CN108829711A - A kind of image search method based on multi-feature fusion - Google Patents
A kind of image search method based on multi-feature fusion Download PDFInfo
- Publication number
- CN108829711A CN108829711A CN201810418660.XA CN201810418660A CN108829711A CN 108829711 A CN108829711 A CN 108829711A CN 201810418660 A CN201810418660 A CN 201810418660A CN 108829711 A CN108829711 A CN 108829711A
- Authority
- CN
- China
- Prior art keywords
- image
- target image
- gray level
- value
- pixel
- 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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention proposes a kind of image search methods based on multi-feature fusion, and its step are as follows:Target image I is obtained, and calculates the characteristics of image of target image I:The color characteristic of target image I is extracted, and is stored in the color characteristic library of image;The shape feature of target image I is extracted, and is stored in the shape feature library of image;The textural characteristics of target image I are extracted, and are stored in the textural characteristics library of image;The similarity calculation of target image I and image data set obtain final search result.Beneficial effects of the present invention are as follows:It allows users to obtain similar house conceptual drawing picture according to retrieval image.This method is directed to the deficiency of existing single features, copes with the combination of the feature set of household industry, under actual household scene, can be improved single features effectiveness of retrieval, improves the insufficient problem of single features covering.It can be improved the search efficiency of family product image, save the time that user searches household scheme, improve the actual search experience of user.
Description
Technical field
The present invention relates to image retrieval technologies fields, particularly relate to a kind of image search method based on multi-feature fusion.
Background technique
Content-based image retrieval method refers to the information such as the visual signature having using image itself and spatial relationship,
The high dimensional feature vectors library for establishing image, is matched according to the high dimensional feature vectors of image, returns to the image retrieval knot of user
Fruit.For this method compared with text based image search method, the result of retrieval is more effective.
Characteristics of image can be divided into color characteristic, textural characteristics and shape feature three categories according to the division of logical type.
Wherein color characteristic is one of the characteristics of image being most widely used, and the function based on color classification can be provided for picture search
Energy;Textural characteristics focus on to describe the texture pattern in image block;And shape feature mainly describes the design feature in image.Scheming
As in search, characteristics of image is as bottom-up information, the retrieval of support target image I.
Image search method based on single features can have higher efficiency in some characteristic aspect, but in face
When different to complicated and changeable, classification natural scene, shortcoming is often just manifested.Therefore, a variety of characteristics of image are merged
Search method has urgent demand in practical implementation.
Under concrete application scene, a problem for merging a variety of characteristics of image is the selection of characteristics of image method, no
Same image characteristic extracting method is not quite similar to the form of expression of feature, eventually leads to and has difference to the understanding of image meaning,
To influence the image retrieval effect under application scenarios.
Summary of the invention
The present invention proposes a kind of image search method based on multi-feature fusion, solves and merges a variety of figures in the prior art
As the incomplete problem of characteristic key method.
