CN104331717B - The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature - Google Patents
The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature Download PDFInfo
- Publication number
- CN104331717B CN104331717B CN201410693888.1A CN201410693888A CN104331717B CN 104331717 B CN104331717 B CN 104331717B CN 201410693888 A CN201410693888 A CN 201410693888A CN 104331717 B CN104331717 B CN 104331717B
- Authority
- CN
- China
- Prior art keywords
- image
- visual signature
- vision word
- coding
- vector
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the image classification method that a kind of integration characteristics dictionary structure and visual signature are encoded, comprise the following steps：Visual Feature Retrieval Process；Characteristics dictionary learns；Visual signature is encoded；Converge in the space of feature coding；Training and classification.The present invention can obtain more accurate image feature representation, lift the accuracy rate of image classification.In addition, by the way that the structural information in characteristics dictionary is incorporated into visual signature cataloged procedure, more being there is the image feature representation of identification, hence in so that the classification to image is more efficient.The present invention realizes efficient, accurate image classification, therefore with higher use value.
Description
Technical field
The present invention relates to image classification field, a kind of integration of Codebook Model (Bag-of-Words, BoW) is based particularly on
The image classification method that characteristics dictionary structure is encoded with visual signature
Background technology
With information technology constantly develop rapidly, every field daily all produce with surprising rapidity it is various types of
Data, including word, image, video, music etc..In colourful data message, image because its show dramatic, in
Appearance is enriched, contained much information, and storage is convenient with transmission, enjoys favor, and have become 21st century most important information
One of carrier.In particular with becoming increasingly popular for the mobile device with camera function such as camera, mobile phone, flat board, Yi Jishe
The rise of network is handed over, the mode that people obtain image is more and more, also further promotes view data sharp increase, quick and precisely
Search required image and efficiently manage but therefore become more and more difficult in ground.The highly desirable computer capacity of people helps the mankind,
The semanteme contained to large nuber of images in internet is analyzed, and fully understands the content expressed by image, so that more effectively
Ground is managed to image, classification annotation, or retrieval image interested.
Image classification has received academia and industry as one of topmost basic technology of computer understanding image
Jie Ge research institutions it is widely studied, be meter and at home and abroad in each authoritative journal and Important Academic meeting as important theme
One epochmaking research topic of calculation machine visual field.Image classification is with referring to image intelligent according to certain sorting criterion
Assign to one group of process having been defined in classification, including object identification, Scene Semantics classification, Activity recognition etc..Image classification is
Important technical through understanding as research image, semantic.Science researcher recognizes the important of problem above gradually
Property is simultaneously constantly analysed in depth.In recent years, Codebook Model is that image high-level semantic represents to bring new inspiration, using Codebook Model as pass
The image classification of key technology has achieved certain achievement, but still has many research points to be not yet related to, and still has huge breakthrough empty
Between.The research of image classification method based on Codebook Model, has become current manual's intelligence, computer vision, machine learning
With the focus of frontier nature in many crossing domains such as data mining, played an important role to actively pushing forward social informatization.In wound
While having made the social value that can not be substituted, the field still has many key technical problems not yet to solve, and still has many functions
Realize that needs are further perfect, therefore, how using Codebook Model, more effectively understand and describe image high-level semantic, with more
The research of image classification is neatly realized, is had far-reaching significance.
The content of the invention
Goal of the invention：The technical problems to be solved by the invention are to integrate spy there is provided one kind in view of the shortcomings of the prior art
The image classification method that dictionary structure is encoded with visual signature is levied, is regarded using the distributed intelligence auxiliary of vision word in characteristics dictionary
Feature coding is felt, so that coding result has more identification, so as to improve the accuracy rate of image classification.
In order to solve the above-mentioned technical problem, encoded the invention discloses a kind of integration characteristics dictionary structure and visual signature
Image classification method, is comprised the following steps：
Step 1, the visual signature of image is extracted：Local sampling is carried out to each image, one group of region unit is obtained, extracts every
The visual signature in block region, obtains the corresponding visual signature set of each image, claim all images visual signature set it is whole
Body is the visual signature collection of all images, is designated as set X；
Step 2, characteristics dictionary learns：Using set X as input, using characteristics dictionary learning method, obtain being had by one group
The characteristics dictionary of representational vision word composition；
Step 3, visual signature is encoded：Each visual signature of each image is expressed as the linear combination of vision word,
One coefficient of each vision word correspondence, this system number is called the coding of visual signature；
Step 4, the space of visual signature coding is converged：Input is encoded to all visual signatures of each image, is made
With statistical method, each image is expressed as a vector, the vector is exactly the image feature representation of correspondence image；
Step 5, the coding of each image step 4 obtained is trained and classified using disaggregated model as input,
Obtain classification results.
