CN107958073B - Particle cluster algorithm optimization-based color image retrieval method - Google Patents

Particle cluster algorithm optimization-based color image retrieval method Download PDF

Info

Publication number
CN107958073B
CN107958073B CN201711281707.4A CN201711281707A CN107958073B CN 107958073 B CN107958073 B CN 107958073B CN 201711281707 A CN201711281707 A CN 201711281707A CN 107958073 B CN107958073 B CN 107958073B
Authority
CN
China
Prior art keywords
image
color
similarity
images
retrieval
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
Application number
CN201711281707.4A
Other languages
Chinese (zh)
Other versions
CN107958073A (en
Inventor
饶云波
刘伟
范柏江
宋佳丽
苟苗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201711281707.4A priority Critical patent/CN107958073B/en
Publication of CN107958073A publication Critical patent/CN107958073A/en
Application granted granted Critical
Publication of CN107958073B publication Critical patent/CN107958073B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G06V10/507—Summing image-intensity values; Histogram projection analysis
    • 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

Abstract

The invention discloses a color image retrieval method based on particle cluster algorithm optimization, which belongs to the technical field of image retrieval and comprises the steps of firstly, respectively extracting bottom layer characteristics of each image in an image library and storing the bottom layer characteristics in the image library; then, distributing different similarity measurement formulas for different image feature descriptions; then, training through a PSO algorithm to obtain a weight of the similarity measurement of the database; when image retrieval processing is carried out, corresponding bottom layer features of a query image are extracted, descriptors of the features are extracted by comparing the query image with a target database, similarity measurement of different features is uniformly ordered based on trained similarity measurement weights, and the previous K most similar pictures are returned as retrieval results. Compared with the prior art, the invention combines and optimizes various feature extraction modes, and improves the retrieval precision of the CBIR retrieval system by combining a plurality of feature descriptors.