The technical proposal of the invention is realized in this way:
A kind of image search method based on multi-feature fusion, method and step are as follows:
(1) target image I is obtained, and calculates the characteristics of image of target image I
1) color characteristic of target image I is extracted, and is stored in the color characteristic library of image
1. target image I is transformed into hsv color space according to standard handovers formula from RGB color;
2. will transition to the target image I in hsv color space according to formulaWithCarry out the
Tone H is quantified to 7 sections, brightness V and saturation degree S to quantify respectively to 3 sections by the quantization of one step;
3. the rgb value of target image I to be mapped to 63 kinds of color spaces of HSV according to formula L=9H+3S+V by quantization
In;
Piecemeal processing, every block of image power different according to how much impartings of its included information content are carried out to target image I
Weight, the color histogram H (I of every block of imagek) indicate, the corresponding weight w of every block of imagekIt indicates, then the piecemeal of whole image
Weighted color histogram isWherein n is the block count to the target image I,
4. the divided group color histogram to the whole image is normalized, and as target image I's
The color characteristic library of color characteristic deposit image;
2) shape feature of target image I is extracted, and is stored in the shape feature library of image
1. the original color image of target image I is subjected to edge enhancing using Domain Transform method, wherein
Parameter sigma_s=10, sigma_r=0.15;
2. the enhanced target image I in edge is converted to gray level image according to normalized form, the gray level image is adopted
It is zoomed in and out with bilinearity difference approach;Edge detection is carried out using canny boundary operator to the gray level image after scaling;
3. calculating the gradient-norm and gradient direction at each profile point of the gray level image:Using formulaTemplate to the gray level image carry out sobel operator filtering, obtain the gray level image
Horizontal gradientAnd vertical gradientAnd then obtain the gradient-norm of the gray level imageIt is described
The gradient direction of gray level image isResult according to above-mentioned edge detection obtains the gray level image
Gradient-norm and gradient direction at each profile point;
4. carrying out multiple dimensioned processing to the gray level image to divide the gray level image using pyramid dividing method
At L layers, each layer is divided into 2l(l=0 ..., L) block adds up certain at n-th piece of layer of the gray level image l of profile point
Statistical value of the gradient modulus value in one gradient direction section as the gradient direction section, traverses all profiles of the gray level image
Point and gradient direction section, the statistical gradient direction histogram of n-th piece of layer of the gray level image l are
5. the gradient orientation histogram of all image blocks of the gray level image is carried out splicing merging, complete gradient is obtained
Direction histogram
6. the complete gradient orientation histogram H (I) of the gray level image is normalized, obtaining dimension is
Shape eigenvectors, and the shape feature as target image I is stored in the shape feature library of image;
