CN108615240B - Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight - Google Patents

Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight Download PDF

Info

Publication number
CN108615240B
CN108615240B CN201810432568.9A CN201810432568A CN108615240B CN 108615240 B CN108615240 B CN 108615240B CN 201810432568 A CN201810432568 A CN 201810432568A CN 108615240 B CN108615240 B CN 108615240B
Authority
CN
China
Prior art keywords
pixel
image
histogram
probability
formula
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.)
Expired - Fee Related
Application number
CN201810432568.9A
Other languages
Chinese (zh)
Other versions
CN108615240A (en
Inventor
唐宏
黄伟
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.)
Beijing Normal University
Original Assignee
Beijing 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 Beijing Normal University filed Critical Beijing Normal University
Priority to CN201810432568.9A priority Critical patent/CN108615240B/en
Publication of CN108615240A publication Critical patent/CN108615240A/en
Application granted granted Critical
Publication of CN108615240B publication Critical patent/CN108615240B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10041—Panchromatic image

Abstract

The invention discloses a non-parametric Bayesian segmentation method combining neighborhood information and distance weight, which comprises the following steps: firstly, collecting a panchromatic remote sensing image or a gray level image of a region to be detected, and carrying out local histogram feature extraction to obtain a histogram (hi) extracted by taking each pixel (i) as a center and a window with a certain size as a feature; then, clustering images by using a sampling mode according to the histogram characteristics to obtain an over-segmentation body; the horizontal axis of the histogram is a sequence formed by pixel values of all pixels in the image; the vertical axis is the number of times each pixel value appears in a window with a certain size; finally, image clustering is carried out on the panchromatic image by using a sampling mode to obtain a panchromatic image segmentation body, and the method comprises the following steps: calculating a parameter vector theta by using a first formulaiLikelihood term of (hi | θ)i). The method fully and effectively utilizes the spatial distance information and the pixel neighborhood information in the panchromatic image, and improves the objectivity and scientificity of the panchromatic image.

Description

Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight
Technical Field
The invention relates to the technical field of image segmentation, in particular to a non-parametric Bayesian segmentation method combining neighborhood information and distance weight.
Background
Image segmentation, a fundamental problem in the field of computer vision, is an important component of image understanding. Meanwhile, the method plays an important role in a plurality of fields such as image processing, pattern recognition and artificial intelligence.
In recent years, over-segmentation has been successfully applied to the field of image segmentation as a rapidly emerging image preprocessing technique. In essence, over-segmentation is the local clustering of pixels in an image. Because the natural image is very complex and the information contained in a single pixel is very limited, the pixels in the image are gathered into small blocks and used as subsequent processing units, so that the execution efficiency of subsequent tasks can be greatly improved, and the over-segmentation body can group the pixels by utilizing the similarity of the characteristics between the pixels, and can acquire more local information of the image compared with the pixels, thereby greatly improving the execution effect of the subsequent tasks and greatly reducing the complexity of the subsequent image processing tasks. Moreover, the pixels in each class of the over-segmentation result have similar image content information such as color, texture, etc., so that the segmentation result can keep most of the boundary of the object in the image, and thus the over-segmentation method is widely applied to the fields of image processing and computer vision.
The existing over-segmentation method is a top-down global segmentation method based on graph theory; there is a gradient descent method starting from the initial coarse clustering; a boundary optimization method for uniformly dividing an image into required over-segmentation bodies as algorithm initialization is provided; a national wind human body generation algorithm based on a level set from inside to outside is also provided; the over-segmentation results generated by the above method, however, depend largely on the initialization of the image and the setting of key parameters. Aiming at different production and experimental requirements, the existing method has strong artificial subjectivity on the setting of the initialization state and the parameters of the image. In addition, for example, in the bottom-up multi-scale segmentation method, parameters such as the segmentation scale, the shape factor, and the compactness factor of the over-segmented region object need to be input by the user to ensure the reliability of the segmentation result.
Disclosure of Invention
The present invention aims to provide a non-parametric bayesian segmentation method combining neighborhood information and distance weight to solve the problems proposed in the background art.
In order to achieve the purpose, the invention provides the following technical scheme: a non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight comprises the following steps:
firstly, collecting a panchromatic remote sensing image of a region to be detected, carrying out local histogram feature extraction, and obtaining a histogram h extracted by taking each pixel i as a center and taking a window with a certain sizeiAs a feature; then, clustering images by using a sampling mode according to the histogram characteristics to obtain an over-segmentation body;
the horizontal axis of the histogram is a sequence formed by pixel values of all pixels in the image; the vertical axis is the number of times each pixel value appears in a window with a certain size;
and finally, carrying out image clustering on the panchromatic remote sensing image by using a sampling mode to obtain a panchromatic remote sensing image segmentation body, wherein the panchromatic remote sensing image segmentation body comprises the following steps:
calculating a parameter vector theta using a first formulaiLikelihood term F (h)i|θi)；
The first formula is:
Figure GDA0002362061850000021
wherein, thetaijIs the probability of occurrence of each abscissa value j in the histogram feature of each pixel i; h isijThe number of occurrences of each pixel value in a histogram generated for a window of a certain size; n is a radical ofcountsThe number of data values recorded in the abscissa of each histogram; n is a radical ofbinsThe number of groups for histogram abscissa; zM(hi) Is equal to thetaijAn unrelated normalization function;
calculating a parameter vector theta using a second formulaiA priori term of (G)0(θi|βπ)；
The second formula is:
Figure GDA0002362061850000031
wherein β is a positive scale factor parameter and π is NbinsA probability vector of dimensions; zD(βπj) Is equal to thetaijAn independent normalization function, () is a gamma function sign, which satisfies (x +1) x (x +1),
Figure GDA0002362061850000032
and calculating the distance between each pixel and the cluster center of the class existing in the peripheral neighborhood by adopting a third formula:
Figure GDA0002362061850000033
where k is 1 …, Nc, xi、yiFor the ordinate and abscissa values of each pixel，xk、ykThe horizontal and vertical coordinate values of the center of each cluster are taken, and Nc is the total number of clusters;
calculating the posterior probability of each class by adopting a fourth formula;
due to the fact that
Figure GDA0002362061850000034
Because:
Figure GDA0002362061850000035
the fourth formula is:
Figure GDA0002362061850000041
wherein Nc is the number of categories existing in the neighborhood;
l is a certain cluster class already existing in Nc, qilA probability value for the ith pixel to be classified as the ith cluster;
q0probability of new clusters;
α is a scale parameter less than 1, and controls the probability of obtaining new category by sampling each pixel;
G0is a base measure of the dirichlet process;
Ω0measure of basis G0A limited domain in which it resides;
qiksampling the probability of the ith pixel to the kth cluster;
m is an interactive item function of the Markov random field;
П is the distribution of the Markov random field;
Fdis a distance metric function;
Fha histogram metric function;
Lithe position of the ith pixel;
Lkthe position of the kth clustering center;
Pikcalculating a result for the probability value of the kth cluster;
Figure GDA0002362061850000042
in order to solve the clustering process, parameters of polynomial distribution sampled in the Dirichlet process are obtained;
n-k ithe total number of the kth class except the ith pixel class in the image;
Figure GDA0002362061850000043
in the k-th class in histogram hkThe probability of occurrence of the jth pixel value;
ZMis the AND parameter β, pi, and the observation vector hiThe independent polynomial allocation functions are used for simplifying the function form and calculating the probability value of the new class;
ZDis the AND parameter β, pi, and the observation vector hiThe distribution functions of all irrelevant dirichlet processes are used for simplifying the function form and calculating to obtain the probability value of a new class;
hithe statistical vector of the number of pixel values formed by 8 neighborhood pixels around the ith image is a vector with the length of 256, the vector represents 0-255 pixel values respectively, and if a certain pixel value appears in 8 neighborhoods, the number of corresponding positions is increased by 1.
Compared with the prior art, the invention has the beneficial effects that: the method can combine neighborhood information of the panchromatic image with a non-parametric Bayes framework, fully and effectively utilize spatial distance information and pixel neighborhood information in the panchromatic image, cluster the regions in a sampling mode, automatically deduce the number of the inherent over-segmentation bodies of the panchromatic image, set algorithm parameters related to the over-segmentation bodies without manual experience, and improve the objectivity and scientificity of the image over-segmentation method.
Drawings
FIG. 1 is a flow chart of the non-parametric Bayesian segmentation method combining neighborhood information and distance weight according to the present invention.
FIG. 2 is a diagram illustrating the result of the non-parametric Bayesian segmentation method combining neighborhood information and distance weight according to the present invention.
FIG. 3 is a diagram illustrating an index for evaluating an over-segmentation result according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a technical scheme that:
example 1:
a non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight comprises the following steps:
firstly, collecting a panchromatic remote sensing image or a gray level image of a region to be detected, and carrying out local histogram feature extraction to obtain a histogram (hi) extracted by taking each pixel (i) as a center and a window with a certain size as a feature; then, clustering images by using a sampling mode according to the histogram characteristics to obtain an over-segmentation body;
the horizontal axis of the histogram is a sequence formed by pixel values of all pixels in the image; the vertical axis is the number of times each pixel value appears in a window with a certain size;
finally, image clustering is carried out on the panchromatic image by using a sampling mode to obtain a panchromatic image segmentation body, and the method comprises the following steps:
calculating a parameter vector theta by using a first formulaiLikelihood term of (hi | θ)i)；
The first formula is:
Figure GDA0002362061850000061
wherein, thetaijIs the probability of occurrence of each abscissa value j in the histogram feature of each pixel i; h isijThe number of occurrences of each pixel value in a histogram generated for a window of a certain size; n is a radical ofcountThe number of data values recorded in the abscissa of each histogram; bins is the number of histogram abscissa bins; zM(hi) Is equal to thetaijAn unrelated normalization function;
calculating a parameter vector theta using a second formulaiA priori term of (G)0(θi|βπ)。
The second formula is:
Figure GDA0002362061850000062
wherein β is a positive scale factor parameter, pi is a bins-dimensional probability vector, and ZD(βπi) Is equal to thetaijAn unrelated normalization function;
calculating the distance between each pixel and the clustering center of the class existing in the peripheral neighborhood by adopting a third formula;
Figure GDA0002362061850000063
where k is 1 …, Nc, xi、yiFor each pixel, xk、ykThe horizontal and vertical coordinate values of the center of each class;
and calculating the posterior probability of each class by adopting a fourth formula.
Due to the fact that
Figure GDA0002362061850000071
Because:
Figure GDA0002362061850000072
the fourth formula is:
Figure GDA0002362061850000073
example 2:
according to the non-parametric bayesian segmentation method combining neighborhood information and distance weight described in embodiment 1, as shown in fig. 1, the classification method for fusing a panchromatic image and a multispectral image in this embodiment is as follows, and a high-resolution remote sensing image of a region to be measured is acquired.
It should be understood that a high-resolution remote sensing image of the region to be measured is acquired by a satellite, and the remote sensing image is a full color image.
And marking the panchromatic image pixel by pixel to obtain an initialized image.
It should be understood that the panchromatic image is labeled with a category label pixel by pixel to obtain an initial clustering result of the panchromatic image, and specifically, the panchromatic image is labeled with a non-repetitive category label for each pixel to obtain N over-segmentation volumes of the panchromatic image, where N is the number of pixels of the panchromatic image.
It will also be appreciated that the full-color image is divided, each pixel being treated as an over-segment, and the over-segment is pre-processed for subsequent acquisition.
And performing characteristic extraction of a local histogram on the panchromatic image, and determining an observed value.
It will be appreciated that the full-color image is traversed pixel-by-pixel, while the gray values in a window of a certain size are extracted and counted for frequency of occurrence, determining a local histogram for each pixel and its neighborhood.
It will also be appreciated that grey value statistics are performed in the 8 neighbourhood of each pixel of the full colour image, allowing a single grey value extension to be achieved, resulting in an observation vector (hi) in the form of a local histogram.
It should be understood that the extracted features are no longer gray values for a single pixel, but rather an observation vector of 9 pixels centered on each pixel.
According to the initialization parameters β and pi, a Dirichlet Process (DP) is defined, and a parameter vector θ i is obtained by sampling according to a first formula.
Wherein the first formula is:
Figure GDA0002362061850000081
where θ i is the probability of occurrence of each abscissa value j in the ith pixel local histogram feature, hi is the number of occurrences of each pixel value in the histogram generated for a window of a certain size, β is a positive scale factor parameter, and π is a multi-dimensional probability vector.
It will be appreciated that the parameter vector θ i is sampled according to an initialized dirichlet process.
According to the parameters β and pi and the observation vector hi, the probability q0 that each pixel sample obtains a new category is calculated by adopting a second formula
Wherein the second formula is:
Figure GDA0002362061850000082
ZD is a normalization function independent of parameters β, pi, and observation vector hi α is a positive number less than 1, and controls the probability of each pixel sample to get a new class.
It will be appreciated that the magnitude of q0 is dependent on α in addition to β, pi, hi.
And according to the parameters β and pi and the observation vector hi, calculating the probability qk of the existing class k being 1, 2 … and Nc in the neighborhood by adopting a third formula.
Wherein the third formula is:
Figure GDA0002362061850000091
where nk-i represents the number of samples belonging to the kth class, and dik is the distance between each pixel and the cluster center of the existing class of the peripheral neighborhood.
It should be understood that qk is determined by the distance to the center of the cluster in combination with the likelihood term F (hi | θ i) of the parameter vector θ i.
Defining probability distribution Pik of category attribution by adopting a fourth formula according to q0 and qk
Wherein the fourth formula is:
Figure GDA0002362061850000092
where Nc is the number of categories present in the neighborhood.
And sampling to obtain the category k of each pixel position by adopting a fifth formula according to the probability distribution Pik of the category attribution.
Wherein the fifth formula is:
k～Pik
109. and if k is not equal to 0, a new category label is created for the current pixel, and if k is not equal to 0, the cluster center coordinate and the cluster parameter thetak are updated by adopting a sixth formula and a seventh formula according to a new clustering result.
Wherein the sixth formula is:
Figure GDA0002362061850000101
where k is 1 …, Nc, xi, yi are horizontal and vertical coordinate values of each pixel, and xk, yk are horizontal and vertical coordinate values of each class center.
The seventh formula is:
Figure GDA0002362061850000102
wherein S is the category given by the last clustering result.
It should be understood that the update of the clustering parameter θ k is obtained by sampling according to the class label given by the last clustering result.
Experimental images: and (4) drawing a satellite image, wherein the resolution of the panchromatic image is two meters.
Area: beijing dense cloud, rural area, as shown in FIG. 2, is a schematic diagram of a full-color image of a rural area shot by a sky-plot satellite, the size of which is as follows: 400*400
The evaluation method for the image over-segmentation comprises the following steps:
1) edge recall: the ratio of the number of over-partition volume edge pixels falling within at least one distance (usually 2 pixels) of the true edge pixels to the total number of the true edge pixels is referred to. In general, the greater the edge recall, the better the edge fit.
2) Under-segmentation error rate: the proportion of the over-partition body region 'overflow' true value region boundary is measured.
3) Optimal target segmentation accuracy: the asa (acceptable segmentation accuracy) standard, defined as the best target segmentation accuracy achievable by over-segmentation as a pre-processing procedure, is a method to "back" evaluate superpixels from the final segmentation result.
4) Landscape crushing degree: is an important component of landscape heterogeneity, which refers to the degree of fragmentation that the landscape is divided into.
The image classification method in the prior art mainly comprises the following steps:
(1) an over-segmentation method based on graph theory; the segmentation problem is converted into an energy function minimization problem, pixel points in the image are taken as graph nodes, edges among the nodes are given with weights, and then the graph is divided by adopting various segmentation criteria, so that an over-segmentation body is formed.
(2) A gradient-based over-segmentation method; and starting from the initial pixel clustering, iteratively correcting the clustering result by adopting a gradient method until a convergence condition is met, thereby forming an over-segmentation body.
The experimental result based on the daily plot of the image of the rural area is a schematic diagram of the over-segmentation result of the panchromatic image shown in fig. 3.
And (4) qualitative evaluation result: firstly, the non-parametric Bayes image over-segmentation method combining the pixel neighborhood information and the distance from the pixel to the clustering center provided by the invention can fully utilize the pixel neighborhood information and the spatial information of the clustering centers with different distances.
From the quantitative point of view, the NBIC performance is superior to that of other methods, the evaluation index corresponding to the over-segmentation result obtained by the method is highest, the error rate is low, and the over-segmentation result obtained by the method is represented to be the highest corresponding consistency degree with the real ground object boundary, namely the over-segmentation accuracy is the highest.
By the non-parametric Bayes image over-segmentation method combining the pixel neighborhood information and the distance from the pixel to the clustering center, the neighborhood information of the panchromatic image can be combined with a non-parametric Bayes framework, the spatial distance information and the pixel neighborhood information in the panchromatic image are fully and effectively utilized, the region is clustered in a sampling mode, the inherent number of over-segmentation bodies of the panchromatic image is automatically deduced, the algorithm parameters related to the over-segmentation bodies do not need to be set through manual experience, and the objectivity and the scientificity of the image over-segmentation method are improved.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (1)

1. A non-parametric Bayes over-segmentation method combining neighborhood information and distance weight is characterized in that: the method comprises the following steps:
firstly, collecting a panchromatic remote sensing image of a region to be detected, carrying out local histogram feature extraction, and obtaining a histogram h extracted by taking each pixel i as a center and taking a window with a certain sizeiAs a feature; then, clustering images by using a sampling mode according to the histogram characteristics to obtain an over-segmentation body;
the horizontal axis of the histogram is a sequence formed by pixel values of all pixels in the image; the vertical axis is the number of times each pixel value appears in a window with a certain size;
and finally, carrying out image clustering on the panchromatic remote sensing image by using a sampling mode to obtain a panchromatic remote sensing image segmentation body, wherein the panchromatic remote sensing image segmentation body comprises the following steps:
calculating a parameter vector theta using a first formulaiLikelihood term F (h)i|θi)；
The first formula is:
Figure FDA0002362061840000011
wherein, thetaijIs the probability of occurrence of each abscissa value j in the histogram feature of each pixel i; h isijThe number of occurrences of each pixel value in a histogram generated for a window of a certain size; n is a radical ofcountsThe number of data values recorded in the abscissa of each histogram; n is a radical ofbinsThe number of groups for histogram abscissa; zM(hi) Is equal to thetaijAn unrelated normalization function;
calculating a parameter vector theta using a second formulaiA priori term of (G)0(θi|βπ)；
The second formula is:
Figure FDA0002362061840000012
wherein β is a positive scale factor parameter and π is NbinsA probability vector of dimensions; zD(βπj) Is equal to thetaijAn independent normalization function, () is a gamma function sign, which satisfies (x +1) x (x +1),
Figure FDA0002362061840000021
and calculating the distance between each pixel and the cluster center of the class existing in the peripheral neighborhood by adopting a third formula:
Figure FDA0002362061840000022
where k is 1 …, Nc, xi、yiFor each pixel, xk、ykThe horizontal and vertical coordinate values of the center of each cluster are taken, and Nc is the total number of clusters;
calculating the posterior probability of each class by adopting a fourth formula;
due to the fact that
Figure FDA0002362061840000023
Because:
Figure FDA0002362061840000024
the fourth formula is:
Figure FDA0002362061840000025
wherein Nc is the number of categories existing in the neighborhood;
l is a certain cluster class already existing in Nc, qilA probability value for the ith pixel to be classified as the ith cluster;
q0probability of new clusters;
α is a scale parameter less than 1, and controls the probability of obtaining new category by sampling each pixel;
G0is a base measure of the dirichlet process;
Ω0measure of basis G0A limited domain in which it resides;
qiksampling the probability of the ith pixel to the kth cluster;
m is an interactive item function of the Markov random field;
П is the distribution of the Markov random field;
Fdis a distance metric function;
Fha histogram metric function;
Lithe position of the ith pixel;
Lkthe position of the kth clustering center;
Pikcalculating a result for the probability value of the kth cluster;
Figure FDA0002362061840000031
in order to solve the parameter of the polynomial distribution sampled in the Dirichlet process in the clustering process;
Figure FDA0002362061840000032
the total number of the kth class except the ith pixel class in the image;
Figure FDA0002362061840000033
in the k-th class in histogram hkThe probability of occurrence of the jth pixel value;
ZMis the AND parameter β, pi, and the observation vector hiThe independent polynomial allocation functions are used for simplifying the function form and calculating the probability value of the new class;
ZDis the AND parameter β, pi, and the observation vector hiThe distribution functions of all irrelevant dirichlet processes are used for simplifying the function form and calculating to obtain the probability value of a new class;
hithe statistical vector of the number of pixel values formed by 8 neighborhood pixels around the ith image is a vector with the length of 256, the vector represents 0-255 pixel values respectively, and if a certain pixel value appears in 8 neighborhoods, the number of corresponding positions is increased by 1.
CN201810432568.9A 2018-05-08 2018-05-08 Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight Expired - Fee Related CN108615240B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810432568.9A CN108615240B (en) 2018-05-08 2018-05-08 Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810432568.9A CN108615240B (en) 2018-05-08 2018-05-08 Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight

Publications (2)

Publication Number Publication Date
CN108615240A CN108615240A (en) 2018-10-02
CN108615240B true CN108615240B (en) 2020-07-21

Family

ID=63662166

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810432568.9A Expired - Fee Related CN108615240B (en) 2018-05-08 2018-05-08 Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight

Country Status (1)

Country Link
CN (1) CN108615240B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111161299B (en) * 2018-11-08 2023-06-30 深圳富泰宏精密工业有限公司 Image segmentation method, storage medium and electronic device
CN113837315B (en) * 2021-10-11 2023-05-23 南昌航空大学 Fault diagnosis method based on hierarchical link infinite hidden Markov model

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914846A (en) * 2014-04-10 2014-07-09 西安电子科技大学 SAR image segmentation method based on triple discrimination random field
CN104766091A (en) * 2015-03-25 2015-07-08 北京师范大学 Space and spectrum synergistic structure and remote-sensing image multi-scale comprehension method and system
CN107730516A (en) * 2017-10-16 2018-02-23 江南大学 A kind of brain MR image division methods based on fuzzy clustering

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101286229B (en) * 2008-05-05 2010-06-02 哈尔滨工程大学 Sonar image self-adapting division method based on stratified MRF
US9414016B2 (en) * 2013-12-31 2016-08-09 Personify, Inc. System and methods for persona identification using combined probability maps
US9454806B2 (en) * 2014-01-21 2016-09-27 Nvidia Corporation Efficient approximate-nearest-neighbor (ANN) search for high-quality collaborative filtering
CN106651865B (en) * 2016-12-23 2019-10-08 湖北工业大学 A kind of optimum segmentation scale automatic selecting method of new high-resolution remote sensing image

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914846A (en) * 2014-04-10 2014-07-09 西安电子科技大学 SAR image segmentation method based on triple discrimination random field
CN104766091A (en) * 2015-03-25 2015-07-08 北京师范大学 Space and spectrum synergistic structure and remote-sensing image multi-scale comprehension method and system
CN107730516A (en) * 2017-10-16 2018-02-23 江南大学 A kind of brain MR image division methods based on fuzzy clustering

Also Published As

Publication number Publication date
CN108615240A (en) 2018-10-02

Similar Documents

Publication Publication Date Title
CN110210463B (en) Precise ROI-fast R-CNN-based radar target image detection method
CN109446992B (en) Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment
CN111428748B (en) HOG feature and SVM-based infrared image insulator identification detection method
CN107330897B (en) Image segmentation method and system
CN109215034B (en) Weak supervision image semantic segmentation method based on spatial pyramid covering pooling
CN106909902B (en) Remote sensing target detection method based on improved hierarchical significant model
CN110264448B (en) Insulator fault detection method based on machine vision
CN105139015B (en) A kind of remote sensing images Clean water withdraw method
CN109446894B (en) Multispectral image change detection method based on probability segmentation and Gaussian mixture clustering
CN107067405B (en) Remote sensing image segmentation method based on scale optimization
CN105574534A (en) Significant object detection method based on sparse subspace clustering and low-order expression
CN110766058B (en) Battlefield target detection method based on optimized RPN (resilient packet network)
CN103246894B (en) A kind of ground cloud atlas recognition methods solving illumination-insensitive problem
CN111652317A (en) Hyper-parameter image segmentation method based on Bayesian deep learning
CN109801305B (en) SAR image change detection method based on deep capsule network
CN111160407A (en) Deep learning target detection method and system
US11037299B2 (en) Region merging image segmentation algorithm based on boundary extraction
CN108596920A (en) A kind of Target Segmentation method and device based on coloured image
CN108615240B (en) Non-parametric Bayesian over-segmentation method combining neighborhood information and distance weight
CN110070545A (en) A kind of method that textural characteristics density in cities and towns automatically extracts cities and towns built-up areas
Zhu et al. Change detection based on the combination of improved SegNet neural network and morphology
CN108230313B (en) SAR image target detection method based on component adaptive selection and discrimination model
CN111199245A (en) Rape pest identification method
CN109558803B (en) SAR target identification method based on convolutional neural network and NP criterion
CN111310566A (en) Static and dynamic multi-feature fusion mountain fire detection method and system

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
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200721