CN110047089A - One kind being based on the matched pattern matching method of texture block - Google Patents
One kind being based on the matched pattern matching method of texture block Download PDFInfo
- Publication number
- CN110047089A CN110047089A CN201910265008.3A CN201910265008A CN110047089A CN 110047089 A CN110047089 A CN 110047089A CN 201910265008 A CN201910265008 A CN 201910265008A CN 110047089 A CN110047089 A CN 110047089A
- Authority
- CN
- China
- Prior art keywords
- pattern
- segmentation
- user
- matching
- similar
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/162—Segmentation; Edge detection involving graph-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20072—Graph-based image processing
Abstract
Based on the matched pattern matching method of texture block, user can allow for pass through the quality for the national culture pattern digital assays such as high-precision is realized in few interaction, the national culture pattern of high consistency is divided, while can be improved vector quantization.The present invention is the following steps are included: find all patterns similar with the pattern that user specifies using global Block- matching first;Then, using the rotation relationship between the similar pattern of localized mass matching detection, and the dense correspondence between them is further established by the Block- matching of belt restraining;Finally, realizing that the segmentation to all similar patterns cooperates with optimization using collaboration Optimized model proposed in this paper.The present invention have many advantages, such as interaction less, precision is high, structure-preserving, moreover it is possible to guarantee the segmentation consistency between similar pattern, have facilitation to the digital assay for improving the national culture patterns such as vector quantization.
Description
Technical field
The present invention relates to a kind of using computer technology based on the matched pattern matching method of texture block.
Background technique
National culture pattern is the symbolic language of national culture, is widely used in ethnic garment, carpet, furniture Deng Ge race
In the daily life of the people.It is the important of national culture, the concentrated reflection of national belief and China's non-material cultural heritage
Component part.Digitlization shooting, collecting is the effective means of persistence national culture pattern, and the digitlization based on image point
Analysis is to inherit and develop the only way which must be passed for residing in the national culture of national pattern again.Since national culture pattern is in the picture
Occur in area format, carrying out accurately segmentation to national culture pattern is the key that carry out subsequent digitation analysis.Such as
The application such as extraction, vector quantization, redesign of national culture pattern all relies on accurate pattern segmentation.
Although the usual color of national culture pattern is clearly demarcated, shape feature is obvious, accurate national pattern is realized
There is lot of challenges for segmentation: 1) due to the limitation of acquisition mode, the picture quality of national culture pattern is not high, causes existing
Automatic division method is difficult to retain main structure feature;2) since national culture pattern uses the diversity of carrier, existing side
Method obtains high-precision segmentation result dependent on a large amount of interaction, but the national culture pattern for needing to divide in digitalization resource manually
Quantity is extremely huge, so that the target of interactive segmentation is difficult to realize;3) national culture pattern is there is many similar patterns, these
Pattern is not striking resemblances, can show certain variation in different application scenarios and decoration, either automatically still
Manual segmentation, their segmentation result usually will appear inconsistent problem.Therefore, it needs a kind of based on the matched figure of texture block
Case matching process improves image segmentation.
Summary of the invention
The present invention will overcome the disadvantages mentioned above of the prior art, improve the segmentation of image, provide a kind of based on texture block
The pattern matching method matched.
The present invention utilizes the repeatability of national pattern, is established by multi-level texture block matching method quasi- between repeat patterns
True dense correspondence, to cooperate with the quality of optimization national culture pattern segmentation.
It is of the invention based on the matched pattern matching method of texture block, the specific steps are as follows:
Step 1 is split input picture I with the image segmentation algorithm based on L0 gradient minimisation, obtains pre-segmentation
Image Ir.
Step 2, single goal rapidly extracting, in input picture I rectangle frame selection target pattern, with improved GrabCut
Method is partitioned into sub-pattern S.
Step 3, multiple target automatically extract, using GPM (Generalized PatchMatch) method on input picture I
Find dense matching point corresponding with pattern.And after clustering to these match points, improved GrabCut method pair is reused
Cluster region carries out the extraction of similar pattern, and iteration is not until having to generate pattern.Finally obtain pattern set C.
Step 4, to all patterns in the pattern set C in step 3, using Shape Context method, with step 2
In pattern S calculate matching similarity, if similarity be less than specified threshold, then it represents that mismatch, delete it from pattern set.