Description

Particle cluster algorithm optimization-based color image retrieval method
Technical Field
The invention belongs to the technical field of image processing and computer vision, and particularly relates to a color image retrieval method.
Background
Content-based image retrieval (CBIR) refers to a query itself being an image or a description of image content, and an index is created by extracting underlying features and then determining the similarity between two images by calculating and comparing the distance between the features and the query. Since the last 70 th century, content-based image retrieval has been a popular research field, and the mainstream search engines have introduced their picture search functions.
The existing CBIR technology is mainly based on knowledge in the fields of computer vision, pattern recognition, image processing, image understanding and the like, and a new media data representation method and a data model are introduced to model picture features. Meanwhile, in order to improve the retrieval precision, the CBIR also gradually relates to the fields of cognitive science, artificial intelligence, man-machine interaction, information retrieval and the like, and the retrieval result is optimized by adopting the technologies of relevance feedback, context analysis and the like so as to design a retrieval system with reliable performance and high operation efficiency. In the prior art, the underlying image features such as color, texture, contour, shape, spatial location, etc. are among the techniques that were used earlier and are well established. Among these underlying features, the most frequently used features are color, shape, texture.
Different features have different emphasis on description of image information, and a CBIR system which only uses a single feature as an image feature descriptor has a very narrow application scene and cannot meet the requirements of the modern society on an image retrieval system, so that a CBIR scheme which integrates a plurality of features is introduced. The CBIR system with multiple features brings another problem that since there is a difference in similarity matching between different features, it is not accurate to perform similarity determination using a uniform similarity metric, which may reduce the search effect. Therefore, the system effect can be improved well by adopting a plurality of similarity measurement standards in the CBIR system fusing multiple characteristics.
In addition to image feature extraction, another difficulty is to model the organization of features, in the process of extracting the underlying features of a digital image, there are various modeling methods, such as RGB, HSV, C L E L a b and C L E L u v, different color models, for example, color space is modeled, and color quantization can be roughly divided into two modes, namely, a segmentation algorithm and a clustering algorithm, the basic idea of the segmentation algorithm is to use M-fold color with the highest frequency of occurrence in the image as a palette, and then map the rest of colors into the palette according to a distance nearest criterion, thereby reconstructing image hierarchy.
Although CBIR technology has been studied for decades, there are still many key issues to be solved. The summary mainly includes three aspects: the method comprises the steps of effective extraction of image features, definition of picture similarity and non-similarity, and semantic gap between the bottom-layer picture features and the high-layer semantic meaning. Therefore, it is necessary to provide a more optimized algorithm to make up for the deficiency of the image retrieval at the present stage, so as to meet the actual multi-aspect requirement of the image retrieval of the user.
Disclosure of Invention
The invention aims to: aiming at the existing problems, the color image retrieval method based on particle cluster algorithm optimization is provided to improve the retrieval precision of the image retrieval method.
The invention discloses a color image retrieval method based on particle cluster algorithm optimization, which comprises the following steps:
1. training similarity measurement weight based on particle cluster algorithm:
101: taking different types of query images as training samples, and extracting color features, texture features and object shape features of the training samples;
the color characteristics are as follows: the color histogram and color moment characteristics of the image in the HSV color space;
the texture features are as follows: filtering the image by adopting a Gabor of a real number part, and extracting a mean value and a standard deviation of filtering output;
the object shape features are: extracting the object shape feature of the image based on an object shape feature extraction method of the region shape;
102: extracting color features, texture features and object shape features of each image to be retrieved in an image library;
