CN110288045A - A kind of semantic visual dictionary optimization method based on Pearson correlation coefficient - Google Patents
A kind of semantic visual dictionary optimization method based on Pearson correlation coefficient Download PDFInfo
- Publication number
- CN110288045A CN110288045A CN201910586981.5A CN201910586981A CN110288045A CN 110288045 A CN110288045 A CN 110288045A CN 201910586981 A CN201910586981 A CN 201910586981A CN 110288045 A CN110288045 A CN 110288045A
- Authority
- CN
- China
- Prior art keywords
- visual
- image
- behavioral characteristics
- phrase
- dictionary
- 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
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of semantic visual dictionary optimization method based on Pearson correlation coefficient extracts the behavioral characteristics of image by the low-level image feature that extracts the color of image, shape and texture and with SURF algorithm first；Using E2LSH clustering algorithm clusters the image behavioral characteristics of acquisition, extracts associated description visual phrase, constructs original visual dictionary；It introduces Pearson correlation coefficient and seeks the degree of correlation between behavioral characteristics and behavioral characteristics and between low-level image feature and behavioral characteristics to the optimization of original visual dictionary, obtain final semantic visual dictionary.The present invention problem complicated for visual phrase redundancy, calculating in semantic visual dictionary, carries out the classification performance that classification improves image to image, reduces the complexity of operation, shorten operation time, while improving the accuracy of classification.
Description
Technical field
The present invention relates to visual dictionary optimisation technique fields, more particularly to a kind of semanteme based on Pearson correlation coefficient
Visual dictionary optimization method.
Background technique
By the concern of lot of domestic and foreign scholar, market and social application are worth and also obtain Image Classfication Technology all the time
The affirmative of people is arrived.Unlike text information, image has bigger information content with video information and is more difficult to understand
Content, to make computer go to understand that piece image and one section of video also have biggish difficulty as the mankind.In reality
In life, the information that people go acquisition image to be included by eyes, the information that then will acquire is located by brain
Reason, human brain can get rid of those noises or useless information, retain corresponding image information.It is of the same race when running into next time
When the image of classification, brain will make corresponding reaction, accurately be identified to image.In artificial intelligence field, to one
Width image is identified and is classified, and will be trained accordingly by computer to it.In the training process, the first step is exactly
The image information that will acquire is input in computer；Then computer can carry out image information as human brain corresponding
Processing, when there is the image of classification of the same race to input again next time, computer will be identified it according to priori knowledge.So
And the complexity of human brain is that people is unthinkable, allowing computer to remove simulation human brain, there are also quite long roads to walk.
In image classification, visual phrase is obtained, must carry out the extraction of feature to image first, then by mentioning
The characteristics of image got carries out the processing, such as clustering algorithm, gauss hybrid models etc. of respective algorithms, finally can just obtain vision
Phrase.However excessive visual dictionary can make the time complexity of image classification increase, and classify to piece image
Perhaps identification inevitably has some picture noises and has an impact to classification or recognition efficiency.These picture noises can not only make
The nicety of grading of image declines, and the scale of visual dictionary can be made to become larger, and excessive visual dictionary will increase computer
The time cost of classification.
The present invention is mainly based upon machine vision and digital image processing techniques, by extract the color of image, shape and
Textural characteristics, and using accelerating robust features algorithm (SURF) to extract image behavioral characteristics, it is breathed out using accurate European local sensitivity
Uncommon clustering algorithm (E2LSH) behavioral characteristics of acquisition are clustered, Pearson correlation coefficient is introduced and optimizes visual dictionary.Structure after optimization
The semantic visual dictionary built carries out the classification performance that classification improves image to image, reduces the complexity of operation, shortens fortune
Evaluation time, while improving the accuracy of classification.
Summary of the invention
The semantic visual dictionary creation method based on Pearson correlation coefficient that the object of the present invention is to provide a kind of.The present invention
It is then to be led to by extracting color, shape and the textural characteristics of image and extract with SURF algorithm the behavioral characteristics of image first
Cross E2LSH clustering algorithm clusters the image behavioral characteristics of acquisition, associated description visual phrase is extracted, to construct original visual
Dictionary, reference Pearson correlation coefficient optimize visual dictionary.
A kind of semantic visual dictionary creation method based on Pearson correlation coefficient, it is characterised in that the following steps are included:
Step A, image behavioral characteristics R={ r is extracted using acceleration robust features algorithm1, r2..., ri..., rN-1, rN,
The characteristics of the underlying image collection S={ s based on color of image, shape and texture is extracted simultaneously1, s2, s3, s1, s2, s3Respectively image
Color, shape and textural characteristics；
Step B, it is clustered using image behavioral characteristics of the accurate European local sensitivity Hash clustering algorithm to acquisition, obtains original
Beginning visual dictionary；
Step C, degree of correlation size between visual phrase is sought in original visual dictionary using the Pearson correlation coefficient degree of correlation |
ρn|, given threshold ratio1If | ρn| < ratio1, this visual phrase is added to newly-built visual phrase set；
If step D, | ρn|≥ratio1, then by the corresponding behavioral characteristics of this visual phrase and characteristics of the underlying image --- face
Color, shape, textural characteristics seek the Pearson came degree of correlation again | ρ 'n|, given threshold ratio2If | ρ 'n| > ratio2, leave this
Visual phrase set is added in visual phrase, optimizes the visual phrase in original visual dictionary, constructs new semantic visual dictionary.
The step B includes:
Step B1, image behavioral characteristics are extracted with acceleration robust features algorithm and constitutes behavioral characteristics collection R={ r1, r2...,
ri..., rN-1, rN, wherein riIt is a behavioral characteristics of image, N is characterized Characteristic Number in collection R；
Step B2, from E2A position sensing function g is randomly selected in collection of functions G in LSH clustering algorithm；
Step B3, characteristics of image r in image behavioral characteristics collection R is sought with position sensing function giCorresponding k dimensional vector g
(ri)；
Step B4, r is calculatediMain cryptographic Hash h1(g(ri)) and time cryptographic Hash h2(g(ri)), all by primary and secondary cryptographic Hash in R
Identical behavioral characteristics are put into the same bucket bkIn, bkIt is visual phrase；
Step B5, the bucket b of all image behavioral characteristics is soughtkConstitute original visual dictionary Tg={ b1, b2..., bk...,
bZ}。
The step C includes:
Step C1, to original visual dictionary Tg={ b1, b2..., bk..., bZIn any one visual phrase bk, according to
Formula Calculate the vision
Related coefficient in phrase and original visual dictionary between other visual phrases | ρn|, obtain correlation matrix hk=[| ρ1
| ..., | ρn| ..., | ρz-1|]；
Step C2, to correlation matrix hkDescending sort；
Step C3, given threshold ratio1, to the correlation matrix h after descendingk, visual phrase is searched for, if | ρn| <
ratio1, then the visual phrase is added to newly-built visual phrase set Bg={ b1, b2..., bk..., bMIn.
Given threshold ratio described in step C1Value be (0.6,0.7).
Given threshold ratio described in step D2Value be (0.5,0.7).
Compared with prior art, the present invention considers the behavioral characteristics that image is considered using SURF algorithm, utilizes Pierre
Inferior related coefficient has measured the degree of correlation size between behavioral characteristics, calculates the dimension that image indicates to reduce, more can be accurate
Indicate the space distribution information of characteristics of image；The low-level image feature of image and behavioral characteristics are used into Pearson correlation coefficient degree simultaneously
Amount, avoids being left out some important features, and dimension is small but describes more accurately semantic visual dictionary for building, solves tradition
Semantic gap and the problem of redundancy in semantic dictionary, the visual dictionary after optimization can reduce the complexity of subsequent image sort operation
Degree shortens operation time, while improving the accuracy of classification.
Detailed description of the invention
Fig. 1 is the semantic visual dictionary creation method flow diagram based on Pearson correlation coefficient.
Specific embodiment
The following further describes the present invention with reference to the drawings.
1, assume that training image collection is D=[d1, d2..., di..., dN], wherein diIndicate the i-th width image.
2, image behavioral characteristics collection R={ r is extracted using SURF algorithm1, r2..., ri..., rN-1, rN, wherein riIt is figure
One behavioral characteristics of picture, N are Characteristic Number in behavioral characteristics collection R；
3, it is clustered using image behavioral characteristics of the accurate European local sensitivity Hash clustering algorithm to acquisition, generates Hash table
Tg={ b1, b2..., bk..., bZ, wherein bkK-th barrel is indicated in Hash table, and Z indicates the total number of bucket in Hash table, Hash
Table TgComplete a particular division to image behavioral characteristics, Hash table Tg={ b1, b2..., bk..., bZIt is exactly original visual
Dictionary.Specific step is as follows:
A, from E2A position sensing function g is randomly selected in LSH clustering algorithm in collection of functions G；
B, behavioral characteristics r in image behavioral characteristics collection R is sought with position sensing function giCorresponding k dimensional vector g (ri)；
C, r is calculatediMain cryptographic Hash h1(g(ri)) and time cryptographic Hash h2(g(ri)), primary and secondary cryptographic Hash in R is all identical
Behavioral characteristics are put into the same bucket bkIn, bkIt is visual phrase；
D, the bucket b of all image behavioral characteristics is soughtkConstitute original visual dictionary Tg={ b1, b2..., bk..., bz}。
4, sentenced using the Pearson correlation coefficient degree of correlation as degree of correlation size between visual phrase in original visual dictionary
Determine mode, takes original visual dictionary TgIn any one visual phrase bk, seek in the visual phrase and original visual dictionary
Related coefficient between other visual phrases, in Pearson correlation coefficient formula (1),
Wherein, the meaning that ρ (X, Y) is represented is the linearly related power of two different image behavioral characteristics vector X and Y
Degree, wherein indicating two feature vector nonlinear correlations if ρ (X, Y)=0 comprising -1≤ρ (X, Y)≤1, ρ's (X, Y) is exhausted
Show that value, correlation is stronger more greatly, i.e., the degree of correlation is bigger.Set a threshold value ratio1, as judgment basis, obtain correlation
Other visual phrases smaller than threshold value are spent, visual phrase set is created.Specific step is as follows:
A, to original visual dictionary Tg={ b1, b2..., bk..., bZIn any one visual phrase bk, according to formula
(1) related coefficient in this feature and visual phrase set between other local features is calculated | ρn|, obtain correlation matrix hk
=[| ρ1| ..., | ρn| ..., | ρZ-1|]；
B, to correlation matrix hkDescending sort；
C, given threshold ratio1=0.65, to the correlation matrix h after descendingk, visual phrase set is searched for, if | ρn
| < ratio1, then the visual phrase is added to newly-built visual phrase set Bg={ b1, b2..., bk..., bMIn；
5, to avoid leaving out some important features, if | ρn|≥ratio1, then the corresponding dynamic of this visual phrase is special
Levy and characteristics of the underlying image --- color, shape, textural characteristics seek the Pearson came degree of correlation again | ρ 'n|, given threshold ratio2,
If | ρ 'n| > ratio2, leave this visual phrase and visual phrase set be added, optimize the visual phrase in original visual dictionary, structure
Build new semantic visual dictionary.Specific step is as follows:
A, characteristics of the underlying image collection S={ s is found out1, s2, s3, s1, s2, s3Respectively color of image, shape and texture are special
Sign calculates riWith the Pearson correlation coefficient between low-level image feature collection S | ρ 'n|=[| ρ '1|, | ρ '2|, | ρ '3|]；
B, given threshold ratio2=0.6, if | ρ 'n| > ratio2, then visual phrase set B is added in the visual phraseg
In, to obtain the semantic visual dictionary after final optimization pass.
Claims (5)
1. a kind of semantic visual dictionary creation method based on Pearson correlation coefficient, it is characterised in that the following steps are included:
Step A, image behavioral characteristics R={ r is extracted using acceleration robust features algorithm1, r2..., ri..., rN-1, rN, it mentions simultaneously
Take the characteristics of the underlying image collection S={ s based on color of image, shape and texture1, s2, s3, s1, s2, s3Respectively color of image,
Shape and textural characteristics；
Step B, it is clustered using image behavioral characteristics of the accurate European local sensitivity Hash clustering algorithm to acquisition, obtains original view
Feel dictionary；
Step C, degree of correlation size between visual phrase is sought in original visual dictionary using the Pearson correlation coefficient degree of correlation | ρn|,
Given threshold ratio1If | ρn| < ratio1, this visual phrase is added to newly-built visual phrase set；
If step D, | ρn|≥ratio1, then by the corresponding behavioral characteristics of this visual phrase and characteristics of the underlying image --- color, shape
Shape, textural characteristics seek the Pearson came degree of correlation again | ρ 'n|, given threshold ratio2If | ρ 'n| > ratio2, then this vision is short
Visual phrase set is added in language, optimizes the visual phrase in original visual dictionary, constructs new semantic visual dictionary.
2. a kind of semantic visual dictionary optimization method based on Pearson correlation coefficient according to claim 1, feature
It is, the step B includes:
Step B1, image behavioral characteristics are extracted with acceleration robust features algorithm and constitutes behavioral characteristics collection R={ r1, r2..., ri...,
rN-1, rN, wherein riIt is a behavioral characteristics of image, N is characterized Characteristic Number in collection R；
Step B2, from E2A position sensing function g is randomly selected in collection of functions G in LSH clustering algorithm；
Step B3, characteristics of image r in image behavioral characteristics collection R is sought with position sensing function giCorresponding k dimensional vector g (ri)；
Step B4, r is calculatediMain cryptographic Hash h1(g(ri)) and time cryptographic Hash h2(g(ri)), primary and secondary cryptographic Hash in R is all identical
Behavioral characteristics be put into the same bucket bkIn, bkIt is visual phrase；
Step B5, the bucket b of all image behavioral characteristics is soughtkConstitute original visual dictionary Tg={ b1, b2..., bk..., bZ}。
3. a kind of semantic visual dictionary optimization method based on Pearson correlation coefficient according to claim 2, feature
It is, the step C includes:
Step C1, to original visual dictionary Tg={ b1, b2..., bk..., bZIn any one visual phrase bk, according to formula Calculate the visual phrase
With the related coefficient between visual phrases other in original visual dictionary | ρn|, obtain correlation matrix hk=[| ρ1| ..., |
ρn| ..., | ρZ-1|]；
Step C2, to correlation matrix hkDescending sort；
Step C3, given threshold ratio1, to the correlation matrix h after descendingk, visual phrase is searched for, if | ρn| < ratio1,
The visual phrase is then added to newly-built visual phrase set Bg={ b1, b2..., bk..., bMIn.