Step 1 specifically includes following steps：
Local sampling is carried out to each image I, intensive sampling is done by the way of unique step, some size identicals are obtained
Region unit, a visual signature is extracted to each region unit, obtains representing the localized mass one using Visual Feature Retrieval Process method
Visual signature, Visual Feature Retrieval Process method includes：Histograms of oriented gradients (Histogram of Oriented Gradient,
HOG), Scale invariant features transform (Scale-invariant feature transform, SIFT) etc..Obtain regarding for image I
Feel characteristic set LFSI, finally give the overall X=[x of the visual signature set of all images1,x2,…,xN]∈Rd×N, wherein,
D represents the dimension of visual signature, and its size is determined by Visual Feature Retrieval Process technology, and N represents the total of the visual signature of all images
Number, xiRepresent i-th of visual signature, 1~N of i values.
Step 2 specifically includes following steps：
Using set X as input, using characteristics dictionary learning method, obtain what one group of representative vision word was constituted
Characteristics dictionary, this feature dictionary is designated as：B=[b1,b2,…,bM]∈Rd×M, wherein M is the number of vision word；bjIt is one
Dimension d column vector, represents j-th of vision word, 1~M of j values.Conventional characteristics dictionary learning method includes：K-means,
K-SVD etc..
Step 3 specifically includes following steps：
This step is encoded to each visual signature in set X one by one, for visual signature xi, its cataloged procedure is as follows：
First, x is selected from characteristics dictionary BiP arest neighbors vision word, i.e., with visual signature xiDistance it is minimum
P vision word, the characteristics dictionary for remembering this p vision word composition is Bi, p values 1~M, i 1~N of value,.
Secondly, characteristics dictionary B is obtainediIn matrix D represented by the distance between each vision wordiWith computation vision feature xi
To characteristics dictionary BiEach vision word the column vector d that represents of distancei, 1~N of i values.Matrix DiM rows s row element be
BiThe distance between middle correspondence vision word, m, s=1,2 ..., p；diN-th of component dinRepresent visual signature xiWith BiIn n-th
The distance between individual vision word, n=1,2 ..., p.It is apart from calculation formula：
σ is a smoothing parameter, the decrease speed of control weight, σ>0.dist(xi,Bi)=[dist (xi,bi1),dist(xi,
bi2),…,dist(xi,bip)]T, bilRepresent Bi l-th of vision word, l=1,2 ..., p；Each componentRepresent visual signature xiWith vision word bilThe distance between；max(dist(xi,Bi)) represent vector
dist(xi,Bi) largest component so that diThe codomain of middle component for (0,1].Calculating a vision word and other visions
During the distance between word, also using same strategy.To accelerate DiSolving speed, disposably obtain each vision word in B
The distance between represent matrix D.Then DiIt is exactly D submatrix, different D can be obtained by direct index Di, i=1,
2,…,N。
3rd, with xi, di, Di, BiIt is input with two parameter lambdas and β, λ, β >=0, minimum following formula obtains xiIn BiOn
Coding
Constraints：
WhereinRepresent that the corresponding component of the vector of dot product, i.e., two is multiplied and obtain a new vector；Solution obtains xiIn this p
The coding result of individual vision word
Finally, to codingIn component sequence, obtain k maximum code coefficientAnd its corresponding k vision list
The characteristics dictionary that word is constitutedK=1,2 ..., p, then visual signature xiCoding ziBe in the vector of M dimension, vector withCorresponding component isRemaining component is all set to 0.
Step 5 specifically includes following steps：
The spatial statistical information of each visual signature in each image is considered, with three layers of spatial pyramid Matching Model
(Spatial Pyramid Matching, SPM), using the coding of piece image I all visual signatures as input, with reference to most
Converge technology greatly, then it is (2 that the spatial pyramid, which exports a dimension,0+22+24) * M vector, the vector is that I image is special
Levy expression.
Step 6 specifically includes following steps：
After the image feature representation of each image is obtained, it is possible to be used to them train and classify.By all images
The set that is constituted of image feature representation be divided into training set and test set two parts, training set is used for train classification models, uses
The model trained is classified to test set.Generally it is used as and is divided from SVMs (Support Vector Machine, SVM)
Class device model.
The present invention is directed to the Image Visual Feature coding method in image classification field, and the present invention has following feature：1)
The present invention not only allows for the relation between visual signature and vision word when being encoded to visual signature, it is also contemplated that vision
The influence that relationship between word is encoded to visual signature；2) the visual signature coding that the present invention is tried to achieve is an analytic solutions, is not required to
Iteration optimization function is wanted, therefore visual signature coding method of the present invention is quick.
Beneficial effect：The present invention has taken into full account this structural information of the distribution of vision word in characteristics dictionary, and this is believed
Cease the coding for visual signature so that the distribution for the vision word that the coding of visual signature can more reflect in characteristics dictionary.Cause
This, the image feature representation of image has very strong identification, so as to lift the accuracy rate of image classification.
Brief description of the drawings
The present invention is done with reference to the accompanying drawings and detailed description and further illustrated, of the invention is above-mentioned
And/or otherwise advantage will become apparent.
Fig. 1 is flow chart of the present invention.
Fig. 2 is Visual Feature Retrieval Process schematic diagram
Fig. 3 is the flow chart to a visual signature coding
Fig. 4 is three sheaf space pyramid structure schematic diagrames.
Embodiment
As shown in figure 1, the invention discloses the image classification side that a kind of integration characteristics dictionary structure and visual signature are encoded
Method, is comprised the following steps：
Step 1, the visual signature of image is extracted：Local sampling is carried out to each image, one group of region unit is obtained, extracts every
The visual signature in block region, obtains the corresponding visual signature set of each image, by the whole of the visual signature set of all images
Body is designated as set X；
Step 2, characteristics dictionary learns：Using set X as input, using characteristics dictionary learning method, obtain being had by one group
The characteristics dictionary of representational vision word composition；
Step 3, visual signature is encoded：Each image visual signature is expressed as to the linear combination of vision word, each regarded
Feel word one coefficient of correspondence, obtain visual signature code set；
Step 4, the space of visual signature coding is converged：Input is encoded to all visions spy of each image, used
Statistical method, a vector is expressed as by each image, and the vector is exactly the image feature representation of correspondence image；
Step 5, the coding of each image step 4 obtained is trained and classified using disaggregated model as input,
Obtain classification results.
1st, step 1 comprises the following steps：
As shown in Fig. 2 for piece image I, generally extracting some size phases from I by the way of unique step intensive sampling
Deng region unit, and to each region unit extract a visual signature, visual signature here is a d dimensional vector.Conventional
Visual Feature Retrieval Process method includes：Histograms of oriented gradients (Histogram of Oriented Gradient, HOG), yardstick
Invariant features conversion (Scale-invariant feature transform, SIFT) etc..Finally give the vision of all images
Overall X=[the x of characteristic set1,x2,…,xN]∈Rd×N, wherein, d represents the dimension of visual signature, and N represents regarding for all images
Feel the sum of feature, xiRepresent i-th of visual signature, 1~N of i values.X is used for step 2 as input, to learn to obtain spy
Levy dictionary.
2nd, step 2 comprises the following steps：
In this step using set X as input, what the vision word for obtaining M d dimension using characteristics dictionary learning method was constituted
Characteristics dictionary B=[b1,b2,…,bM]∈Rd×M, wherein M is the number of vision word；bjIt is dimension d column vector, represents
J-th of vision word, 1~M of j values.By taking k-means methods as an example, set X is gathered for M class, each class using k-means
Center is exactly a vision word.
3rd, step 3 comprises the following steps：
This step is encoded to each visual signature in set X one by one.
Flow chart as shown in Figure 3 describes the cataloged procedure of a visual signature, for visual signature xi, choose vision
Feature xiThe characteristics dictionary B obtained by step 2 in p arest neighbors vision word, i.e., with visual signature xiDistance it is minimum
P vision word, p 1~M of value, the characteristics dictionary for remembering this p vision word composition is Bi, i 1~N of value obtain feature
Dictionary BiIn matrix D represented by the distance between each vision wordi, matrix DiM rows s row element be BiMiddle correspondence is regarded
Feel the distance between word, m, s=1,2 ..., p, then computation vision feature xiTo characteristics dictionary BiEach vision word distance
The column vector d of expressioni, diN-th of component dinRepresent visual signature xiWith BiIn the distance between n-th vision word, n=
1,2,…,p；With xi, di, Di, BiIt is input with two parameter lambdas and β, λ, β >=0, minimum following formula obtains xiIn BiOn coding
Constraints：
WhereinRepresent that the corresponding component of the vector of dot product, i.e., two is multiplied and obtain a new vector；Solution obtains xiIn this p
The coding result of individual vision wordFinally to codingIn component sequence, obtain k maximum code coefficientAnd its
Corresponding k vision wordK=1,2 ..., p, then xiCoding ziBe in the vector of M dimension, vector withIt is corresponding
Component isRemaining component is all set to 0.