Step 5 takes out a pattern from the pattern set C in step 3 first, establishes between the pattern and other patterns
GPM dense matching, improve the figure by analyzing topological structure corresponding relationship between the pattern on pre-segmentation image Ir
Case divides quality, then further takes out next pattern, repeats the step, until having traversed all patterns in set, thus
To final improvement result.Further, in the step 1, because the present invention is to carry out on the basis of image over-segmentation to image
Improve.So we adjust the parameter of L0 gradient minimisation method, make image segmentation result over-segmentation.Involved in the step
L0 gradient minimisation method is the article Feature- for being published in Computers&Graphics periodical the 38th phase of volume 38
The method that preserving filtering with L0 gradient minimization is proposed.
Further, in the step 2, the invention proposes a kind of improved GrabCut methods, and user is allowed to pass through picture frame
Method fast selecting optimization aim, which is a rectangle, interior zone Ruser, i.e. prospect, perimeter Rback=
Renlarge-Ruser, i.e. background.Wherein RenlargeIt is by rectangular area RuserIt is up and down respectively, left and right while extending w pixel
Enlarged area, w default takes 1st/20th of input picture catercorner length.Then Gaussian Mixture is established for foreground and background
Model (abbreviation GMM model), the probability of the fore/background to calculate each pixel in region to be split are final to utilize figure segmentation side
Method realizes segmentation.GrabCut method involved in the step is the article being published in ACM Siggraph meeting in 2004
Grabcut:Interactive Grabcut:Interactive foreground extraction using iterated
The method that graph cuts is proposed.
Further, in the step 3, take target using GPM method the single choice split in step 2, wherein needing
Using the multiple dimensioned and multi-angle function of search of GPM method, the nearest neighbor search quantity k of each pixel is set as 16, and will
The target area split is set as pure white to avoid repeated matching.Then, by spatial connectivity to these searching positions
Space clustering is carried out, i.e., is 5 roundlet in each searching position one radius of strokes and dots, and each company is obtained by se ed filling algorithm
The region of region area very little is rejected in logical region.The lesser judgement of region area is not using absolute threshold, which is
State variation, default takes single choice to take 1/5th of target area.Because of the pattern that target pattern and user's interactive segmentation obtain
More similar, therefore, we help the segmentation of pattern using following GMM model, to improve the stability and consistency of segmentation.
GMMi=α GMMuser+(1-α)GMMi
Wherein, α is hybrid subscriber interaction background GMMuserModel and current segmentation patterned background GMMiThe weight of model because
Son takes 0.5 herein.GPM method involved in the step is to be published in European Conference in 2010
The article The generalized patchmatch correspondence algorithm of Computer Vision meeting
The method of proposition.
Further, in the step 4 because dense matching also can there is a certain error, the present invention use Shape
The matched method of Context shape.This method is that the single goal pattern that the similar pattern that detection automatically extracts is extracted with user is
It is no similar, if dissimilar, delete, the shape matching method that the present invention uses has certain invariable rotary shape.In the step
The Shape Context shape matching method being related to is the Neural Information Processing for being published in 2000
The article Shape context:A new descriptor for shape matching and object of Systems meeting
The method that recognition is proposed.
Further, in the step 5, to all similar purpose patterns that step 4 generates, the present invention passes through local Block- matching
Establish the dense correspondence between these patterns.In order to solve pattern from AXIALLY SYMMETRIC PROBLEMS, herein be each target pattern estimation one
A direction first rotates to same direction and is carrying out dense matching.For two target pattern a and b with dense matching, pass through
GPM method establishes dense mapping relations ψ (a) → b;Wherein nearest neighbor search quantity k is set as 1.Then, with the center of pattern
For the center of circle, the mapping rotation angle ρ (a of each position is calculatedi), wherein aiIt is a location of pixels any in target pattern a;So
Afterwards, establish near border pixel in target pattern a to target pattern b mapping rotation angle histogram, and with numerical value highest order
Set integral-rotation angle θ (a, b) of the corresponding angle between target pattern a and b.
Assuming that shared in input picture N number of target pattern similar with user specifies pattern (specify including user that
It is a), it is denoted as { Ek, pattern E will be belonged tokInterior pre-segmentation region is denoted as the element of the pattern, is denoted asDue to the side of pattern
Boundary and the zone boundary of pre-segmentation are not fully consistent, and the present invention is only by 80% or more in EkInternal region is regarded as EkElement.
To determine arbitrary graphic pattern elementWhether need to be merged optimization, inquired in pattern element adjacent thereto whether
There are area ratiosGreatly, color withSimilar EkPattern element.If it does, might as well be denoted asIt willWithIt reflects respectively
It is mapped to other pattern EsIn (s ≠ k), check that their mapping pattern element whether there is intersection.If at any one its
There is intersections in its pattern element, just by pattern elementIt is merged intoIn.
The technical concept that the present invention is sent out is: making full use of the repeatability of national pattern, passes through multi-level texture block match party
Method establishes accurate dense correspondence between repeat patterns, to cooperate with the quality of optimization national pattern segmentation.Firstly, proposing one kind
In conjunction with the method for background segment before texture block global registration and Grabcut, user is allowed easily to choose all phases to be optimized
Like segmentation pattern;Relative rotational relationship is detected followed by local grain Block- matching, is established between multiple similar segmentation patterns
Dense matching;Finally, establishing the collaboration Optimized model and method for solving of pre-segmentation pattern, the Block- matching of part and belt restraining is utilized
Realize that the collaboration to segmentation effect optimizes.
The present invention has the advantages that establish a collaboration Optimized model for national pattern segmentation, by global and
Part repeatedly texture Block- matching establish the dense correspondence between similar pattern, thus cooperate with Optimized Segmentation accuracy with it is consistent
Property;As soon as proposing a segmentation framework for national culture pattern, allow user that can be quickly obtained high-precision by interaction on a small quantity, protect
The pattern segmentation effect of structure.
Detailed description of the invention
Fig. 1 is the general flow chart of the method for the present invention.
Fig. 2 a~Fig. 2 b is the schematic diagram of pattern segmentation result of the present invention, and wherein Fig. 2 a is input picture and initially interaction
Schematic diagram, Fig. 2 b are the patterns that the method for the present invention is divided.
Specific embodiment
Referring to attached drawing, the present invention is further illustrated:
Based on the matched pattern matching method of texture block, comprising the following steps:
Step 1 is split input picture I with the image segmentation algorithm based on L0 gradient minimisation, obtains pre-segmentation
Image Ir.
Step 2, single goal rapidly extracting, in input picture I rectangle frame selection target pattern, with improved GrabCut
Method is partitioned into pattern S.
Step 3, multiple target automatically extract, using GPM (Generalized PatchMatch) method on input picture I
Find dense matching point corresponding with pattern.And after clustering to these match points, improved GrabCut method pair is reused
Cluster region carries out the extraction of similar pattern, and iteration is not until having to generate pattern.Finally obtain pattern set C.
Step 4, to all patterns in the pattern set C in step 3, use
ShapeContext method calculates matching similarity with the pattern S in step 2, if similarity is less than specified threshold,
It then indicates to mismatch, deletes it from pattern set.
Step 5 takes out a pattern from the pattern set C in step 3 first, establishes between the pattern and other patterns
GPM dense matching, improve the figure by analyzing topological structure corresponding relationship between the pattern on pre-segmentation image Ir
Case divides quality, then further takes out next pattern, repeats the step, until having traversed all patterns in set, thus
To final improvement result.In the step 1, because the present invention is to improve on the basis of image over-segmentation to image.Institute
The parameter that L0 gradient minimisation method is adjusted with us makes image segmentation result over-segmentation.L0 gradient involved in the step
Minimum method is the article Feature-preserving for being published in Computers&Graphics periodical the 38th phase of volume 38
The method that filtering with L0 gradient minimization is proposed.
In the step 2, the invention proposes a kind of improved GrabCut methods, allow method of the user by picture frame
Fast selecting optimization aim, the frame are a rectangle, interior zone Ruser, i.e. prospect, perimeter Rback=Renlarge-
Ruser, i.e. background.Wherein RenlargeIt is by rectangular area RuserExpansion area that is up and down respectively, left and right while extending w pixel
Domain, w default take 1st/20th of input picture catercorner length.Then gauss hybrid models (letter is established for foreground and background
Claim GMM model), the probability of the fore/background to calculate each pixel in region to be split is finally realized using figure dividing method and is divided
It cuts.GrabCut method involved in the step is the article Grabcut being published in ACM Siggraph meeting in 2004:
Interac-tive Grabcut:Interactive foreground extraction using iterated graph
The method that cuts is proposed.