4. a kind of semantic visual dictionary optimization method based on Pearson correlation coefficient according to claim 1, feature
It is, given threshold ratio described in step C1Value be (0.6,0.7).
5. a kind of semantic visual dictionary optimization method based on Pearson correlation coefficient according to claim 1, feature
It is, given threshold ratio described in step D2Value be (0.5,0.7).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910586981.5A CN110288045B (en) | 2019-07-02 | 2019-07-02 | Semantic visual dictionary optimization method based on Pearson correlation coefficient |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910586981.5A CN110288045B (en) | 2019-07-02 | 2019-07-02 | Semantic visual dictionary optimization method based on Pearson correlation coefficient |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110288045A true CN110288045A (en) | 2019-09-27 |
| CN110288045B CN110288045B (en) | 2023-03-24 |
Family
ID=68021541
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910586981.5A Active CN110288045B (en) | 2019-07-02 | 2019-07-02 | Semantic visual dictionary optimization method based on Pearson correlation coefficient |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110288045B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102693311A (en) * | 2012-05-28 | 2012-09-26 | 中国人民解放军信息工程大学 | Target retrieval method based on group of randomized visual vocabularies and context semantic information |
| US20150269191A1 (en) * | 2014-03-20 | 2015-09-24 | Beijing University Of Technology | Method for retrieving similar image based on visual saliencies and visual phrases |
| CN107480718A (en) * | 2017-08-17 | 2017-12-15 | 南京信息工程大学 | A kind of high-resolution remote sensing image sorting technique of view-based access control model bag of words |
-
2019
- 2019-07-02 CN CN201910586981.5A patent/CN110288045B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102693311A (en) * | 2012-05-28 | 2012-09-26 | 中国人民解放军信息工程大学 | Target retrieval method based on group of randomized visual vocabularies and context semantic information |
| US20150269191A1 (en) * | 2014-03-20 | 2015-09-24 | Beijing University Of Technology | Method for retrieving similar image based on visual saliencies and visual phrases |
| CN107480718A (en) * | 2017-08-17 | 2017-12-15 | 南京信息工程大学 | A kind of high-resolution remote sensing image sorting technique of view-based access control model bag of words |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110288045B (en) | 2023-03-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111523462B (en) | Video sequence expression recognition system and method based on self-attention enhanced CNN | |
| CN101315663B (en) | Nature scene image classification method based on area dormant semantic characteristic | |
| CN109101938B (en) | Multi-label age estimation method based on convolutional neural network | |
| CN110097060B (en) | Open set identification method for trunk image | |
| CN102385592B (en) | Image concept detection method and device | |
| CN110415071B (en) | Automobile competitive product comparison method based on viewpoint mining analysis | |
| CN110751027B (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| CN109886161A (en) | A kind of road traffic index identification method based on possibility cluster and convolutional neural networks | |
| CN106997379B (en) | Method for merging similar texts based on click volumes of image texts | |
| CN111325237B (en) | Image recognition method based on attention interaction mechanism | |
| CN110472652A (en) | A small amount of sample classification method based on semanteme guidance | |
| CN110414513A (en) | Vision significance detection method based on semantically enhancement convolutional neural networks | |
| CN112784929A (en) | Small sample image classification method and device based on double-element group expansion | |
| CN115309860B (en) | False news detection method based on pseudo twin network | |
| CN112883931A (en) | Real-time true and false motion judgment method based on long and short term memory network | |
| CN110765285A (en) | Multimedia information content control method and system based on visual characteristics | |
| CN111860278A (en) | Human behavior recognition algorithm based on deep learning | |
| CN115130538A (en) | Training method of text classification model, text processing method, equipment and medium | |
| Cosovic et al. | Classification methods in cultural heritage | |
| Zhang et al. | Cross-dataset learning for age estimation | |
| CN113779283B (en) | Fine-grained cross-media retrieval method with deep supervision and feature fusion | |
| CN114625908A (en) | Text expression package emotion analysis method and system based on multi-channel attention mechanism | |
| CN114662586A (en) | Method for detecting false information based on common attention multi-mode fusion mechanism | |
| CN112200260B (en) | Figure attribute identification method based on discarding loss function | |
| CN113535960A (en) | Text classification method, device and equipment |
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 |