103: and (3) optimizing the weight of similarity measurement of three types of image features by adopting a particle clustering algorithm:
initializing a similarity measure D of color featurescWeight value omega ofcSimilarity measure of texture features DtWeight value omega oftSimilarity measure of object shape features DsWeight value omega ofsAnd defining the particle position of each particle as (omega)c,ωt,ωs) Wherein 0 is not more than ωc,ωt,ωsLess than or equal to 1; initializing the number k of particles, and the local optimal position of each particle and the global optimal position of a particle swarm;
and (3) performing image query processing by taking n training samples of the same type as query images each time: respectively calculating similarity measure D of color features between each training sample and each retrieved image in the image librarycSimilarity measure of texture features DtSimilarity measure D with object shape featuressAnd based on the weight ωc,ωt,ωsCarrying out weighted summation on the three similarity measures to obtain a total similarity measure D; taking the searched images corresponding to the first K maximum total similarity measures D as search results;
iteratively updating the particle position, the local optimal position and the global optimal position based on the current retrieval precision until iteration is converged, and storing the most recently updated particle position;
the retrieval precision is as follows: the ratio of the number of the searched images matched with the query image in the search result to the value K;
2. and (3) image retrieval:
extracting color features, texture features and object shape features of a current query image;
and respectively calculating similarity measure D with color features between the images to be retrieved in the image librarycSimilarity measure of texture features DtSimilarity measure D with object shape featuress；
Weight omega obtained based on trainingc,ωt,ωsObtaining the current total similarity measure D ═ ωcDc+ωtDt+ωsDs；
And taking the searched images corresponding to the first K maximum total similarity measures D as search results and outputting the search results.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
compared with the prior art, the invention combines and optimizes various feature extraction modes, and improves the retrieval precision of the CBIR retrieval system by combining a plurality of feature descriptors. By adopting an image retrieval mode of fusing a plurality of characteristics, images similar to each characteristic can be retrieved, the requirements of users can be expanded to different characteristics, the similarity measurement results calculated by the characteristics are uniformly sorted, and the results are returned to the users according to the size sequence of the similarity measurement. The invention does not adopt a uniform similarity measurement mode, but adopts different measurement methods aiming at different structure modes of different characteristic vectors. The description emphasis points of different characteristics on the image are different, the method adopts a weighted combination mode, obtains the optimal weight combination through a particle clustering algorithm, and obtains the optimal similarity measurement mode.
Drawings
FIG. 1 is a framework flow diagram of an embodiment of the present invention;
FIG. 2 is a schematic diagram of an HSV color space configuration;
FIG. 3 is a schematic diagram of Pseudo-Zernike Moment shape feature extraction;
FIG. 4 is a schematic diagram of Gabor texture feature extraction;
FIG. 5 is a schematic view showing the results of the treatment, wherein (5-a) is an African resident, (5-b) is a beach, and (5-c) is a building.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings.
The color image retrieval method based on particle cluster algorithm optimization is characterized in that various bottom layer characteristics of a single image in a target database are normalized into an image information descriptor with a specific length, and the image characteristics are stored in a characteristic database to establish an index table. In the process of executing picture search by a user, extracting various single-low-layer features of the image serving as a search term to form a query vector. As shown in fig. 1, similarity matching is performed by querying the feature vectors and the feature vectors in the feature database, and finally, the retrieved image results are sorted and output in a unified manner. Because the features described by different features have different emphasis points, in order to avoid the problems of low accuracy of retrieval results and the like caused by factors such as similarity calculation, incomplete matching of single features and the like, the content-based image retrieval can cover more matched results by using multi-feature retrieval so as to improve the retrieval rate, and the accuracy of the matching results in the retrieval process is improved by using the normalization of similarity measurement, namely the accuracy rate is improved.
The method comprises the following specific implementation steps:
(1) and (5) extracting color features.
In this embodiment, an HSV color space is used as a model for representing the color characteristics of an image. Firstly, converting an original picture into an HSV color space picture, and then extracting color features.
Referring to fig. 2, HSV color space describes colors of an object according to Hue (H), Saturation (S), and brightness (V), and is more suitable for the visual effect of human observation than the conventional RGB color space. In order to more effectively represent Color feature information of an image, the present invention models Color features using a Color histogram (ColorHistogram) and Color moments (Color moments), respectively.
The HSV color space is quantized using the (8,2,2) regime, i.e., H components are divided into an 8-channel representation on average, and S and V are quantized using 2 and 2 channels, respectively, to obtain a histogram vector of 8 х 2,2 х 2 ═ 32 dimensions.
Color moments are a simple and effective method for representing color features, i.e. using a first moment muiMean, second order momenti(variance, viarance) represents the image color distribution condition, which is defined as:
Figure GDA0002468522700000041
wherein p isijIs the value of j pixel points of the picture on the ith channel, and N is the total number of the picture pixels. HSV is a three-dimensional color space, and a 6-dimensional color moment vector can be obtained by calculating the color moments of the three color spaces, respectively.
By combining the color histogram and the color moments, a combined color feature descriptor of (32+6) ═ 38 dimensions is obtained.
(2) And extracting texture features.
In this embodiment, the extraction of the texture features of the image is realized by using gabor filtering. Since the 2-dimensional Gabor filter function is a complex function, the complexity of calculation is increased in practical use. To simplify the calculation, while preserving the feature extraction characteristics of the Gabor filter function. The invention uses Gabor of real part to filter, which is defined as follows:
Figure GDA0002468522700000051
wherein
Figure GDA0002468522700000052
x and y are coordinate positions of pixels in the picture, lambda and theta respectively represent wavelength and the offset direction of filtering, sigma represents the standard deviation of a Gaussian function in a Gabor kernel function, the parameter determines the size of an acceptable area of the Gabor kernel, and the value of sigma is related to b (Bandwidth, difference of high and low frequencies) and lambda; γ represents an aspect ratio, i.e., a spatial aspect ratio, representing the ellipticity of the Gabor filter; psi represents the phase parameter of cosine function in Gabor kernel function, the effective value is-180 degrees, the corresponding equation of 0 degree and 180 degrees is symmetrical with the origin, and the power of-90 degree and 90 degreeThe ranges are respectively centrosymmetric to the origin. In the present embodiment, 5 parameters, 0.1, 0.8, 2, 5, and 11, are used as the wavelength of the filter, and four parameters, 0, pi/4, pi/2, and 3 pi/4, are used as the offset of the Gabor filtering. The images are then subjected to texture feature extraction by their combined filtering. And finally, obtaining the vector of the texture feature by calculating the mean value and the standard deviation of the filtering result:
Figure GDA0002468522700000053
where M is 0,1, …, M-1, N is 0,1, …, N-1. M, N representing the number of wavelengths and offset parameters of the Gabor filter, respectively, P × Q representing the original size of the picturemn(x, y) is a texture feature value subjected to Gabor filtering processing at the coordinate point (x, y) by the above processing, 5 × 4 × 2 ═ 40-dimensional texture feature vector F can be obtainedT＝{μ00,σ00,μ01,σ01,...,μM-1N-1,σM-1N-1As shown in fig. 4.
(3) And extracting the shape features of the object.
Image shape descriptors can be divided into two categories: one type is a set of pixels that describe the contour of the target region boundary of the shape, called a contour-based shape descriptor; one type is a set of all pixels within a target region that describes a shape, called a region-based shape descriptor. The specific implementation mode adopts pseudo-Zernike monomer as an image shape feature extraction method. The Pseudo-Zernick momentas is an object shape feature extraction algorithm based on region shapes, can represent images in multiple stages, and has strong anti-noise capability.
In a digital image of size M × N, a Pseudo-Zernike Moment of order N and repetition M is defined as:
Figure GDA0002468522700000054
where f (x, y) is a functional representation used to model the image and (x, y) represents the coordinate position in the picture pixel. Pseudo-Zernike moment by using a polynomial { V }nm(x, y) } sets are transformed with polar coordinates,mapping f (x, y) to x2+y2≤1：
Figure GDA0002468522700000061
Vnm(x,y)＝Vnm(ρ,θ)＝Rnm(ρ)exp(jmθ)，
Figure GDA0002468522700000062
Wherein the content of the first and second substances,
Figure GDA0002468522700000063
θ is arctan (y/x). The order n of the moment is a non-negative integer, and the value of the repetition degree m is | m | < n, j.
The order of use of the invention being 5
Figure GDA0002468522700000064
As the shape vector of the image texture feature, finally obtaining a 21-dimensional image shape feature vector:
Figure GDA0002468522700000065
(4) a similarity measure.
The present invention uses three different image feature descriptors. To compare the similarity of the relevant image feature vectors to the images in the target image library. The invention adopts different similarity measurement methods for different image characteristics.
Color feature similarity measure:
Figure GDA0002468522700000066
texture moment similarity measure:
Figure GDA0002468522700000067
shape feature similarity measure:
Figure GDA0002468522700000068
in the above similarity measurement formula, Q and I represent two pictures for similarity measurement. Wherein
Figure GDA0002468522700000069
Respectively representing the color characteristic components of the pictures Q and I;
Figure GDA00024685227000000610
and
Figure GDA00024685227000000611
respectively representing a mean vector and a standard deviation vector in the texture characteristic moments of the Q picture and the I picture;
Figure GDA00024685227000000612
representing the Pseudo-Zernike Moment value of the picture at the order of 5.
Different characteristic descriptors have different description emphasis on the picture, for example, if the object to be retrieved is a blue sky and a sea, the color descriptor of the image is more emphasized, but if the picture with objects such as bananas or plates is to be retrieved, the shape characteristics of the objects in the picture are more emphasized, and the objects with sickle shapes or round shapes in the picture are searched for and compared. In order to further improve the retrieval precision, the similarity measurement results of different characteristics are weighted and combined to be used as the similarity measurement value of the final query image and the final target image, and the similarity measurement value is expressed as follows: d (Q, I) ═ ωcDColor characteristics+ωtDTexture moment+ωsDShape ofWherein ω isc，ωt，ωsRepresenting the weight of three characteristic components in the similarity measurement process, and simplifying the calculated amount for equalizing measurement results simultaneously to make omegac+ωt+ωs＝1。
(5) And optimizing the similarity metric by using a particle clustering algorithm.
The Particle clustering algorithm (PSO) is a common method for solving the optimal solution, and the velocity and Particle position update formula is as follows:
V[i]＝ω*V[i]+c1*rand()*(pbest[i]-present[i])+c2*rand()*(gbest-present[i])present[i+1]＝present[i]+V[i]
the functional formula is an equation for obtaining the optimal solution by moving the particles to the optimal solution, where V [ i ] represents the velocity of the ith particle, w represents the inertia weight, c1 and c2 represent learning parameters, rand () represents a random number between 0 and 1, pbest [ i ] represents the optimal value searched for by the ith particle (local optimal solution), gbest represents the optimal value searched for by the entire cluster (global optimal solution), and present [ i ] represents the current position of the ith particle.
In this embodiment, Corel-10 is used as an experimental image database. Corel-10 is an open source picture library with 10000 pictures, containing 10 categories, with label names Africa, beach, mountain, car, dinosaur, elephant, flower, horse, food and building, respectively. There are 1000 pictures for each door category.
And 3 weights in the similarity measurement process are selected as the motion particles. In order to make the weight value size not generate out-of-range error, let:
Figure GDA0002468522700000071
wherein, 0 is not less than omegac,ωt,ωs≤1。
For image retrieval based on multi-feature fusion, different feature descriptors have different description emphasis on the same image, and the query result can be optimized by well utilizing the advantages of the different feature descriptors by optimizing the combination mode of each different feature. The invention solves the problem by utilizing query result feedback information and PSO algorithm optimization, namely, the adjustment of the weight is guided by analyzing the result of test query feedback. Specifically, the invention randomly selects 20 different pictures from each category of a test database (Corel database) as query pictures, and the previous K (preset value) pictures obtained by the CBIR system are used for calculating retrieval precision each time. The retrieval precision is as follows: and the ratio of the number of the images matched with the query image in the returned K images to the value K. The invention uses the retrieval precision as the self-adaptive function, the higher the retrieval precision is, the better the value is, and the closer the position of the particle is to the position of the local optimal solution.
