103996047A (en) * | 2014-03-04 | 2014-08-20 | 西安电子科技大学 | Hyperspectral image classification method based on compression spectrum clustering integration |
| CN105488509A (en) * | 2015-11-19 | 2016-04-13 | Tcl集团股份有限公司 | Image clustering method and system based on local chromatic features |
| CN106650660A (en) * | 2016-12-19 | 2017-05-10 | 深圳市华尊科技股份有限公司 | Vehicle type recognition method and terminal |
| CN108509862A (en) * | 2018-03-09 | 2018-09-07 | 华南理工大学 | Anti- angle and the fast human face recognition for blocking interference |
| CN108446729A (en) * | 2018-03-13 | 2018-08-24 | 天津工业大学 | Egg embryo classification method based on convolutional neural networks |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222724A (en) * | 2019-05-15 | 2019-09-10 | 平安科技（深圳）有限公司 | A kind of picture example detection method, apparatus, computer equipment and storage medium |
| WO2020228179A1 (en) * | 2019-05-15 | 2020-11-19 | 平安科技（深圳）有限公司 | Picture instance detection method and apparatus, computer device, and storage medium |
| CN110222724B (en) * | 2019-05-15 | 2023-12-19 | 平安科技（深圳）有限公司 | Picture instance detection method and device, computer equipment and storage medium |
| CN111242230A (en) * | 2020-01-17 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Image processing method and image classification model training method based on artificial intelligence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109460777B (en) | 2021-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020253629A1 (en) | Detection model training method and apparatus, computer device, and storage medium | |
| Zhang et al. | Detection of co-salient objects by looking deep and wide | |
| Guo et al. | A fast and refined cancer regions segmentation framework in whole-slide breast pathological images | |
| JP6862579B2 (en) | Acquisition of image features | |
| Lu et al. | Dense and sparse reconstruction error based saliency descriptor | |
| Springenberg et al. | Improving deep neural networks with probabilistic maxout units | |
| CN108647588A (en) | Goods categories recognition methods, device, computer equipment and storage medium | |
| CN113762138B (en) | Identification method, device, computer equipment and storage medium for fake face pictures | |
| CN112464865A (en) | Facial expression recognition method based on pixel and geometric mixed features | |
| CN110427970A (en) | Image classification method, device, computer equipment and storage medium | |
| CN113570029A (en) | Method for obtaining neural network model, image processing method and device | |
| CN110889450B (en) | Super-parameter tuning and model construction method and device | |
| CN112036395A (en) | Text classification identification method and device based on target detection | |
| CN110210543B (en) | Image classification system, method, apparatus and storage medium | |
| Bai et al. | Weakly supervised object localization via transformer with implicit spatial calibration | |
| Acharya et al. | Deep convolutional network for breast cancer classification: enhanced loss function (ELF) | |
| CN111292377A (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN109816030A (en) | A kind of image classification method and device based on limited Boltzmann machine | |
| CN109460777A (en) | Picture classification method, device and computer readable storage medium | |
| CN111860056B (en) | Blink-based living body detection method, blink-based living body detection device, readable storage medium and blink-based living body detection equipment | |
| Albert et al. | Embedding contrastive unsupervised features to cluster in-and out-of-distribution noise in corrupted image datasets | |
| CN111694954A (en) | Image classification method and device and electronic equipment | |
| CN113096080B (en) | Image analysis method and system | |
| Dornaika et al. | Object-centric contour-aware data augmentation using superpixels of varying granularity | |
| CN116171462A (en) | Object sequence identification method, network training method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |