CN109740008B - Color texture image retrieval method based on non-Gaussian multi-correlation HMT model - Google Patents

Color texture image retrieval method based on non-Gaussian multi-correlation HMT model Download PDF

Info

Publication number
CN109740008B
CN109740008B CN201811607246.XA CN201811607246A CN109740008B CN 109740008 B CN109740008 B CN 109740008B CN 201811607246 A CN201811607246 A CN 201811607246A CN 109740008 B CN109740008 B CN 109740008B
Authority
CN
China
Prior art keywords
image
hmt
coefficient
vector
parameter
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
CN201811607246.XA
Other languages
Chinese (zh)
Other versions
CN109740008A (en
Inventor
杨红颖
高思洋
牛盼盼
王向阳
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.)
Liaoning Normal University
Original Assignee
Liaoning Normal University
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 Liaoning Normal University filed Critical Liaoning Normal University
Priority to CN201811607246.XA priority Critical patent/CN109740008B/en
Publication of CN109740008A publication Critical patent/CN109740008A/en
Application granted granted Critical
Publication of CN109740008B publication Critical patent/CN109740008B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a color texture image retrieval method based on a non-Gaussian multi-correlation HMT model, which comprises the steps of firstly, carrying out DWT conversion on R, G and B channels of a texture image; secondly, binding high-frequency sub-band coefficients of nodes in the same scale, the same direction and the same position in the R, G and B color components obtained by transformation into a coefficient vector; then modeling a VB-HMT statistical model to obtain an image characteristic parameter quintuple; and finally, selecting the parameter set to represent the image characteristics for image retrieval. Experimental results show that the method of the invention takes the parameter tuple as the characteristic representing any image, reduces the characteristic dimension, thereby reducing the time of similarity calculation and obtaining better retrieval effect.