In the step 3, take target using GPM method the single choice split in step 2, wherein needing using GPM
The nearest neighbor search quantity k of each pixel is set as 16, and will split by the multiple dimensioned and multi-angle function of search of method
Target area be set as pure white to avoid repeated matching.Then, space is carried out to these searching positions by spatial connectivity
It clusters, i.e., the roundlet for being 5 in each searching position one radius of strokes and dots, and obtains the area of each connection by se ed filling algorithm
The region of region area very little is rejected in domain.The lesser judgement of region area is not using absolute threshold, which is dynamic change
, default takes single choice to take 1/5th of target area.Because of the pattern that target pattern is obtained with user's interactive segmentation more phase
Seemingly, therefore, we help the segmentation of pattern using following GMM model, to improve the stability and consistency of segmentation.
GMMi=α GMMuser+(1-α)GMMi
Wherein, α is hybrid subscriber interaction background GMMuserModel and current segmentation patterned background GMMiThe weight of model because
Son takes 0.5 herein.GPM involved in the step is to be published in European Conference on Computer in 2010
The side that the article The generalized patchmatch correspondence algorithm of Vision meeting is proposed
Method.
In the step 4 because dense matching also can there is a certain error, the present invention use Shape
The matched method of Context shape.This method is that the single goal pattern that the similar pattern that detection automatically extracts is extracted with user is
It is no similar, if dissimilar, delete, the shape matching method that the present invention uses has certain invariable rotary shape.In the step
The Shape Context shape matching method being related to is the Neural Information Processing for being published in 2000
The article Shape context:A new descriptor for shape matching and object of Systems
The method that recognition is proposed.
In the step 5, to all similar purpose patterns that step 4 generates, the present invention establishes this by local Block- matching
Dense correspondence between a little patterns.In order to solve pattern from AXIALLY SYMMETRIC PROBLEMS, herein be each target pattern estimate a direction,
It first rotates to same direction and is carrying out dense matching.For two target pattern a and b with dense matching, built by GPM method
Found dense mapping relations ψ (a) → b;Wherein nearest neighbor search quantity k is set as 1.Then, using the center of pattern as the center of circle, meter
Calculate the mapping rotation angle ρ (a of each positioni), wherein aiIt is a location of pixels any in target pattern a;Then, mesh is established
In case of marking on a map a near border pixel to target pattern b mapping rotation angle histogram, and with the corresponding angle in numerical value extreme higher position
Spend the integral-rotation angle θ (a, b) between target pattern a and b.
Assuming that shared in input picture N number of target pattern similar with user specifies pattern (specify including user that
It is a), it is denoted as { Ek, pattern E will be belonged tokInterior pre-segmentation region is denoted as the element of the pattern, is denoted asDue to the side of pattern
Boundary and the zone boundary of pre-segmentation are not fully consistent, and the present invention is only by 80% or more in EkInternal region is regarded as EkElement.
To determine arbitrary graphic pattern elementWhether need to be merged optimization, inquiring in pattern element adjacent thereto is
It is no that there are area ratiosGreatly, color withSimilar EkPattern element.If it does, might as well be denoted asIt willWithPoint
It is not mapped to other pattern EsIn (s ≠ k), check that their mapping pattern element whether there is intersection.If any one
There is intersections in a other pattern elements, just by pattern elementIt is merged intoIn.
The national pattern segmentation of high quality is to carry out number to non-material cultural heritages such as ethnic garment, nationality carpets
Change one of the premise of analysis.The invention proposes a kind of national culture pattern cooperative optimization method based on multi-level Block- matching,
Similar pattern can be divided into the minutiae patterns that are mutually related automatically, national culture pattern repeatability can be overcome strong, part
Have deformation etc. challenge, have many advantages, such as interaction less, precision height, structure-preserving, moreover it is possible to guarantee the segmentation consistency between similar pattern,
There is facilitation to the digital assay for improving the national culture patterns such as vector quantization.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in art technology
Personnel conceive according to the present invention it is conceivable that equivalent technologies mean.