Visual signature xiSpecific coding method on B is as follows：
Input：Image Visual Feature xi, characteristics dictionary B=[b1,b2,…,bM]∈Rd×M, M be B in vision word number with
And xiThe dimension of coding on B.xiArest neighbors word number p, parameter k, λ and β.
Cataloged procedure：
1) computation vision feature xiWith the vectorial d ' of the M dimensions represented by the distance of all vision wordsi；
2) to d '′Middle component is sorted in ascending order, and selects the set B that the p minimum vision word of distance is constitutedi, and correspondingly
Apart from di；
3) B is obtainediIn matrix D represented by the distance between each vision wordi；
4) coding is obtained according to following formula
Ψ=(xi1T-Bi)T(xi1T-Bi)
Θ=Ψ+λ * diag2(di)+βDi
α=- (1TΘ-11)
Wherein diag (di) represent diagoned vector be di diagonal matrix.1 expression component is all 1 column vector herein；
5) it is rightIn component sort in descending order, obtain k maximum code coefficientAnd its corresponding k vision word
The characteristics dictionary of compositionThen xiCoding ziBe in the vector of M dimension, vector withCorresponding component isRemaining point
Amount is all set to 0.Use formula zi=(1Tzi)-1ziNormalize zi；
Output：Visual signature xiCoding zi。
4th, step 4 comprises the following steps：
One three layers of spatial pyramid Matching Model is illustrated in figure 4, is compiled in all visual signatures for obtaining piece image
After code, using spatial pyramid Matching Model (Spatial Pyramid Matching, SPM), converge (Max with reference to maximum
Pooling) technology is converged in this space, and input is encoded to all visual signatures of piece image, obtains a vector, the vector
It is exactly the image feature representation of diagram picture.Concrete operations are：Using picture centre as origin, using different scale, recursively divide
Three layers of spatial pyramid Matching Model is used for some subregions, such as in Fig. 4, one has 20+22+24=21 sub-regions.It is right
In a-th of region, a values 1~21 converge the coding that technology obtains the region using maximum
The formula represents that this image region one has t visual signature；atRepresent the coding of h-th of visual signature in the region, h
1~t of value；z′aIt is a dimension and zahIdentical column vector, i.e. its dimension are M, and its q-th of component is matrixCorresponding row maximum, i.e.,1~M of q values.Further by z 'qReturn
One changes, for example, normalize to obtain z ' using 2 normsq=z 'q/||z′q||2.Finally the coding of all subregion is spliced successively, obtained
The image feature representation of the image.
5th, step 5 comprises the following steps：
After the image feature representation of all images is obtained, the image feature representation training of the image of training set is used as
Svm classifier model, reuses the SVM models trained and the image feature representation of the image as test set is classified.
Embodiment 1
The present embodiment includes following part：
1st, first by image down to the size no more than 300 × 300, and gray-scale map is converted into, is then adopted using intensive
Sample strategy, extracts the image block of 16 × 16 pixels from image, every 6 pixel decimations once, and one is extracted to each image block
SIFT feature.Therefore piece image may comprising hundreds and thousands of features, depending on tile size when extracting feature and
Every size.
2nd, all Image Visual Features are gathered for M cluster first by k- averages (k-means), each cluster center is just represented
One vision word.Set arest neighbors vision word number p, intensive neighbour's vision word number k, apart from smoothing parameter σ, canonical
Change parameter lambda and β.Each visual signature is encoded.
3rd, converge technology using space gold Matching Model and maximum, all visual signatures coding of each image is converged
For image feature representation of the vector as the image.And image is trained and classified using supporting vector machine model.
Embodiment 2
To the visual signature that image zooming-out dimension is 128, the size of characteristics dictionary and the quantity of vision word are set to
1024.P and k are set to 10 and 5 respectively.Other parameter settings also include：λ=10-4, β=10-4.Use 3 layers of space
Pyramid is matched and maximum converges technology.Obtain the image feature representation of 21504 dimensions of each image.Using being used as training set
The image feature representation training svm classifier model of image, and it is special to the image of the image as test set with the model trained
Presentation class is levied, final classification results are obtained.