Description

Color texture image retrieval method based on non-Gaussian multi-correlation HMT model
Technical Field
The invention belongs to the technical field of digital image retrieval, relates to an image retrieval method based on contents, and particularly relates to a color texture image retrieval method based on a non-Gaussian multi-correlation HMT model.
Background
With the rapid development of network multimedia technology, information circulates at a high speed, digital image data is rapidly increasing, people can capture images at any time and any place to form image information and immediately transmit the image information to the internet, and the image as an information transmission medium is integrated into the daily life of people. For the increasing number of digital image sources on the network, how to perform more reasonable and efficient retrieval and classification management on the digital image sources is a hot problem which needs to be solved urgently.
The text-based image retrieval method (TBIR) is widely applied in the early stage of image retrieval technology, and essentially, the image retrieval is converted into the retrieval of image keywords by manually performing text annotation on the image. However, the network technology is continuously developed nowadays, the information amount of digital images is increased greatly, and the TBIR method has certain disadvantages: it is difficult to fully and accurately depict the growing contents of images and visual information by a keyword retrieval method, and a content-based image retrieval technology (CBIR) is one of effective methods for solving the above problems. Compared with the TBIR, the CBIR represents the image information through calculation, and further performs feature matching and searching on the similar images, and has the advantage of extracting and converting visual information such as image color, outline and the like by using an optimal algorithm. The characteristic extraction and matching process can replace manual labeling by using computer software equipment, and has high efficiency and objectivity, so that a large amount of labor and time cost is saved, and the influence of subjective factors on various aspects such as interests, knowledge reserves, occupation and the like is eliminated.
In the current CBIR technology, research based on a shape and color feature retrieval method has been successful, however, there are many difficulties in research on image texture feature extraction, and feature extraction is a key for determining the quality of a CBIR algorithm, but there are no reports on a color texture image retrieval method based on a non-gaussian multi-correlation HMT model so far.
Disclosure of Invention
The invention provides a color texture image retrieval method based on a non-Gaussian multi-correlation HMT model, aiming at solving the technical problems in the prior art.
The technical solution of the invention is as follows: a color texture image retrieval method based on a non-Gaussian multi-correlation HMT model is characterized by comprising the following steps:
appointing: l represents a low-frequency subband obtained by discrete wavelet transform, H represents a high-frequency subband;
Figure 833750DEST_PATH_IMAGE001
represents a coefficient vector; />
Figure 176876DEST_PATH_IMAGE002
Representing hidden states of the HMT model coefficients; />
Figure 114876DEST_PATH_IMAGE003
Mean covarianceA matrix; />
Figure 254870DEST_PATH_IMAGE004
Represents a gamma function; />
Figure 84155DEST_PATH_IMAGE005
Representing a set of image parameter features; k represents the number of wavelet coefficient vector trees; />
Figure 625995DEST_PATH_IMAGE006
Indicates the fifth->
Figure 476DEST_PATH_IMAGE007
A set of hidden state variables of the vector tree; />
Figure 627766DEST_PATH_IMAGE008
Is the iteration number; />
Figure 73791DEST_PATH_IMAGE009
Representing the image characteristic parameter set after the parameter updating; />
Figure 863280DEST_PATH_IMAGE010
Represents a wavelet coefficient node->
Figure 470979DEST_PATH_IMAGE011
A parent node of (1); />
Figure 975778DEST_PATH_IMAGE012
Represents the probability quality function of the edge state of the wavelet transform subband coefficient, < >>
Figure 897598DEST_PATH_IMAGE013
Representing a joint probability quality function between parent nodes and child nodes of wavelet transform sub-band coefficients;
Figure 335401DEST_PATH_IMAGE005
a set of parameters representing image feature values; v is an image texture feature library; i refers to an image to be retrieved; j refers to an image in the image library;
Figure 176318DEST_PATH_IMAGE014
is the relative entropy divergence distance between I and J; />
Figure 715884DEST_PATH_IMAGE015
The representative image I is in->
Figure 175815DEST_PATH_IMAGE016
