CN111178361A - Interactive image segmentation method based on label group propagation - Google Patents

Interactive image segmentation method based on label group propagation Download PDF

Info

Publication number
CN111178361A
CN111178361A CN201911261106.6A CN201911261106A CN111178361A CN 111178361 A CN111178361 A CN 111178361A CN 201911261106 A CN201911261106 A CN 201911261106A CN 111178361 A CN111178361 A CN 111178361A
Authority
CN
China
Prior art keywords
super
pair
superpixel
pixel
probability
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.)
Pending
Application number
CN201911261106.6A
Other languages
Chinese (zh)
Inventor
孙权森
齐圣哲
王涛
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.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
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 Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN201911261106.6A priority Critical patent/CN111178361A/en
Publication of CN111178361A publication Critical patent/CN111178361A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an interactive image segmentation method based on label group propagation. The method comprises the following steps: firstly, dividing an image into a plurality of superpixels, constructing superpixel pairs, and constructing a relation matrix of the superpixel pairs; then calculating the initial probability that each super pixel pair belongs to the label pair; then, the initial probability of the super pixel pairs is transmitted to the whole by utilizing the relation matrix of the super pixel pairs; and finally, determining the category of the superpixel according to the superpixel pair probability transmitted to the whole situation, and finishing image segmentation. The method can better detect the object contour under limited prior knowledge, and improves the segmentation effect of low-contrast and weak boundary images.