The invention provides the image classification method that a kind of integration characteristics dictionary structure and visual signature are encoded, implement
The method and approach of the technical scheme are a lot, and described above is only the preferred embodiment of the present invention, it is noted that for this skill
For the those of ordinary skill in art field, under the premise without departing from the principles of the invention, some improvements and modifications can also be made,
These improvements and modifications also should be regarded as protection scope of the present invention.Each part being not known in the present embodiment can use existing
Technology is realized.
Claims (3)
1. the image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature, it is characterised in that including following step
Suddenly：
Step 1, the visual signature of image is extracted：Local sampling is carried out to each image, one group of region unit is obtained, every piece of area is extracted
The visual signature in domain, obtains the corresponding visual signature set of each image, claims the visual signature set of all images generally
The visual signature collection of all images, is designated as set X；
Step 2, characteristics dictionary learns：Using set X as input, using characteristics dictionary learning method, obtain that there is representative by one group
Property vision word composition characteristics dictionary；
Step 3, visual signature is encoded：Each visual signature of each image is expressed as the linear combination of vision word, each
Vision word one coefficient of correspondence, the coefficient is called the coding of visual signature；
Step 4, the space of visual signature coding is converged：Input is encoded to all visual signatures of each image, system is used
Meter method, a vector is expressed as by each image, and the vector is exactly the image feature representation of correspondence image；
Step 5, the coding of each image step 4 obtained is trained and classified using disaggregated model as input, is obtained
Classification results；
Step 1 comprises the following steps：
Local sampling is carried out for image I, sampling every time obtains a region unit, and each region unit extracts a visual signature,
Obtain image I visual signature set LFSI, finally give the visual signature set X=[x of all images1,x2,…,xN]∈Rd ×N, wherein, d represents the dimension of visual signature, and N represents the sum of the visual signature of all images, xiI-th of visual signature is represented,
1~N of i values；
Step 2 comprises the following steps：
Using set X as input, using characteristics dictionary learning method, the spy being made up of one group of representative vision word is obtained
Dictionary is levied, this feature dictionary is designated as：B=[b1,b2,…,bj,…,bM]∈Rd×M, wherein M is the number of vision word；bjIt is
One dimension d column vector, represents j-th of vision word, 1~M of j values；
Step 3 comprises the following steps：
For visual signature xi, choose visual signature xiThe characteristics dictionary B obtained by step 2 in p arest neighbors vision list
Word, i.e., with visual signature xiThe minimum p vision word of distance, p 1~M of value remember the feature that this p vision word is constituted
Dictionary is Bi, i 1~N of value obtain characteristics dictionary BiIn matrix D represented by the distance between each vision wordi, matrix Di's
The element of m rows s row is characterized dictionary BiIn the distance between m-th and s-th vision word, m, s=1,2 ..., p；Count again
Calculate visual signature xiTo characteristics dictionary BiEach vision word the column vector d that represents of distancei, diN-th of component dinExpression is regarded
Feel feature xiWith BiIn the distance between n-th vision word, n=1,2 ..., p, with xi, di, Di, BiIt is with β with two parameter lambdas
Input, λ, β >=0 minimizes following formula, obtains xiIn BiOn coding
Constraints：
WhereinRepresent that the corresponding component of the vector of dot product, i.e., two is multiplied and obtain a new vector；Solution obtains xiRegarded at this p
Feel the coding result of wordFinally to codingIn component sequence, obtain k maximum code coefficientAnd its it is corresponding
The characteristics dictionary that k vision word is constitutedK=1,2 ..., p, then visual signature xiCoding ziIt is the vector of a M dimension,
In vector withCorresponding component isRemaining component is all set to 0.
2. according to the method described in claim 1, it is characterised in that step 5 comprises the following steps：Matched using spatial pyramid
Model, a vector is merged into as the image feature representation of the image using the coding of all visual signatures of each image.