Wherein, the training weight omegac，ωt，ωsThe specific process is as follows:
firstly, randomly selecting N pictures from each category of a Corel database as training data. The number of particles n ═ k (k is a preset value, set based on the database size), the initial position of the particles present [ i ] is generated by a random number, and 0 ≦ present [ i ] ≦ 1, for example set to 0.01;
the local optimal position gbest and global optimal position pbest [ i ] of the particle are then initialized]By specifying the initial position of any particle motion by a random number
Figure GDA0002468522700000081
Where i is the identifier of the particle,
Figure GDA0002468522700000082
represents the corresponding weight omega of the current particle ic、ωt、ωsA value of (d);
performing a query operation on each particle
Figure GDA0002468522700000083
Calculate a query parameter as
Figure GDA0002468522700000084
Accuracy of time search, wherein IiFeature description vectors (color, texture, shape, etc.) representing the query picture;
according to the current highest retrieval precision
Figure GDA0002468522700000085
Update the values of gbest and pbest [ i ]]A value of (d);
judging whether the iteration convergence condition is met, if not, continuing the query operation, and updating the gbest and pbest [ i ] based on the retrieval progress](ii) a Otherwise, the current is output
Figure GDA0002468522700000086
In the PSO algorithm, the present embodiment sets the convergence condition of the clustering algorithm to be that the query accuracy for querying the same picture twice is less than 0.01 or the algorithm cycle number exceeds 1000, that is, when the deviation of the query accuracy meeting the last two times is less than the threshold (0.01) or reaches the maximum iteration number, the updating is stopped, and the weight of the optimal combination is obtained.
When the method is used for real-time query processing, various bottom-layer features (color histograms, color moments, texture moments and shapes) of a query object are obtained based on a currently input image to be queried, feature description vectors of the image to be queried are obtained, then vectors are extracted according to different features, similarity comparison is carried out between the vectors and a feature library of a picture library by adopting different feature measurement methods, measurement results of similarity between pictures corresponding to different features are obtained, final similarity is obtained based on a weight of a trained optimal combination, finally the obtained final similarity is ranked (for example, similarity result ranking output is increased), and the first K (for example, 35) most similar K pictures are returned to a user, as shown in fig. 5.
The method fuses multiple feature vectors as feature descriptors of the image, and allocates a similarity measurement formula which is suitable for the structure of each feature vector to each feature vector according to the characteristics of each different feature descriptor. Since the picture to be queried by the user is unknown, different pictures have different requirements on the type of feature descriptor. A single feature cannot meet all image retrieval requirements, and images similar in different features can be retrieved simultaneously by fusing multiple feature descriptions. Meanwhile, because a semantic gap (semantic gap) exists between the bottom-layer feature description and the high-layer semantics, different images may have the same feature descriptor on a certain aspect, and errors caused by the problem can be effectively overcome by fusing a plurality of features. Meanwhile, in the process of calculating the feature similarity, the invention combines the measurement results of the features of the components in a weighting mode. And training the optimal weight of each similarity component through a particle clustering algorithm. Thereby avoiding errors caused by similarity calculation and incomplete description of matching of single features.
The present invention relates to content-based image retrieval, primarily with a view to optimizing CBIR techniques. The CBIR system fusing multiple features is optimized by using the PSO algorithm, so that the accuracy of image retrieval can be effectively improved, and the method has certain practical value.
While the invention has been described with reference to specific embodiments, any feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise; all of the disclosed features, or all of the method or process steps, may be combined in any combination, except mutually exclusive features and/or steps.