The eigenvalues of the eigenvectors at the component positions; />
Figure 546754DEST_PATH_IMAGE017
The image features to be retrieved; />
Figure 214364DEST_PATH_IMAGE018
Features in the image feature library;
a. initial setting
Acquiring an image J, and performing parameter initialization operation;
b. high frequency subband acquisition
Respectively executing three levels of DWT on three color channels of R, G and B of an image J, wherein each channel obtains 1L and 9H composed of matrixes with different sizes;
c. coefficient BKF-VB-HMT modeling
c.1 Binding the H coefficients of the same-scale, same-direction and same-position nodes in the R, G and B color channels into a coefficient vector
Figure 178909DEST_PATH_IMAGE001
；
c.2 Probability density function using BKF-VB-HMT model
Figure 225887DEST_PATH_IMAGE019
The 9H coefficient vectors per image were statistically modeled, device for selecting or keeping>
Figure 123436DEST_PATH_IMAGE019
The definition is as follows:
Figure 306156DEST_PATH_IMAGE020
；
in the formula (I), the compound is shown in the specification,
Figure 741685DEST_PATH_IMAGE021
、/>
Figure 871315DEST_PATH_IMAGE022
representing a shape parameter and a scale parameter; />
Figure 216846DEST_PATH_IMAGE023
In a second modified manner>
Figure 445833DEST_PATH_IMAGE024
The BKF function of order, calculated as follows:
Figure 165396DEST_PATH_IMAGE025
；
Figure 895455DEST_PATH_IMAGE026
is a state probability matrix, based on the state>
Figure 970858DEST_PATH_IMAGE027
Representing the scale after the discrete wavelet transform,
Figure 682331DEST_PATH_IMAGE028
，/>
Figure 843185DEST_PATH_IMAGE029
for the initial state probability, the HMT model state transition probability matrix, < >>
Figure 111356DEST_PATH_IMAGE030
Is represented as follows:
Figure 78482DEST_PATH_IMAGE031
；
wherein
Figure 977167DEST_PATH_IMAGE032
Represents a hidden state value range, wherein>
Figure 625318DEST_PATH_IMAGE033
；/>
c.3 Formalizing parameters in the probability density function as quintuple
Figure 697179DEST_PATH_IMAGE005
Is expressed as follows:
Figure 730863DEST_PATH_IMAGE034
；
c.4 Estimating the quintuple using a maximum expectation algorithm according to a maximum likelihood criterion
Figure 128346DEST_PATH_IMAGE005
In EM step, the set of coefficients of each vector tree is->
Figure 998213DEST_PATH_IMAGE001
As observation data, a->
Figure 873765DEST_PATH_IMAGE035
And with
Figure 761956DEST_PATH_IMAGE036
(ii) a In ML step, update->
Figure 799182DEST_PATH_IMAGE009
The parameters in (1) are:
Figure 421924DEST_PATH_IMAGE037
；
Figure 570009DEST_PATH_IMAGE038
；
Figure 578285DEST_PATH_IMAGE039
；
Figure 51992DEST_PATH_IMAGE040
；
Figure 162030DEST_PATH_IMAGE041
；
c.5 Parameter set of BKF-VB-HMT distribution model to be obtained
Figure 379385DEST_PATH_IMAGE005
Storing the image texture feature library V as a feature vector of each image J to be retrieved and used;
d. image processing operation to be retrieved
d.1 Inputting I, and respectively implementing three-level DWT conversion on three color components of R, G and B, wherein each color component is decomposed into 1L and 9H composed of matrixes with different sizes;
d.2 And c, repeating the step c, modeling the high-frequency sub-band coefficient of the image I to be retrieved through VB-HMT by utilizing BKF distribution, and calculating a parameter characteristic five-element group
Figure 245097DEST_PATH_IMAGE005
Obtaining a feature vector of the image I;
e. similarity calculation
e.1 Definition of
Figure 889705DEST_PATH_IMAGE014
Represents the relative entropy divergence distance between I and J>
Figure 221460DEST_PATH_IMAGE015
Finger I is in->
Figure 976927DEST_PATH_IMAGE011
The features of the feature vector at the component positionValue,. Or>
Figure 959795DEST_PATH_IMAGE042
Is J in>
Figure 509725DEST_PATH_IMAGE011
The characteristic value of a characteristic vector in the case of a component position, is evaluated>
Figure 328777DEST_PATH_IMAGE043
Refers to the total number of images, so the relative entropy divergence distance between I and J can be calculated as follows: />
Figure 887934DEST_PATH_IMAGE044
；
e.2 The similarity results are arranged from small to large in sequence, and a better search result is output within a set search result range;
f. and d-e are repeated until the images I and J are completely matched.
The invention firstly executes DWT conversion to channels R, G and B of the texture image; secondly, binding the high-frequency subband coefficients of the nodes with the same scale, the same direction and the same position in the R, G and B color components obtained by transformation into a coefficient vector; then modeling a VB-HMT statistical model to obtain an image characteristic parameter quintuple; and finally, selecting the parameter set to represent the image characteristics for image retrieval. The experimental result shows that the method reduces the feature dimension by taking the parameter tuple as the feature representing any image, thereby reducing the time for calculating the similarity and obtaining better retrieval effect.
Compared with the prior art, the invention has the following advantages:
firstly, a multi-correlation statistical modeling method is adopted, the defect that information representation is incomplete when a traditional single method is used for depicting texture images is overcome, time complexity is reduced, and image information is depicted and represented comprehensively and accurately, so that an optimal retrieval result is obtained;
secondly, the parameter tuple is used as a characteristic value to represent any image, so that the characteristic dimension is reduced, and the time consumption of similarity calculation is reduced;
thirdly, a WD-BKF-VB-HMT model is constructed by utilizing the R, G and B components, model parameters are estimated by adopting more correlations, and the coefficient correlation among color channels is fully considered, so that the representation of the color image texture information is more comprehensive, and the more accurate modeling effect is achieved.
Drawings
Fig. 1 is a graph of the fitting effect of wavelet domain BKF edge distribution PDF according to an embodiment of the present invention.
FIG. 2 is a PDF fitting effect diagram of a color image WD-BKF-VB-HMT model according to an embodiment of the invention.
Fig. 3 to 6 are graphs showing the result of searching the color texture image library according to the embodiment of the present invention.
FIG. 7 is a comparative analysis chart of the searching performance of the embodiment of the invention and the method of the comparative literature.
FIG. 8 is a flow chart of an embodiment of the present invention.
Detailed Description
The method comprises four stages in total: the method comprises the steps of image DWT high-frequency sub-band acquisition, high-frequency sub-band coefficient BKF-VB-HMT modeling, image processing operation to be retrieved and similarity calculation. Specifically, as shown in fig. 8, the following steps are sequentially performed:
appointing: l denotes a low frequency subband obtained by Discrete Wavelet Transform (DWT), H denotes a high frequency subband;
Figure 725309DEST_PATH_IMAGE001
represents a coefficient vector; />
Figure 711719DEST_PATH_IMAGE002
Representing hidden states of the HMT model coefficients; />
Figure 752488DEST_PATH_IMAGE003
A mean covariance matrix; />
Figure 849757DEST_PATH_IMAGE004
Represents a gamma function; />
Figure 807217DEST_PATH_IMAGE005
Representing a set of image parameter features; k represents the number of wavelet coefficient vector trees; />
Figure 964529DEST_PATH_IMAGE006
Indicates the fifth->
Figure 492594DEST_PATH_IMAGE007
A set of hidden state variables of the vector tree; />
Figure 596816DEST_PATH_IMAGE008
Is the iteration number; />
Figure 284149DEST_PATH_IMAGE009
Representing the image characteristic parameter set after the parameter updating; />
Figure 739926DEST_PATH_IMAGE010
Represents a wavelet coefficient node->
Figure 879920DEST_PATH_IMAGE011
A parent node of (1); />
Figure 194358DEST_PATH_IMAGE012
Representing the wavelet transform subband coefficient edge state probability quality function,
Figure 736198DEST_PATH_IMAGE013
representing a joint probability quality function between parent nodes and child nodes of wavelet transform sub-band coefficients; />
Figure 625525DEST_PATH_IMAGE005
A set of parameters representing image feature values; v is an image texture feature library; i refers to an image to be retrieved; j refers to an image in the image library;
Figure 252816DEST_PATH_IMAGE014
is the relative entropy divergence distance between I and J; />
Figure 370944DEST_PATH_IMAGE015
The representative image I is in->
Figure 767291DEST_PATH_IMAGE011
The eigenvalues of the eigenvectors at the component positions; />
Figure 640569DEST_PATH_IMAGE017
The image features to be retrieved; />
Figure 879789DEST_PATH_IMAGE018
Features in the image feature library;
a. initial setting
Acquiring an image J, and performing parameter initialization operation;
b. high frequency subband acquisition
Respectively executing three stages of DWT on three color channels R, G and B of an image J, wherein each channel obtains 1L and 9H composed of matrixes with different sizes;
c. coefficient BKF-VB-HMT modeling
c.1 Binding the H coefficients of the same-scale, same-direction and same-position nodes in the R, G and B color channels into a coefficient vector
Figure 926242DEST_PATH_IMAGE001
；
c.2 Probability density function using BKF-VB-HMT model
Figure 52461DEST_PATH_IMAGE019
The 9H coefficient vectors per image were statistically modeled, device for selecting or keeping>