Description

Interactive image segmentation method based on label group propagation
Technical Field
The invention relates to the technical field of interactive image segmentation, in particular to an interactive image segmentation method based on label group propagation.
Background
The interactive image segmentation is to separate an object interested by a user from a complex image background environment based on a certain similarity criterion under the prior knowledge provided by the user, and is a key problem in the fields of image analysis, pattern recognition, computer vision and the like, and the quality of segmentation quality directly influences subsequent related applications.
The method based on the graph theory has the excellent characteristics of multi-feature fusion, global optimization, high execution efficiency and the like, so that the method has received wide attention of scholars at home and abroad in recent years, and becomes one of the mainstream interactive image segmentation methods. When a user provides enough seed points, the traditional graph theory-based method can obtain better segmentation results when a simple color image is segmented. However, this type of method is sensitive to the number and location of seed points, and it is difficult to obtain satisfactory segmentation results when user interaction is limited. In addition, the traditional graph theory method only utilizes the local pixel relation of the image to guide the segmentation, and the image containing noise and texture is difficult to segment because the structural information of the image is not fully utilized. These defects seriously affect the accuracy and robustness of the method and limit the application range of the method.
Disclosure of Invention
The invention aims to provide an interactive image segmentation method based on label group propagation, which has strong applicability, good segmentation effect, high efficiency and strong robustness.
The technical solution for realizing the purpose of the invention is as follows: an interactive image segmentation method based on label group propagation comprises the following steps:
step 1, dividing an image into a plurality of super pixels and constructing a super pixel pair;
step 2, constructing a relation matrix of the superpixel pairs;
step 3, calculating the initial probability of each super pixel pair belonging to the label pair;
step 4, transferring the initial probability of the super pixel pair to the whole situation by using the relation matrix of the super pixel pair;
and 5, determining the category of the superpixel according to the superpixel pair probability transmitted to the whole situation, and finishing image segmentation.
Further, the step 1 of dividing the image into a plurality of super pixels and constructing super pixel pairs includes the following specific steps:
step 1.1, dividing an image into M super pixels by using an unsupervised segmentation algorithm;
step 1.2, for any super pixel, combining with the rest M-1 super pixels and the super pixels to construct M-1 super pixel pairs, wherein M super pixels construct M2A super pixel pair.
Further, the step 2 constructs a relation matrix of the super-pixel pairs, specifically as follows:
step 2.1, taking the average value of the characteristics of all pixel points in the super pixel as the characteristics of the super pixel;
and 2.2, acquiring the neighborhood relation of the superpixels, and establishing a relation matrix of the superpixel pairs according to the characteristic values by using a Gaussian function.
Further, the step 3 of calculating the initial probability that each superpixel pair belongs to a tag pair specifically includes the following steps:
performing multivariate relation learning through a high-order tensor product graph, capturing the interaction between the super pixels of the image, estimating the initial probability of each super pixel pair belonging to a label pair, and adding a high-order graph Laplacian operator to smooth a prior label;
superpixel pairs s based on fusion relationshipsijBelong to the label pair lmnPrior probability of (2)
Figure BDA0002311596230000021
The definition is as follows:
Figure BDA0002311596230000022
where i, j is 1, … M, pair of superpixels sijFrom super-pixels siAnd SjForm, label pair lmnFrom a label lmAnd lnConstitution p(s)i∈lm|sj∈ln) For conditional probability, the following is defined:
Figure BDA0002311596230000023
wherein,
Figure BDA0002311596230000024
representing a super-pixel SiAnd SjSimilarity between them, the greater its value, the higher the similarity; l represents the number of tags;
Figure BDA0002311596230000025
is a super pixel SiBelong to the label lmIs defined as follows:
Figure BDA0002311596230000026
wherein, ciIs a super pixel siIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA0002311596230000027
for clustering algorithm to label lmAnd (4) clustering the seed point characteristics to obtain the kth clustering center, wherein K is the total number of the clustering centers.
Further, the step 4 of transferring the initial probability of the superpixel pair to the global by using the relation matrix of the superpixel pair specifically includes:
P(t)＝αQP(t-1)+(1-α)P
q is a row normalization matrix of the super-pixel pair relation matrix obtained in the step 2, P is an initial probability matrix of the super-pixel pair, t is iteration times, and P is(t)and alpha is a regulation parameter as a result of the tth iteration of the P.
Further, step 5 determines the class of the superpixel according to the superpixel pair probability transmitted to the global, and completes the image segmentation, specifically as follows:
when determining the super pixel category, according to the sum of the element pair probabilities of the super pixel and the super pixel itself and other M-1 super pixels, if the probability that the element pairs belong to the target is greater than the probability that the element pairs belong to the background, the super pixel belongs to the target, otherwise, the super pixel belongs to the background, and the classification is finished according to the probability.
Compared with the prior art, the invention has the remarkable advantages that: (1) extending the local neighborhood relationship to the global relationship on the image and exploring more complex interactions between element pairs and label pairs, which facilitates better detection of object contours with limited a priori knowledge; (2) the label priors are propagated according to the structure of a high-order tensor product map, and can capture the inherent data structure between image element pairs, which helps to better overcome the problems of low contrast and weak boundaries; (3) by adopting the non-local relation, the internal global relation of the image is captured by transmitting the similarity relation between the adjacent pixels, and the image segmentation effect is improved.
Drawings
FIG. 1 is a flow chart of an interactive image segmentation method based on tag group propagation according to the present invention.
FIG. 2 is a graph comparing the segmentation results obtained using the method of the present invention and the prior art method in the example of the present invention.
Detailed Description
The invention constructs a unified multivariate label group relation propagation framework, performs multivariate relation learning based on a high-order tensor product diagram, and naturally derives a binary label pair propagation algorithm. The method comprises the following steps: firstly, in order to ensure the efficiency of image segmentation, taking a super-pixel pair as a unit, dividing an image into a plurality of super-pixels by using the existing unsupervised algorithm, and constructing a super-pixel pair set; then, performing multivariate relation learning through a high-order tensor product diagram, capturing the interaction between image elements (superpixels), and estimating the prior probability of each superpixel pair belonging to a label pair; then, designing an iterative label pair diffusion model to improve optimization efficiency; finally, the individual superpixels are classified (foreground/background) by univariate probability transformation.
With reference to fig. 1, the present invention provides an interactive image segmentation method based on tag group propagation, which includes the following steps:
step 1, dividing an image into a plurality of super pixels and constructing a super pixel pair;
step 2, constructing a relation matrix of the superpixel pairs;
step 3, calculating the initial probability of each super pixel pair belonging to the label pair;
step 4, transferring the initial probability of the super pixel pair to the whole situation by using the relation matrix of the super pixel pair;
and 5, determining the category of the superpixel according to the superpixel pair probability transmitted to the whole situation, and finishing image segmentation.
As a specific implementation manner, the step 1 divides the image into a plurality of super pixels, and constructs a super pixel pair, specifically as follows:
step 1.1, dividing an image into M super pixels by using an unsupervised segmentation algorithm;
step 1.2, for any super pixel, combining with the rest M-1 super pixels and the super pixels to construct M-1 super pixel pairs, wherein M super pixels construct M2A super pixel pair.
As a specific implementation, the constructing the relationship matrix of the super-pixel pairs in step 2 is as follows:
step 2.1, taking the average value of the characteristics of all pixel points in the super pixel as the characteristics of the super pixel;
and 2.2, acquiring the neighborhood relation of the superpixels, and establishing a relation matrix of the superpixel pairs according to the characteristic values by using a Gaussian function.
As a specific implementation, the step 3 of calculating the initial probability that each superpixel pair belongs to a tag pair specifically includes the following steps:
performing multivariate relation learning through a high-order tensor product graph, capturing the interaction between the super pixels of the image, estimating the initial probability of each super pixel pair belonging to a label pair, and adding a high-order graph Laplacian operator to smooth a prior label;
superpixel pairs s based on fusion relationshipsijBelong to the label pair lmnPrior probability of (2)
Figure BDA0002311596230000041
The definition is as follows:
Figure BDA0002311596230000042
where i, j is 1, … M, pair of superpixels sijBy super-imageElemental SiAnd sjForm, label pair lmnFrom a label lmAnd lnConstitution p(s)i∈lm|sj∈ln) For conditional probability, the following is defined:
Figure BDA0002311596230000043
wherein,
Figure BDA0002311596230000044
representing a super-pixel siAnd sjSimilarity between them, the greater its value, the higher the similarity; l represents the number of tags;
Figure BDA0002311596230000045
is a super pixel SiBelong to the label lmIs defined as follows:
Figure BDA0002311596230000046
wherein, CiIs a super pixel SiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA0002311596230000047
for clustering algorithm to label lmAnd (4) clustering the seed point characteristics to obtain the kth clustering center, wherein K is the total number of the clustering centers.
As a specific implementation, the relationship matrix of the superpixel pair in step 4 is used to transfer the initial probability of the superpixel pair to the global, which specifically includes the following steps:
P(t)＝αQP(t-1)+(1-α)P
q is a row normalization matrix of the super-pixel pair relation matrix obtained in the step 2, P is an initial probability matrix of the super-pixel pair, t is iteration times, and P is(t)and alpha is a regulation parameter as a result of the tth iteration of the P.
As a specific implementation manner, the determining the class of the superpixel according to the superpixel pair probability passed to the global in step 5, and completing the image segmentation specifically includes:
in the method, when a certain super pixel class is determined, according to the sum of the probabilities of the super pixel and the elements of the super pixel and other M-1 super pixels, if the probability that the element pairs belong to a target is greater than the probability that the element pairs belong to a background, the super pixel belongs to the target, otherwise, the super pixel belongs to the background, and the classification is finished according to the probability.
The method has the following advantages: due to the introduction of high-order information, the relationship between unmarked data and marked data can be more accurately learned based on a high-order tensor product diagram; secondly, in the tag group propagation process, a finer relationship between the image element group and the tag group can be obtained based on more complex interactions between the image elements.
The invention is described in further detail below with reference to the figures and specific examples.
Example 1
In the embodiment, a common color image is used as input, a plurality of super pixels are generated by using a SLIC algorithm, and a super pixel pair is constructed; taking the pixel characteristic mean value in the superpixels as superpixel characteristics, and calculating a relation matrix of superpixel pairs; obtaining foreground and background seed points through user line drawing interaction, and obtaining pixel prior probability on the basis; taking the probability mean value of all pixels in the superpixels as the probability of the superpixels, and then estimating the initial probability that the paired superpixels belong to the label pair; then, an iterative optimization strategy based on transmission is used for transmitting the initial probability to the whole situation to obtain the probability of the whole super-pixel pair; and finally, determining the super-pixel category according to the transmitted super-pixel pair probability to finish image segmentation.
With reference to fig. 1, the process of the embodiment of the present invention includes the following steps:
step 1, dividing an image into a plurality of super pixels, and constructing super pixel pairs, wherein the method specifically comprises the following steps:
the embodiment adopts an image selected from a Berkeley image data set, the image is a common color image, the image is selected before segmentation, the image is scribed, a red line mark is used for a foreground, and a green line mark is used for a background, so that the seed point category is obtained. Using the SLIC algorithm, a number of superpixels are generated, and then superpixel pairs are constructed.
Step 2, constructing a relation matrix of the super pixel pairs, which comprises the following specific steps:
and taking the average value of the characteristics of all pixel points in the super pixels as the characteristics of the super pixels, and calculating a relation matrix of the super pixel pairs by using a Gaussian function.
Step 3, calculating the initial probability of each super pixel pair belonging to the label pair, which is as follows:
the multivariate relation learning is carried out through a high-order tensor product diagram, the interaction between the super pixels of the image is captured, the initial probability that each super pixel pair belongs to a label pair is estimated through calculation, and a high-order graph Laplacian operator is added to smooth the prior label.
And 4, transferring the initial probability of the superpixel pair to the whole by using the relation matrix of the superpixel pair, wherein the method specifically comprises the following steps:
the prior probability of the superpixel pair is transmitted to the whole world, and the specific method comprises the following steps:
P(t)＝αQP(t-1)+(1-α)P
where Q is the row normalization matrix of the superpixel pair relationship matrix, P is the initial probability matrix of the superpixel pair, t is the number of iterations, P is the number of iterations(t)Is the result of the tth iteration of P.
And 5, determining the category of the superpixel according to the superpixel pair probability transmitted to the whole situation, and finishing image segmentation.
The experimental result is shown in fig. 2, column 1 is a test image, lines in the image are seed points of the foreground and the background, and columns 2 to 7 are the latest interactive segmentation algorithms: tensor product graph transfer algorithm (TPG), regularized tensor product graph transfer algorithm (RTPG), self-transfer algorithm (SD), probability transfer algorithm (PD), Laplace coordinate algorithm (LC) and sub Markov algorithm (SMRW), and the last 1 is the segmentation result obtained by the algorithm. As can be seen from fig. 2, under the condition of limited seed points, compared with several existing methods, the method of the present invention has a significantly better segmentation effect, expands the local neighborhood relationship to the global relationship on the image, and explores more complex interactions between element pairs and label pairs, which is helpful for better detecting the object contour under limited prior knowledge; second, the tag priors are propagated in the structure of a higher order tensor product map, which can capture the intrinsic data structure between pairs of image elements, which helps to better overcome the problems of low contrast and weak boundaries. The complexity of the image segmentation algorithm and the sensitivity of the segmentation result to the seed point are superior to those of the conventional method, so that the method realizes a better image segmentation effect and has important practical significance on aspects such as image matting, target tracking and the like in image processing.