Claims (4)

1. A color image retrieval method based on particle cluster algorithm optimization is characterized by comprising the following steps:
1. training similarity measurement weight based on particle cluster algorithm:
101: taking different types of query images as training samples, and extracting color feature vectors, texture feature vectors and object shape feature vectors of the training samples;
the color feature vector is: the color histogram and color moment characteristics of the image in the HSV color space;
the texture feature vector is: carrying out Gabor filtering on the image by adopting Gabor of a real number part, and extracting a mean value and a standard deviation of filtering output;
the object shape feature vector is: extracting the object shape feature of the image based on an object shape feature extraction method of the region shape;
102: extracting color characteristic vectors, texture characteristic vectors and object shape characteristic vectors of all searched images in an image library;
103: and (3) optimizing the weight of the similarity measurement of the three types of image feature vectors by adopting a particle clustering algorithm:
initializing a similarity measure D of color feature vectorscWeight value omega ofcSimilarity measure of texture feature vectors DtWeight value omega oftCharacteristic of object shapeMeasure of similarity of quantities DsWeight value omega ofsAnd defining the particle position of each particle as (omega)c,ωt,ωs) Wherein 0 is not more than ωc,ωt,ωsLess than or equal to 1; initializing the number k of particles, and the local optimal position of each particle and the global optimal position of a particle swarm;
and (3) performing image query processing by taking n training samples of the same type as query images each time: respectively calculating similarity measure D of color feature vectors between each training sample and each image to be searched in the image librarycSimilarity measurement of texture feature vectors DtSimilarity measure D with object shape feature vectorsAnd based on the weight ωc,ωt,ωsCarrying out weighted summation on the three similarity measures to obtain a total similarity measure D; taking the searched images corresponding to the first K maximum total similarity measures D as search results;
iteratively updating the particle position, the local optimal position and the global optimal position based on the current retrieval precision until iteration is converged, and storing the most recently updated particle position;
the retrieval precision is as follows: the ratio of the number of the searched images matched with the query image in the search result to the value K;
wherein the similarity measure D of the color feature vectorscSimilarity measurement of texture feature vectors DtSimilarity measure D with object shape feature vectorsThe calculation method is as follows:
definition Q, I represents two images for which a similarity measure is performed, the similarity measure D between images Q, Ic、DtAnd DsThe method specifically comprises the following steps:
Figure FDA0002468522690000021
Figure FDA0002468522690000022
Figure FDA0002468522690000023
representing the color characteristic components, N, of the images Q, I, respectivelycA dimension number representing a color feature vector;
Figure FDA0002468522690000024
Figure FDA0002468522690000025
and
Figure FDA0002468522690000026
mean and standard deviation vectors in the texture feature moments, N, representing images Q and I, respectivelytRepresenting the dimension number of the texture feature vector;
Figure FDA0002468522690000027
Figure FDA0002468522690000028
representing the components of the object shape feature vectors, M, of images Q and I, respectivelysRepresenting the order of the object shape feature vector;
2. and (3) image retrieval:
extracting color characteristic vectors, texture characteristic vectors and object shape characteristic vectors of the current query image;
and respectively calculating similarity measure D with color feature vector between the images to be searched in the image librarycSimilarity measurement of texture feature vectors DtSimilarity measure D with object shape feature vectors；
Weight omega obtained based on trainingc,ωt,ωsObtaining the current total similarity measure D ═ ωcDc+ωtDt+ωsDs；
And taking the searched images corresponding to the first K maximum total similarity measures D as search results and outputting the search results.
2. The method according to claim 1, wherein the color histogram and color moment features of the image in HSV color space are specifically:
quantizing the HSV color space by adopting a system of (8,2,2), namely, averagely dividing H components into 8 channels for representation, and averagely dividing S and V components into 2 channels for representation respectively;
extracting the color histogram characteristics of each quantized channel;
respectively calculating H, S, V first moment mu of component according to formulaiAnd second momentiBy a first moment muiAnd second momentiColor moment characteristics of the components, wherein
Figure FDA0002468522690000029
Wherein p isijAnd the pixel value of the jth pixel point of the image on the ith channel is represented, and N represents the total amount of the pixel points of the image.
3. The method of claim 1, wherein the Gabor filter has a wavelength of 0.1, 0.8, 2, 5, 11 and an offset of 0, pi/4, pi/2, 3 pi/4.
4. The method of claim 1, wherein the object shape features are extracted by: the Pseudo-Zernick Moments algorithm of order 5.
CN201711281707.4A 2017-12-07 2017-12-07 Particle cluster algorithm optimization-based color image retrieval method Active CN107958073B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711281707.4A CN107958073B (en) 2017-12-07 2017-12-07 Particle cluster algorithm optimization-based color image retrieval method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711281707.4A CN107958073B (en) 2017-12-07 2017-12-07 Particle cluster algorithm optimization-based color image retrieval method