3. method according to claim 2, it is characterised in that step 6 comprises the following steps：The image for obtaining all images is special
Levy after the set for representing constituted, the set be divided into training set and test set two parts, training set is used for train classification models,
Test set is classified with the model trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410693888.1A CN104331717B (en) | 2014-11-26 | 2014-11-26 | The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410693888.1A CN104331717B (en) | 2014-11-26 | 2014-11-26 | The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104331717A CN104331717A (en) | 2015-02-04 |
| CN104331717B true CN104331717B (en) | 2017-10-17 |
Family
ID=52406438
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410693888.1A Active CN104331717B (en) | 2014-11-26 | 2014-11-26 | The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104331717B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105224942B (en) * | 2015-07-09 | 2020-02-04 | 华南农业大学 | RGB-D image classification method and system |
| CN105224619B (en) * | 2015-09-18 | 2018-06-05 | 中国科学院计算技术研究所 | A kind of spatial relationship matching process and system suitable for video/image local feature |
| CN105808757B (en) * | 2016-03-15 | 2018-12-25 | 浙江大学 | The Chinese herbal medicine picture retrieval method of BOW model based on multi-feature fusion |
| CN110598776A (en) * | 2019-09-03 | 2019-12-20 | 成都信息工程大学 | Image classification method based on intra-class visual mode sharing |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646256A (en) * | 2013-12-17 | 2014-03-19 | 上海电机学院 | Image characteristic sparse reconstruction based image classification method |
| CN103699436A (en) * | 2013-12-30 | 2014-04-02 | 西北工业大学 | Image coding method based on local linear constraint and global structural information |
-
2014
- 2014-11-26 CN CN201410693888.1A patent/CN104331717B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646256A (en) * | 2013-12-17 | 2014-03-19 | 上海电机学院 | Image characteristic sparse reconstruction based image classification method |
| CN103699436A (en) * | 2013-12-30 | 2014-04-02 | 西北工业大学 | Image coding method based on local linear constraint and global structural information |
Non-Patent Citations (2)
| Title |
|---|
| Locality-constrained Linear Coding for Image Classfication;Jinjun Wang 等;《Computer Vision and Pattem Recognition》;IEEE;20101231;第3360-3367页 * |
| Structurally Enhanced Incremental Neural Learning for Image Classification with Subgraph Extraction;Yu-Bin Yang 等;《International Journal of Neural Systems》;20140812;第24卷(第7期);第1450024-1 - 1450024-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104331717A (en) | 2015-02-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107122375B (en) | Image subject identification method based on image features | |
| CN111160350B (en) | Portrait segmentation method, model training method, device, medium and electronic equipment | |
| CN106022392B (en) | A kind of training method that deep neural network sample is accepted or rejected automatically | |
| CN107644235A (en) | Image automatic annotation method based on semi-supervised learning | |
| CN103116766B (en) | A kind of image classification method of encoding based on Increment Artificial Neural Network and subgraph | |
| CN108897778B (en) | Image annotation method based on multi-source big data analysis | |
| CN107943967A (en) | Algorithm of documents categorization based on multi-angle convolutional neural networks and Recognition with Recurrent Neural Network | |
| CN111931061B (en) | Label mapping method and device, computer equipment and storage medium | |
| CN107103326A (en) | The collaboration conspicuousness detection method clustered based on super-pixel | |
| CN113515632B (en) | Text classification method based on graph path knowledge extraction | |
| CN103116762A (en) | Image classification method based on self-modulated dictionary learning | |
| CN110825850B (en) | Natural language theme classification method and device | |
| CN104331717B (en) | The image classification method that a kind of integration characteristics dictionary structure is encoded with visual signature | |
| CN110210468A (en) | A kind of character recognition method based on the migration of convolutional neural networks Fusion Features | |
| CN109213853A (en) | A kind of Chinese community's question and answer cross-module state search method based on CCA algorithm | |
| CN105808757B (en) | The Chinese herbal medicine picture retrieval method of BOW model based on multi-feature fusion | |
| CN106960176A (en) | A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion | |
| Qi et al. | Personalized sketch-based image retrieval by convolutional neural network and deep transfer learning | |
| CN112507800A (en) | Pedestrian multi-attribute cooperative identification method based on channel attention mechanism and light convolutional neural network | |
| CN106227836B (en) | Unsupervised joint visual concept learning system and unsupervised joint visual concept learning method based on images and characters | |
| CN107577994A (en) | A kind of pedestrian based on deep learning, the identification of vehicle auxiliary product and search method | |
| CN110765960A (en) | Pedestrian re-identification method for adaptive multi-task deep learning | |
| CN115563327A (en) | Zero sample cross-modal retrieval method based on Transformer network selective distillation | |
| CN112541347A (en) | Machine reading understanding method based on pre-training model | |
| CN114398485B (en) | Expert portrait construction method and device based on multi-view fusion |
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 | ||
| GR01 | Patent grant |