Figure 158958DEST_PATH_IMAGE019
The definition is as follows:
Figure 885474DEST_PATH_IMAGE020
；
in the formula (I), the compound is shown in the specification,
Figure 470039DEST_PATH_IMAGE021
、/>
Figure 716344DEST_PATH_IMAGE022
representing a shape parameter and a scale parameter; />
Figure 931424DEST_PATH_IMAGE023
In a second modified manner>
Figure 20603DEST_PATH_IMAGE024
The BKF function of order, calculated as follows:
Figure 270844DEST_PATH_IMAGE025
；
Figure 496288DEST_PATH_IMAGE026
is a state probability matrix, based on the state>
Figure 85533DEST_PATH_IMAGE027
Representing the scale after the discrete wavelet transform,
Figure 130849DEST_PATH_IMAGE028
，/>
Figure 447430DEST_PATH_IMAGE029
the initial state probability, the HMT model state transition probability matrix,
Figure 792961DEST_PATH_IMAGE030
is represented as follows:
Figure 21948DEST_PATH_IMAGE031
；
wherein
Figure 820139DEST_PATH_IMAGE032
Represents a hidden state value range, wherein>
Figure 753460DEST_PATH_IMAGE033
；
c.3 Formalizing parameters in the probability density function as quintuple
Figure 78131DEST_PATH_IMAGE005
Is expressed as follows: />
Figure 602654DEST_PATH_IMAGE034
；
c.4 Estimating the quintet using an Expectation Maximization (EM) algorithm according to a Maximum Likelihood (ML) criterion
Figure 29087DEST_PATH_IMAGE005
In the EM step, the coefficient set of each vector tree is ≥ v>
Figure 297257DEST_PATH_IMAGE001
As observation data, a->
Figure 476435DEST_PATH_IMAGE035
And/or>
Figure 437437DEST_PATH_IMAGE036
(ii) a In ML step, update->
Figure 413484DEST_PATH_IMAGE009
The parameters in (1) are:
Figure 95132DEST_PATH_IMAGE037
；
Figure 269761DEST_PATH_IMAGE038
；
Figure 275368DEST_PATH_IMAGE039
；
Figure 535448DEST_PATH_IMAGE040
；
Figure 286367DEST_PATH_IMAGE041
；
c.5 Parameter set of BKF-VB-HMT distribution model to be obtained
Figure 315503DEST_PATH_IMAGE005
Storing the image texture feature library V as a feature vector of each image J to be retrieved and used;
d. image processing operation to be retrieved
d.1 Inputting I, and respectively implementing three-level DWT conversion on three color components of R, G and B, wherein each color component is decomposed into 1L and 9H composed of matrixes with different sizes;
d.2 And c, repeating the step c, modeling the high-frequency sub-band coefficient of the image I to be retrieved through VB-HMT by utilizing BKF distribution, and calculating parameter characteristic quintuple
Figure 742942DEST_PATH_IMAGE005
Obtaining a feature vector of the image I;
e. similarity calculation
e.1 Definition of
Figure 224739DEST_PATH_IMAGE014
Represents the relative entropy divergence distance between I and J>
Figure 513769DEST_PATH_IMAGE015
Means I in>
Figure 600673DEST_PATH_IMAGE011
The characteristic value of a characteristic vector in the case of a component position, is evaluated>
Figure 74380DEST_PATH_IMAGE042
Is J in>
Figure 168107DEST_PATH_IMAGE011
The characteristic value of the characteristic vector at the component position is greater or less>
Figure 385462DEST_PATH_IMAGE043
Refers to the total number of images, so the relative entropy divergence distance between I and J can be calculated as follows:
Figure 264556DEST_PATH_IMAGE044
；/>
e.2 Arranging the similarity results from small to large according to the sequence, and outputting a better search result in a set search result range;
f. and d-e are repeated until the images I and J are completely matched.
Experimental testing and parameter setting:
the experiment is implemented in a Matlab R2011a environment, images related to the experiment are sourced from VisteX, brodatz, ALOT and STex texture image libraries, are published, and can be searched and downloaded on the internet. The sizes of the images in the library are different, and the design of the invention can process various images with different sizes, thereby achieving the same retrieval effect.
Fig. 1 is a graph of the fitting effect of wavelet domain BKF edge distribution PDF according to an embodiment of the present invention.
FIG. 2 is a PDF fitting effect diagram of a color image WD-BKF-VB-HMT model according to an embodiment of the invention.
Fig. 3 to 6 are graphs showing the result of searching the color texture image library according to the embodiment of the present invention.
Wherein FIG. 3 is the ALOTSPLIT image library search result; FIG. 4 is a Brodatz image library search result.
FIG. 5 is the result of STex image library search; fig. 6VisTex Split image library search results.
FIG. 7 is a comparative analysis chart of the searching performance of the embodiment of the invention and the method of the comparative literature.
The reference: chaorongLi, yuanyuanHuang, lihongZhu, color texture image reconstructed based on Gaussian copula models of Gabor, pattern Recognition, 2017, 64: 118-129.

Claims (1)

1. A color texture image retrieval method based on a non-Gaussian multi-correlation HMT model is characterized by comprising the following steps:
appointing: l represents a low-frequency subband obtained by discrete wavelet transform, H represents a high-frequency subband; x represents a coefficient vector; m and m' represent the hidden states of the HMT model coefficients; sigma refers to a covariance matrix; Γ () denotes the gamma function; theta represents an image parameter feature set; k represents the number of wavelet coefficient vector trees; s. the k A set of hidden state variables representing the kth vector tree; l is the number of iterations; theta l+1 Representing the image characteristic parameter set after the parameter updating; ρ (i) represents a parent node of wavelet coefficient node i;
Figure FDA0004065572150000011
represents the probability quality function of the edge state of the wavelet transform subband coefficient, < >>
Figure FDA0004065572150000012
Representing a joint probability quality function between parent nodes and child nodes of wavelet transform sub-band coefficients; v is an image texture feature library; i refers to an image to be retrieved; j refers to an image in the image library; d (I, J) is the relative entropy divergence distance between I and J; f. of i (I) Representing the characteristic value of the characteristic vector when the image I is at the position of the I component; a. initial setting
Acquiring an image J, and performing parameter initialization operation;
b. high frequency subband acquisition
Respectively executing three levels of DWT on three color channels of R, G and B of an image J, wherein each channel obtains 1L and 9H composed of matrixes with different sizes;
c. coefficient BKF-VB-HMT modeling
c.1 binding H coefficients of nodes with the same scale, the same direction and the same position in three color channels of R, G and B into a coefficient vector x;
c.2 statistically modeling 9H coefficient vectors per image using probability density function p of BKF-VB-HMT model, p being defined as follows:
Figure FDA0004065572150000013
in the formula, psi and Λ represent shape parameters and scale parameters; k v () for the second modified vkth BKF function, the calculation is as follows:
Figure FDA0004065572150000014
p is a state probability matrix, j represents the scale after discrete wavelet transformation,
Figure FDA0004065572150000021
h 1 as initial state probability, HMT model state transition probability matrix, A j Is represented as follows:
Figure FDA0004065572150000022
wherein M =1, 2.. M denotes a hidden state value range, wherein M =2;
c.3 formalizing the parameters in the probability density function in the form of the five-tuple Θ, as follows:
Figure FDA0004065572150000023
c.4, estimating parameters in the five-tuple theta by using a maximum expectation algorithm according to a maximum likelihood criterion, and in the EM step, calculating to obtain a coefficient set X of each vector tree as observation data
Figure FDA0004065572150000024
And with
Figure FDA0004065572150000025
In ML step, update Θ according to ML l+1 The parameters in (1) are:
Figure FDA0004065572150000026
Figure FDA0004065572150000027
Figure FDA0004065572150000028
Figure FDA0004065572150000029
Figure FDA00040655721500000210
c.5, storing the obtained parameter set theta of the BKF-VB-HMT distribution model into an image texture feature library V to serve as a feature vector of each image J to be retrieved for use;
d. image processing operation to be retrieved
d.1, inputting I, and respectively implementing three-level DWT conversion on the R, G and B color components, wherein each color component is decomposed into 1L and 9H composed of matrixes with different sizes;
d.2, repeating the step c, modeling the high-frequency sub-band coefficient of the image I to be retrieved through VB-HMT by utilizing BKF distribution, and calculating a parameter feature quintuple theta to obtain a feature vector of the image I;
e. similarity calculation
e.1 definition D (I, J) represents the relative entropy divergence distance between I and J, f i (I) The eigenvalues f of the eigenvectors of I at the I-component position i (J) Is J at the component bit of iThe temporal feature vector has a feature value, R refers to the total number of images, so the relative entropy divergence distance between I and J can be calculated as follows:
Figure FDA0004065572150000031
e.2, arranging the similarity results from small to large in sequence, and outputting a better search result in a set search result range;
f. and d-e are repeated until the images I and J are completely matched.
CN201811607246.XA 2018-12-27 2018-12-27 Color texture image retrieval method based on non-Gaussian multi-correlation HMT model Active CN109740008B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811607246.XA CN109740008B (en) 2018-12-27 2018-12-27 Color texture image retrieval method based on non-Gaussian multi-correlation HMT model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811607246.XA CN109740008B (en) 2018-12-27 2018-12-27 Color texture image retrieval method based on non-Gaussian multi-correlation HMT model