3) textural characteristics of target image I are extracted, and are stored in the textural characteristics library of image
1. calculating separately the roughness of each pixel of target image I, contrast and directionality;
2. by a channel of target image IThe square mean filter that size is 2*m is carried out, wherein m=1,2,3,4,
5, to obtain 5 different mean filter imagesIt is equal to calculate separately 5 differences
The level difference image of value filtering imageWithWherein
3. by the channelEach pixel (x, y) at 10 E are calculatedT,R(x, y) value, therefrom selects maximum value
As the roughness value at pixel (x, y), i.e.,Wherein m=1,2,3,4,
5;
4. in the channelThe mean value of statistical pixel (x, y) in the 7*7 window of pixel (x, y)VarianceIn the channelThe 7*7 window of pixel (x, y)
Four differences of statistical pixel (x, y) in mouthfulIt is then described
ChannelContrast value at pixel (x, y) is
5. by the channelAccording to formulaMask convolution obtain the channel's
Horizontal gradientAnd vertical gradientIt further calculates to obtain the channelDirectionality value at pixel (x, y)
6. roughness value, contrast value and the directionality value in tri- channels R, G, B, obtain at the pixel (x, y) of adding up
The unrelated roughness value in channel, contrast value and directionality value at the pixel (x, y);
7. by roughness, contrast and directionality all uniform quantizations of the target image I to g section, then the mesh
The value interval of logo image I roughness, contrast and directionality becomes [0, g-1];
8. joint roughness, contrast and directionality determine correspondence section of the pixel (x, y) in Texture similarityEach pixel is carried out on corresponding section tired
Add, obtains cumulative Texture similarity HT(IT), wherein HT(IT) dimension be g*g*g;
9. the Texture similarity H that the g*g*g is tieed upT(IT) be normalized, the texture for obtaining the target image I is special
Sign, is stored in the textural characteristics library of image;
(2) similarity calculation of target image I and image data set
1) a retrieval image Q is inputted, extracts its color characteristic X respectivelyC, shape feature XSWith textural characteristics XT, calculate XC
With characteristic Y each in color characteristic libraryCLinear kernel range lineCalculate XSWith shape spy
Levy each characteristic Y in librarySEuclidean distanceCalculate XTWith characteristic Y each in textural characteristics libraryTJSD
Distance
Wherein d is the dimension of character pair;
2) image two-by-two in image library is randomly selected to Qr1And Qr2, calculate the color characteristic distance of described image pairObtain sampling setFurther obtain sampling setSample average and sample standard deviation, repeat this mistake
Journey obtains the average value of sample average, as dCGaussian Profile mean μC, dCStandard deviation sigmaC;According to aforesaid operations step,
D can similarly be obtainedSGaussian Profile mean μS, standard deviation sigmaS, dTGaussian Profile mean μT, standard deviation sigmaT;FoundationRespectively by dC、dS、dTIt is transformed into standard gaussian distribution;
3) by above-mentioned three kinds of distance measurements dC、dS、dTIt merges to obtain d using the method for weightmerge=wCdC+wSdS+wTdT,wC
+wS+wT=1;
4) to calculated distance dmergeIt is ranked up, P data before taking, that is, P width image is as search result before taking.