Claims (6)

1. An interactive image segmentation method based on label group propagation is characterized by comprising the following steps:
step 1, dividing an image into a plurality of super pixels and constructing a super pixel pair;
step 2, constructing a relation matrix of the superpixel pairs;
step 3, calculating the initial probability of each super pixel pair belonging to the label pair;
step 4, transferring the initial probability of the super pixel pair to the whole situation by using the relation matrix of the super pixel pair;
and 5, determining the category of the superpixel according to the superpixel pair probability transmitted to the whole situation, and finishing image segmentation.
2. The method of claim 1, wherein the step 1 of dividing the image into a plurality of superpixels and constructing superpixel pairs comprises the following steps:
step 1.1, dividing an image into M super pixels by using an unsupervised segmentation algorithm;
step 1.2, for any super pixel, combining with the rest M-1 super pixels and the super pixels to construct M-1 super pixel pairs, wherein M super pixels construct M2'Sichuan' superA pair of pixels.
3. The method of claim 1, wherein the step 2 of constructing a relation matrix of superpixel pairs comprises the following steps:
step 2.1, taking the average value of the characteristics of all pixel points in the super pixel as the characteristics of the super pixel;
and 2.2, acquiring the neighborhood relation of the superpixels, and establishing a relation matrix of the superpixel pairs according to the characteristic values by using a Gaussian function.
4. The method of claim 1, wherein the step 3 of calculating the initial probability that each superpixel pair belongs to a tag pair comprises the following steps:
performing multivariate relation learning through a high-order tensor product graph, capturing the interaction between the super pixels of the image, estimating the initial probability of each super pixel pair belonging to a label pair, and adding a high-order graph Laplacian operator to smooth a prior label;
superpixel pairs s based on fusion relationshipsijBelong to the label pair lmnPrior probability of (2)
Figure FDA0002311596220000011
The definition is as follows:
Figure FDA0002311596220000012
where i, j is 1, … M, pair of superpixels sijFrom super-pixels siAnd sjForm, label pair lmnFrom a label lmAnd lnConstitution p(s)i∈lm|sj∈ln) For conditional probability, the following is defined:
Figure FDA0002311596220000021
wherein,
Figure FDA0002311596220000022
representing a super-pixel SiAnd SjSimilarity between them, the greater its value, the higher the similarity; l represents the number of tags;
Figure FDA0002311596220000023
for super-pixel Si belonging to label lmIs defined as follows:
wherein, ciIs a super pixel siIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure FDA0002311596220000025
for clustering algorithm to label lmAnd (4) clustering the seed point characteristics to obtain the kth clustering center, wherein K is the total number of the clustering centers.
5. The method of claim 1, wherein the initial probability of a superpixel pair is transmitted to the global using the relation matrix of the superpixel pair in step 4, specifically as follows:
P(t)＝αQP(t-1)+(1-α)P
q is a row normalization matrix of the super-pixel pair relation matrix obtained in the step 2, P is an initial probability matrix of the super-pixel pair, t is iteration times, and P is(t)and alpha is a regulation parameter as a result of the tth iteration of the P.
6. The interactive image segmentation method based on tag group propagation as claimed in claim 1, wherein the step 5 determines the class of superpixels according to the superpixel pair probability transmitted to the global, and completes image segmentation, specifically as follows:
when determining the super pixel category, according to the sum of the element pair probabilities of the super pixel and the super pixel itself and other M-1 super pixels, if the probability that the element pairs belong to the target is greater than the probability that the element pairs belong to the background, the super pixel belongs to the target, otherwise, the super pixel belongs to the background, and the classification is finished according to the probability.
CN201911261106.6A 2019-12-10 2019-12-10 Interactive image segmentation method based on label group propagation Pending CN111178361A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911261106.6A CN111178361A (en) 2019-12-10 2019-12-10 Interactive image segmentation method based on label group propagation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911261106.6A CN111178361A (en) 2019-12-10 2019-12-10 Interactive image segmentation method based on label group propagation