Publications (2)

Publication Number Publication Date
CN107958073A CN107958073A (en) 2018-04-24
CN107958073B true CN107958073B (en) 2020-07-17

Family

ID=61958120

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711281707.4A Active CN107958073B (en) 2017-12-07 2017-12-07 Particle cluster algorithm optimization-based color image retrieval method

Country Status (1)

Country Link
CN (1) CN107958073B (en)

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108829711B (en) * 2018-05-04 2021-06-01 上海得见计算机科技有限公司 Image retrieval method based on multi-feature fusion
CN108875813B (en) * 2018-06-04 2021-10-08 北京工商大学 Three-dimensional grid model retrieval method based on geometric image
CN109063656B (en) * 2018-08-08 2021-08-24 厦门市美亚柏科信息股份有限公司 Method and device for carrying out face query by using multiple face engines
CN109259965A (en) * 2018-09-29 2019-01-25 重庆大学 Adaptive medical POCT system and detection method
CN109360289B (en) * 2018-09-29 2021-09-28 南京理工大学 Power meter detection method fusing inspection robot positioning information
CN109846479A (en) * 2019-03-06 2019-06-07 武汉几古几古科技有限公司 One kind being based on child image Analysis of Cognition Psychology system
WO2020210996A1 (en) * 2019-04-17 2020-10-22 深圳大学 Image query method and system, computing device and storage medium
CN110502660B (en) * 2019-08-28 2024-02-13 南京大学 Multi-distance measurement image retrieval method under weak supervision
CN111930844B (en) * 2020-08-11 2021-09-24 肖岩 Financial prediction system based on block chain and artificial intelligence
CN112101430B (en) * 2020-08-28 2022-05-03 电子科技大学 Anchor frame generation method for image target detection processing and lightweight target detection method
CN112528061A (en) * 2020-10-12 2021-03-19 西安理工大学 Multi-target image retrieval method based on selective convolution descriptor aggregation
CN112395451A (en) * 2020-11-17 2021-02-23 厦门博海中天信息科技有限公司 Classification retrieval method, system, medium and device based on image features
CN112667842B (en) * 2020-12-29 2022-08-02 中国电子科技集团公司第五十八研究所 Image texture retrieval method
CN112685591A (en) * 2020-12-31 2021-04-20 荆门汇易佳信息科技有限公司 Accurate picture retrieval method for user interest area and feedback guidance
CN114455255A (en) * 2022-01-27 2022-05-10 山东仁功智能科技有限公司 Abnormal cigarette sorting error detection method based on multi-feature recognition
CN116737982B (en) * 2023-08-11 2023-10-31 拓锐科技有限公司 Intelligent screening management system for picture search results based on data analysis

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024155A (en) * 2010-12-06 2011-04-20 广州科易光电技术有限公司 Rapid matching method of multispectral images based on edge detection
CN102426606A (en) * 2011-11-11 2012-04-25 南京财经大学 Method for retrieving multi-feature image based on particle swarm algorithm

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004334339A (en) * 2003-04-30 2004-11-25 Canon Inc Information processor, information processing method, and storage medium, and program
US20110202543A1 (en) * 2010-02-16 2011-08-18 Imprezzeo Pty Limited Optimising content based image retrieval

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024155A (en) * 2010-12-06 2011-04-20 广州科易光电技术有限公司 Rapid matching method of multispectral images based on edge detection
CN102426606A (en) * 2011-11-11 2012-04-25 南京财经大学 Method for retrieving multi-feature image based on particle swarm algorithm

Also Published As

Publication number Publication date
CN107958073A (en) 2018-04-24

Similar Documents

Publication Publication Date Title
CN107958073B (en) Particle cluster algorithm optimization-based color image retrieval method
JP5236785B2 (en) Color image search method, color image search apparatus, color image search system, and computer executable program
Srivastava et al. A review: color feature extraction methods for content based image retrieval
Bui et al. Scalable sketch-based image retrieval using color gradient features
CN110188225B (en) Image retrieval method based on sequencing learning and multivariate loss
WO2021082168A1 (en) Method for matching specific target object in scene image
CN110188763B (en) Image significance detection method based on improved graph model
Saad et al. Image retrieval based on integration between YC b C r color histogram and shape feature
Erkut et al. HSV color histogram based image retrieval with background elimination
Kam et al. Content based image retrieval through object extraction and querying
CN112561976A (en) Image dominant color feature extraction method, image retrieval method, storage medium and device
Pujari et al. Content-based image retrieval using color and shape descriptors
Wu et al. Generic proposal evaluator: A lazy learning strategy toward blind proposal quality assessment
CN108319959A (en) A kind of corps diseases image-recognizing method compressed based on characteristics of image with retrieval
Gao et al. SHREC’15 Track: 3D object retrieval with multimodal views
James Face Image retrieval with HSV color space using clustering techniques
Yang et al. Weakly supervised class-agnostic image similarity search based on convolutional neural network
Huet et al. Inexact graph retrieval
Arjunan et al. Image Classification in CBIR systems with color histogram features
Bajaj Image indexing and retrieval in compressed domain using color clusters
Zhao et al. Indoor and outdoor scene classification method based on Fourier transform
CN110750672A (en) Image retrieval method based on depth metric learning and structure distribution learning loss
Nie Research on image network retrieval application based on swarm optimization algorithm
Tang Intelligent colour matching method for 3D character animation based on texture features
Tai et al. Image retrieval based on color and texture

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