Publications (2)

Publication Number Publication Date
CN109740008A CN109740008A (en) 2019-05-10
CN109740008B true CN109740008B (en) 2023-04-14

Family

ID=66361383

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811607246.XA Active CN109740008B (en) 2018-12-27 2018-12-27 Color texture image retrieval method based on non-Gaussian multi-correlation HMT model

Country Status (1)

Country Link
CN (1) CN109740008B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101329736A (en) * 2008-06-20 2008-12-24 西安电子科技大学 Method of image segmentation based on character selection and hidden Markov model
CN103761727A (en) * 2013-12-28 2014-04-30 辽宁师范大学 Robust image segmentation method based on self-adaption Gaussian mixture model
CN108053360A (en) * 2017-12-18 2018-05-18 辽宁师范大学 The digital image watermark detection method of HMT models is closed based on multiphase
CN108090907A (en) * 2017-12-18 2018-05-29 辽宁师范大学 Image partition method based on non-gaussian HMT models

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8515171B2 (en) * 2009-01-09 2013-08-20 Rochester Institute Of Technology Methods for adaptive and progressive gradient-based multi-resolution color image segmentation and systems thereof

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101329736A (en) * 2008-06-20 2008-12-24 西安电子科技大学 Method of image segmentation based on character selection and hidden Markov model
CN103761727A (en) * 2013-12-28 2014-04-30 辽宁师范大学 Robust image segmentation method based on self-adaption Gaussian mixture model
CN108053360A (en) * 2017-12-18 2018-05-18 辽宁师范大学 The digital image watermark detection method of HMT models is closed based on multiphase
CN108090907A (en) * 2017-12-18 2018-05-29 辽宁师范大学 Image partition method based on non-gaussian HMT models

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Image denoising in dual contourlet domain using hidden Markov tree models;Hamid Reza Shahdoosti;《Digital Signal Processing》;20170427;全文 *
Weibull statistical modeling for textured image retrieval using nonsubsampled contourlet transform;hongying yang;《Soft Computing》;20180314;全文 *
基于二元BKF统计建模的双树复数小波域数字水印检测算法;王向阳;《计算机学报》;20171229;全文 *

Also Published As

Publication number Publication date
CN109740008A (en) 2019-05-10

Similar Documents

Publication Publication Date Title
Wang et al. Texture analysis and classification with linear regression model based on wavelet transform
Sorzano et al. A survey of dimensionality reduction techniques
Storath et al. Jump-sparse and sparse recovery using Potts functionals
CN111709435B (en) Discrete wavelet transform-based countermeasure sample generation method
Gu et al. Deep learning network for blind image quality assessment
Trujillo et al. Evolving estimators of the pointwise Hölder exponent with genetic programming
Peng Joint and direct optimization for dictionary learning in convolutional sparse representation
CN105426543A (en) Image retrieval method based on Bessel statistic model
Gorokhovatskyi et al. Transforming image descriptions as a set of descriptors to construct classification features
CN112231482A (en) Long and short text classification method based on scalable representation learning
CN105894462A (en) Remote sensing image de-noising method based on shearing-wave-domain hidden markov tree model
CN109740008B (en) Color texture image retrieval method based on non-Gaussian multi-correlation HMT model
CN108090914A (en) Color image segmentation method based on statistical modeling and pixel classifications
Mosleh et al. Texture image retrieval using contourlet transform
CN116343032A (en) Classification method combining Gaussian regression mixed model and MRF hyperspectral function data
CN105608470A (en) Image retrieval method based on Weibull distribution statistical model
Harini et al. Image retrieval system based on feature extraction and relevance feedback
Balestriero et al. Scattering decomposition for massive signal classification: from theory to fast algorithm and implementation with validation on international bioacoustic benchmark
CN106815314A (en) Image search method based on amplitude phase hybrid modeling
CN115937006A (en) Hyperspectral image denoising method based on Gaussian-Weishate prior
Azam et al. Bounded Laplace mixture model with applications to image clustering and content based image retrieval
CN108256581B (en) Gabor wavelet domain copula model image classification method
CN112766297A (en) Image classification method based on scalable representation learning
Zhao et al. Comparison of data reduction techniques based on SVM classifier and SVR performance
Gandhe et al. Image mining using wavelet transform

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