Publications (1)

Publication Number Publication Date
CN111178361A true CN111178361A (en) 2020-05-19

Family

ID=70651933

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911261106.6A Pending CN111178361A (en) 2019-12-10 2019-12-10 Interactive image segmentation method based on label group propagation

Country Status (1)

Country Link
CN (1) CN111178361A (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107633522A (en) * 2017-08-30 2018-01-26 山东财经大学 Brain image dividing method and system based on local similarity movable contour model

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107633522A (en) * 2017-08-30 2018-01-26 山东财经大学 Brain image dividing method and system based on local similarity movable contour model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
TAO WANG等: "Probabilistic Diffusion for Interactive Image Segmentation", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *

Similar Documents

Publication Publication Date Title
CN108388896B (en) License plate identification method based on dynamic time sequence convolution neural network
CN108549891B (en) Multi-scale diffusion well-marked target detection method based on background Yu target priori
Zeng et al. Learning to promote saliency detectors
CN108537239B (en) Method for detecting image saliency target
CN110796026A (en) Pedestrian re-identification method based on global feature stitching
CN111639564B (en) Video pedestrian re-identification method based on multi-attention heterogeneous network
CN111160407B (en) Deep learning target detection method and system
CN110728694B (en) Long-time visual target tracking method based on continuous learning
CN108846404B (en) Image significance detection method and device based on related constraint graph sorting
CN110188763B (en) Image significance detection method based on improved graph model
CN111507337A (en) License plate recognition method based on hybrid neural network
CN110942468B (en) Interactive image segmentation method based on superpixel pair probability transfer
CN116310098A (en) Multi-view three-dimensional reconstruction method based on attention mechanism and variable convolution depth network
Annum et al. Saliency detection using contrast enhancement and texture smoothing operations
CN116310095A (en) Multi-view three-dimensional reconstruction method based on deep learning
CN110634142B (en) Complex vehicle road image boundary optimization method
Katkar et al. A novel approach for medical image segmentation using PCA and K-means clustering
CN115471682A (en) Image matching method based on SIFT fusion ResNet50
Wang et al. End-to-end trainable network for superpixel and image segmentation
CN115035097A (en) Cross-scene strip steel surface defect detection method based on domain adaptation
Moradi et al. A salient object segmentation framework using diffusion-based affinity learning
CN116612385B (en) Remote sensing image multiclass information extraction method and system based on depth high-resolution relation graph convolution
CN113421210A (en) Surface point cloud reconstruction method based on binocular stereo vision
CN117152470A (en) Space-sky non-cooperative target pose estimation method and device based on depth feature point matching
CN111862159A (en) Improved target tracking and segmentation method, system and medium for twin convolutional network

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20200519

RJ01 Rejection of invention patent application after publication