Preferably, the gray level image scaling of step 2. is not more than 500 in step 2) in the step (1).
Three kinds of color of image, shape, texture category features respectively correspond the picture material attribute of mankind's difference perception.This
In scheme, the other characteristics of image of three types is individually calculated first, then three kinds of feature letters of fusion during similarity calculation
Breath, obtains comprehensive similarity calculation result.
The color characteristic for extracting image using color histogram first extracts image using laminated gradient direction histogram
Shape feature extracts the textural characteristics of image using tamura texture representation, finally uses the method for weight fusion by color, shape
Shape and Texture Feature Fusion are retrieved.
Beneficial effects of the present invention are:
It allows users to obtain similar house conceptual drawing picture according to retrieval image.This method is directed to existing single features
Deficiency, cope with household industry feature set combination, under actual household scene, can be improved single features retrieval effect
Rate improves the insufficient problem of single features covering.
It can be improved the search efficiency of family product image, save the time that user searches household scheme, improve user's
Actual search experience.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the uneven method of partition schematic diagram that the present invention carries out piecemeal processing to image.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment
A kind of image search method based on multi-feature fusion, specific step is as follows:
(1) target image I is obtained, and calculates the characteristics of image of target image I
1) color characteristic of target image I is extracted, and is stored in the color characteristic library of image
1. target image I is transformed into hsv color space according to standard handovers formula from RGB color;
2. will transition to the target image I in hsv color space according to formulaWithCarry out the
Tone H is quantified to 7 sections, brightness V and saturation degree S to quantify respectively to 3 sections by the quantization of one step;
3. the rgb value of target image I to be mapped to 63 kinds of color spaces of HSV according to formula L=9H+3S+V by quantization
In;
The information content that the color of different location provides in one block of image is different, and the information of a usual sub-picture is concentrated mainly on
Image centre, edge are often used as background, therefore carry out simple piecemeal processing to target image I, every block of image according to
The different weight of how much impartings of its included information content, using uneven method of partition as shown in Figure 1, a-quadrant position in Fig. 1
In picture centre, the main information of image is contained, assigns biggish weight, the image letter that B, C, D, E, F, G, H, I are included
It ceases less, assigns lesser weight;
The color histogram of every block of image H (Ik) indicate, the corresponding weight w of every block of imagekIt indicates, then whole image
Divided group color histogram beWherein n is the block count to the target image I,
4. the divided group color histogram to the whole image is normalized, and as target image I's
The color characteristic library of color characteristic deposit image;
2) shape feature of target image I is extracted, and is stored in the shape feature library of image
1. the original color image of target image I is subjected to edge enhancing using Domain Transform method, wherein
Parameter sigma_s=10, sigma_r=0.15;
2. the enhanced target image I in edge is converted to gray level image according to normalized form, the gray level image is adopted
It is zoomed in and out with bilinearity difference approach, scaling maximum value is 500;The edge canny is used to the gray level image after scaling
Operator carries out edge detection, and wherein parameter Low threshold is 46, and high threshold is 115, apertureSize=3;
3. calculating the gradient-norm and gradient direction at each profile point of the gray level image:Using formulaTemplate to the gray level image carry out sobel operator filtering, obtain the gray level image
Horizontal gradientAnd vertical gradientAnd then obtain the gradient-norm of the gray level imageIt is described
The gradient direction of gray level image isResult according to above-mentioned edge detection obtains the gray level image
Gradient-norm and gradient direction at each profile point, gradient direction range is 0~180 degree in the present embodiment, and gradient direction is equal
It is even to be quantified as KSA section, wherein KS=20;
4. carrying out multiple dimensioned processing to the gray level image to divide the gray level image using pyramid dividing method
At L layers, each layer is divided into 2l(l=0 ..., L) block adds up certain at n-th piece of layer of the gray level image l of profile point
Statistical value of the gradient modulus value in one gradient direction section as the gradient direction section, traverses all profiles of the gray level image
Point and gradient direction section, the statistical gradient direction histogram of n-th piece of layer of the gray level image l are
5. the gradient orientation histogram of all image blocks of the gray level image is carried out splicing merging, complete gradient is obtained
Direction histogram
6. the complete gradient orientation histogram H (I) of the gray level image is normalized, obtaining dimension is
Shape eigenvectors, and the shape feature as target image I is stored in the shape feature library of image;
3) textural characteristics of target image I are extracted, and are stored in the textural characteristics library of image
1. calculating separately the roughness of each pixel of target image I, contrast and directionality;
2. by a channel of target image IThe square mean filter that size is 2*m is carried out, wherein m=1,2,3,4,
5, to obtain 5 different mean filter imagesIt is equal to calculate separately 5 differences
The level difference image of value filtering imageWithWherein
3. by the channelEach pixel (x, y) at 10 E are calculatedT,R(x, y) value, therefrom selects maximum value
As the roughness value at pixel (x, y), i.e.,Wherein m=1,2,3,4,
5;
4. in the channelThe mean value of statistical pixel (x, y) in the 7*7 window of pixel (x, y)VarianceIn the channelThe 7*7 window of pixel (x, y)
Four differences of statistical pixel (x, y) in mouthfulIt is then described
ChannelContrast value at pixel (x, y) is
5. by the channelAccording to formulaMask convolution obtain the channel's
Horizontal gradientAnd vertical gradientIt further calculates to obtain the channelDirectionality value at pixel (x, y)
6. roughness value, contrast value and the directionality value in tri- channels R, G, B, obtain at the pixel (x, y) of adding up
The unrelated roughness value in channel, contrast value and directionality value at the pixel (x, y);
7. by roughness, contrast and directionality all uniform quantizations of the target image I to g section, then the mesh
The value interval of logo image I roughness, contrast and directionality becomes [0, g-1];
8. joint roughness, contrast and directionality determine correspondence section of the pixel (x, y) in Texture similarityEach pixel is carried out on corresponding section tired
Add, obtains cumulative Texture similarity HT(IT), wherein HT(IT) dimension be g*g*g;
9. the Texture similarity H that the g*g*g is tieed upT(IT) be normalized, the texture for obtaining the target image I is special
Sign, is stored in the textural characteristics library of image;
(2) similarity calculation of target image I and image data set
1) a retrieval image Q is inputted, extracts its color characteristic X respectivelyC, shape feature XSWith textural characteristics XT, calculate XC
With characteristic Y each in color characteristic libraryCLinear kernel range lineCalculate XSWith shape spy
Levy each characteristic Y in librarySEuclidean distanceCalculate XTWith characteristic Y each in textural characteristics libraryTJSD
Distance
Wherein d is the dimension of character pair;
2) image two-by-two in image library is randomly selected to Qr1And Qr2, calculate the color characteristic distance of described image pairObtain sampling setFurther obtain sampling setSample average and sample standard deviation, repeat this mistake
Journey obtains the average value of sample average, as dCGaussian Profile mean μC, dCStandard deviation sigmaC;According to aforesaid operations step,
D can similarly be obtainedSGaussian Profile mean μS, standard deviation sigmaS, dTGaussian Profile mean μT, standard deviation sigmaT;FoundationRespectively by dC、dS、dTIt is transformed into standard gaussian distribution;
3) by above-mentioned three kinds of distance measurements dC、dS、dTIt merges to obtain w using the method for weightC=wS=wT=1/3, this implementation
W in exampleC=wS=wT=1/3;
4) to calculated distance dmergeIt is ranked up, P data before taking, that is, P width image is as search result before taking.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (2)
1. a kind of image search method based on multi-feature fusion, which is characterized in that its method and step is as follows:
(1) target image I is obtained, and calculates the characteristics of image of target image I
1) color characteristic of target image I is extracted, and is stored in the color characteristic library of image
1. target image I is transformed into hsv color space according to standard handovers formula from RGB color;
2. will transition to the target image I in hsv color space according to formulaWithCarry out the first step
Tone H is quantified to 7 sections, brightness V and saturation degree S to quantify respectively to 3 sections by quantization;
3. the rgb value of target image I is mapped in 63 kinds of color spaces of HSV by quantization according to formula L=9H+3S+V;
To target image I carry out piecemeal processing, every block of image weight different according to how much impartings of its included information content, often
The color histogram of block image H (Ik) indicate, the corresponding weight w of every block of imagekIt indicates, then the divided group of whole image
Color histogram isWherein n is the block count to the target image I,
4. the divided group color histogram to the whole image is normalized, and the color as target image I
The color characteristic library of feature deposit image;
2) shape feature of target image I is extracted, and is stored in the shape feature library of image
1. the original color image of target image I is carried out edge enhancing using Domain Transform method, wherein parameter
Sigma_s=10, sigma_r=0.15;
2. the enhanced target image I in edge is converted to gray level image according to normalized form, by the gray level image using double
Linear difference method zooms in and out;Edge detection is carried out using canny boundary operator to the gray level image after scaling;
3. calculating the gradient-norm and gradient direction at each profile point of the gray level image:Using formula
Template to the gray level image carry out sobel operator filtering, obtain the horizontal gradient of the gray level imageAnd vertical gradientAnd then obtain the gradient-norm of the gray level imageThe gradient direction of the gray level image isResult according to above-mentioned edge detection obtains gradient of the gray level image at each profile point
Mould and gradient direction;
4. carrying out multiple dimensioned processing to the gray level image is divided into L for the gray level image using pyramid dividing method
Layer, each layer are divided into 2l(l=0 ..., L) block, at n-th piece of layer of the gray level image l of profile point, add up a certain ladder
Statistical value of the gradient modulus value as the gradient direction section for spending Direction interval, traverse the gray level image all profile points and
The statistical gradient direction histogram in gradient direction section, n-th piece of layer of the gray level image l is
5. the gradient orientation histogram of all image blocks of the gray level image is carried out splicing merging, complete gradient direction is obtained
Histogram
6. the complete gradient orientation histogram H (I) of the gray level image is normalized, obtaining dimension isShape
Shape feature vector, and the shape feature as target image I are stored in the shape feature library of image;
3) textural characteristics of target image I are extracted, and are stored in the textural characteristics library of image
1. calculating separately the roughness of each pixel of target image I, contrast and directionality;
2. by a channel of target image IThe square mean filter that size is 2*m is carried out, wherein m=1,2,3,4,5, from
And obtain 5 different mean filter imagesCalculate separately 5 different mean value filters
The level difference image of wave imageWithWherein
3. by the channelEach pixel (x, y) at 10 E are calculatedT,R(x, y) value, therefrom select maximum value as
Roughness value at pixel (x, y), i.e.,Wherein m=1,2,3,4,5;
4. in the channelThe mean value of statistical pixel (x, y) in the 7*7 window of pixel (x, y)VarianceIn the channelThe 7*7 window of pixel (x, y)
Four differences of middle statistical pixel (x, y)It is then described logical
RoadContrast value at pixel (x, y) is
5. by the channelAccording to formulaMask convolution obtain the channelLevel
GradientAnd vertical gradientIt further calculates to obtain the channelDirectionality value at pixel (x, y)
6. roughness value, contrast value and the directionality value in tri- channels R, G, B at the pixel (x, y) of adding up obtain described
The unrelated roughness value in channel, contrast value and directionality value at pixel (x, y);
7. by roughness, contrast and directionality all uniform quantizations of the target image I to g section, then the target figure
As the value interval of I roughness, contrast and directionality becomes [0, g-1];
8. joint roughness, contrast and directionality determine correspondence section of the pixel (x, y) in Texture similarityEach pixel is carried out on corresponding section tired
Add, obtains cumulative Texture similarity HT(IT), wherein HT(IT) dimension be g*g*g;
9. the Texture similarity H that the g*g*g is tieed upT(IT) be normalized, the textural characteristics of the target image I are obtained, are deposited
Enter the textural characteristics library of image;
(2) similarity calculation of target image I and image data set
1) a retrieval image Q is inputted, extracts its color characteristic X respectivelyC, shape feature XSWith textural characteristics XT, calculate XCWith face
Each characteristic Y in color characteristic libraryCLinear kernel range lineCalculate XSWith shape feature library
In each characteristic YSEuclidean distanceCalculate XTWith characteristic Y each in textural characteristics libraryTJSD distance
Wherein d is the dimension of character pair;
2) image two-by-two in image library is randomly selected to Qr1And Qr2, calculate the color characteristic distance of described image pairIt obtains
Sampling setFurther obtain sampling setSample average and sample standard deviation, repeat this process, obtain sample
This mean of mean, as dCGaussian Profile mean μC, dCStandard deviation sigmaC;According to aforesaid operations step, d can be similarly obtainedS
Gaussian Profile mean μS, standard deviation sigmaS, dTGaussian Profile mean μT, standard deviation sigmaT;FoundationPoint
Not by dC、dS、dTIt is transformed into standard gaussian distribution;
3) by above-mentioned three kinds of distance measurements dC、dS、dTIt merges to obtain d using the method for weightmerge=wCdC+wSdS+wTdT,wC+wS+wT
=1;
4) to calculated distance dmergeIt is ranked up, P data before taking, that is, P width image is as search result before taking.
2. a kind of image search method based on multi-feature fusion according to claim 1, which is characterized in that the step
(1) the gray level image scaling of step 2. is not more than 500 in step 2) in.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810418660.XA CN108829711B (en) | 2018-05-04 | 2018-05-04 | Image retrieval method based on multi-feature fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810418660.XA CN108829711B (en) | 2018-05-04 | 2018-05-04 | Image retrieval method based on multi-feature fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108829711A true CN108829711A (en) | 2018-11-16 |
CN108829711B CN108829711B (en) | 2021-06-01 |
Family
ID=64148349
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810418660.XA Active CN108829711B (en) | 2018-05-04 | 2018-05-04 | Image retrieval method based on multi-feature fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108829711B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109782214A (en) * | 2019-01-26 | 2019-05-21 | 段俊杰 | Electric energy meter status remote transmitting mechanism |
CN110135440A (en) * | 2019-05-15 | 2019-08-16 | 北京艺泉科技有限公司 | A kind of image characteristic extracting method suitable for magnanimity Cultural Relics Image Retrieval |
CN110826446A (en) * | 2019-10-28 | 2020-02-21 | 衢州学院 | Method and device for segmenting field of view region of texture-free scene video |
CN113095341A (en) * | 2019-12-23 | 2021-07-09 | 顺丰科技有限公司 | Image matching method, device and storage medium |
CN114170418A (en) * | 2021-11-30 | 2022-03-11 | 吉林大学 | Automobile wire harness connector multi-feature fusion image retrieval method by searching images through images |
CN116403419A (en) * | 2023-06-07 | 2023-07-07 | 贵州鹰驾交通科技有限公司 | Traffic light control method based on vehicle-road cooperation |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101551823A (en) * | 2009-04-20 | 2009-10-07 | 浙江师范大学 | Comprehensive multi-feature image retrieval method |
CN101763429A (en) * | 2010-01-14 | 2010-06-30 | 中山大学 | Image retrieval method based on color and shape features |
CN101770644A (en) * | 2010-01-19 | 2010-07-07 | 浙江林学院 | Forest-fire remote video monitoring firework identification method |
US20110255589A1 (en) * | 2009-08-03 | 2011-10-20 | Droplet Technology, Inc. | Methods of compressing data and methods of assessing the same |
CN102663391A (en) * | 2012-02-27 | 2012-09-12 | 安科智慧城市技术(中国)有限公司 | Image multifeature extraction and fusion method and system |
CN104298775A (en) * | 2014-10-31 | 2015-01-21 | 北京工商大学 | Multi-feature content-based image retrieval method and system |
US20150363945A1 (en) * | 2011-01-24 | 2015-12-17 | Alon Atsmon | System and process for automatically finding objects of a specific color |
CN105404657A (en) * | 2015-11-04 | 2016-03-16 | 北京工业大学 | CEDD feature and PHOG feature based image retrieval method |
CN105518668A (en) * | 2013-09-04 | 2016-04-20 | 沙祖拉公司 | Content based image retrieval |
CN106202338A (en) * | 2016-06-30 | 2016-12-07 | 合肥工业大学 | Image search method based on the many relations of multiple features |
CN107958073A (en) * | 2017-12-07 | 2018-04-24 | 电子科技大学 | A kind of Color Image Retrieval based on particle swarm optimization algorithm optimization |
-
2018
- 2018-05-04 CN CN201810418660.XA patent/CN108829711B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101551823A (en) * | 2009-04-20 | 2009-10-07 | 浙江师范大学 | Comprehensive multi-feature image retrieval method |
US20110255589A1 (en) * | 2009-08-03 | 2011-10-20 | Droplet Technology, Inc. | Methods of compressing data and methods of assessing the same |
CN101763429A (en) * | 2010-01-14 | 2010-06-30 | 中山大学 | Image retrieval method based on color and shape features |
CN101770644A (en) * | 2010-01-19 | 2010-07-07 | 浙江林学院 | Forest-fire remote video monitoring firework identification method |
US20150363945A1 (en) * | 2011-01-24 | 2015-12-17 | Alon Atsmon | System and process for automatically finding objects of a specific color |
CN102663391A (en) * | 2012-02-27 | 2012-09-12 | 安科智慧城市技术(中国)有限公司 | Image multifeature extraction and fusion method and system |
CN105518668A (en) * | 2013-09-04 | 2016-04-20 | 沙祖拉公司 | Content based image retrieval |
CN104298775A (en) * | 2014-10-31 | 2015-01-21 | 北京工商大学 | Multi-feature content-based image retrieval method and system |
CN105404657A (en) * | 2015-11-04 | 2016-03-16 | 北京工业大学 | CEDD feature and PHOG feature based image retrieval method |
CN106202338A (en) * | 2016-06-30 | 2016-12-07 | 合肥工业大学 | Image search method based on the many relations of multiple features |
CN107958073A (en) * | 2017-12-07 | 2018-04-24 | 电子科技大学 | A kind of Color Image Retrieval based on particle swarm optimization algorithm optimization |
Non-Patent Citations (3)
Title |
---|
YUE GAO ET AL.: "Mean-shift algorithm fusing multi feature", 《2017 IEEE 2ND ADVANCED INFORMATION TECHNOLOGY, ELECTRONIC AND AUTOMATION CONTROL CONFERENCE》 * |
晋杰: "基于多特征融合的服装图片检索关键技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
邵天日: "基于多特征DS融合策略的图像检索技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109782214A (en) * | 2019-01-26 | 2019-05-21 | 段俊杰 | Electric energy meter status remote transmitting mechanism |
CN110135440A (en) * | 2019-05-15 | 2019-08-16 | 北京艺泉科技有限公司 | A kind of image characteristic extracting method suitable for magnanimity Cultural Relics Image Retrieval |
CN110826446A (en) * | 2019-10-28 | 2020-02-21 | 衢州学院 | Method and device for segmenting field of view region of texture-free scene video |
CN110826446B (en) * | 2019-10-28 | 2020-08-21 | 衢州学院 | Method and device for segmenting field of view region of texture-free scene video |
CN113095341A (en) * | 2019-12-23 | 2021-07-09 | 顺丰科技有限公司 | Image matching method, device and storage medium |
CN114170418A (en) * | 2021-11-30 | 2022-03-11 | 吉林大学 | Automobile wire harness connector multi-feature fusion image retrieval method by searching images through images |
CN114170418B (en) * | 2021-11-30 | 2024-05-24 | 吉林大学 | Multi-feature fusion image retrieval method for automobile harness connector by means of graph searching |
CN116403419A (en) * | 2023-06-07 | 2023-07-07 | 贵州鹰驾交通科技有限公司 | Traffic light control method based on vehicle-road cooperation |
CN116403419B (en) * | 2023-06-07 | 2023-08-25 | 贵州鹰驾交通科技有限公司 | Traffic light control method based on vehicle-road cooperation |
Also Published As
Publication number | Publication date |
---|---|
CN108829711B (en) | 2021-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108829711A (en) | A kind of image search method based on multi-feature fusion | |
JP3740065B2 (en) | Object extraction device and method based on region feature value matching of region-divided video | |
CN102567731B (en) | Extraction method for region of interest | |
CN102663382B (en) | Video image character recognition method based on submesh characteristic adaptive weighting | |
CN101763429B (en) | Image retrieval method based on color and shape features | |
CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
CN107066916B (en) | Scene semantic segmentation method based on deconvolution neural network | |
Xiao et al. | Efficient shadow removal using subregion matching illumination transfer | |
CN104598933A (en) | Multi-feature fusion based image copying detection method | |
CN102194241B (en) | Internet-based design method of artistic picture splicing system | |
JP2003157438A (en) | Method for detecting subject matter regions in image | |
JP2003173442A (en) | Method of segmenting multidimensional image and computer system | |
CN106937120B (en) | Object-based monitor video method for concentration | |
CN105321177B (en) | A kind of level atlas based on image importance pieces method together automatically | |
CN103578119A (en) | Target detection method in Codebook dynamic scene based on superpixels | |
CN107507206B (en) | Depth map extraction method based on significance detection | |
CN102136064A (en) | System for recognizing characters from image | |
CN103473551A (en) | Station logo recognition method and system based on SIFT operators | |
CN105957124B (en) | With the natural image color edit methods and device for repeating situation elements | |
Yadav et al. | Survey on content-based image retrieval and texture analysis with applications | |
JP2009535683A (en) | Saliency estimation of object-based visual attention model | |
CN103198479A (en) | SAR image segmentation method based on semantic information classification | |
CN105404657A (en) | CEDD feature and PHOG feature based image retrieval method | |
CN107909079A (en) | One kind collaboration conspicuousness detection method | |
CN105868683A (en) | Channel logo identification method and apparatus |
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 |