Claims (1)
1. one kind is based on the matched pattern matching method of texture block, comprising the following steps:
Step 1 is split input picture I with the image segmentation algorithm based on L0 gradient minimisation, by adjusting L0 gradient
The parameter of minimum method makes image segmentation result over-segmentation, obtains pre-segmentation image Ir;
Step 2, single goal rapidly extracting, in input picture I rectangle frame selection target pattern, with improved GrabCut method
It is partitioned into pattern S, the improved GrabCut method specifically includes: allowing user excellent by the method fast selecting of picture frame
Change target, which is a rectangle, interior zone Ruser, i.e. prospect, perimeter Rback=Renlarge-Ruser, that is, carry on the back
Scape;Wherein RenlargeIt is by rectangular area RuserEnlarged area that is up and down respectively, left and right while extending w pixel, w default
Take 1st/20th of input picture catercorner length;Then gauss hybrid models GMM is established for foreground and background, to calculate
The probability of the fore/background of each pixel in region to be split finally realizes segmentation using figure dividing method;
Step 3, multiple target automatically extract, and are found on input picture I using GPM (Generalized PatchMatch) method
Dense matching point corresponding with pattern;It specifically includes:, will be each using the multiple dimensioned and multi-angle function of search of GPM method
The nearest neighbor search quantity k of pixel is set as 16, and the target area split is set as pure white to avoid repeated matching;
Then, space clustering is carried out to these searching positions by spatial connectivity, i.e., is 5 in each searching position one radius of strokes and dots
Roundlet, and obtain by se ed filling algorithm the region of each connection, reject the region of region area very little;Region area is smaller
Judgement be not using absolute threshold, which is dynamic change, and default takes single choice to take 1/5th of target area.Because
Target pattern and the obtained pattern of user's interactive segmentation are more similar, therefore, point of pattern are helped using following GMM model
It cuts, to improve the stability and consistency of segmentation.
GMMi=α GMMuser+(1-α)GMMi
Wherein, α is hybrid subscriber interaction background GMMuserModel and current segmentation patterned background GMMiThe weight of model, takes
0.5;
And after being clustered to these match points, reuses improved GrabCut method and similar pattern is carried out to cluster region
Extraction, iteration until do not have generate pattern until;Finally obtain pattern set C;
Step 4, to all patterns in the pattern set C in step 3, using Shape Context method, and in step 2
Pattern S calculates matching similarity, detects the similar pattern automatically extracted and whether the single goal pattern that user extracts be similar, if phase
It is less than specified threshold like degree, then it represents that mismatch, delete it from pattern set;
Step 5 takes out a pattern from the pattern set C in step 3 first, establishes between the pattern and other patterns
GPM dense matching improves the pattern by analyzing the topological structure corresponding relationship between the pattern on pre-segmentation image Ir
Divide quality, then further take out next pattern, the step is repeated, until having traversed all patterns in set, to obtain
Final improvement result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910265008.3A CN110047089A (en) | 2019-04-03 | 2019-04-03 | One kind being based on the matched pattern matching method of texture block |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910265008.3A CN110047089A (en) | 2019-04-03 | 2019-04-03 | One kind being based on the matched pattern matching method of texture block |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110047089A true CN110047089A (en) | 2019-07-23 |
Family
ID=67275965
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910265008.3A Withdrawn CN110047089A (en) | 2019-04-03 | 2019-04-03 | One kind being based on the matched pattern matching method of texture block |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110047089A (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104778242A (en) * | 2015-04-09 | 2015-07-15 | 复旦大学 | Hand-drawn sketch image retrieval method and system on basis of image dynamic partitioning |
CN107657625A (en) * | 2017-09-11 | 2018-02-02 | 南京信息工程大学 | Merge the unsupervised methods of video segmentation that space-time multiple features represent |
-
2019
- 2019-04-03 CN CN201910265008.3A patent/CN110047089A/en not_active Withdrawn
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104778242A (en) * | 2015-04-09 | 2015-07-15 | 复旦大学 | Hand-drawn sketch image retrieval method and system on basis of image dynamic partitioning |
CN107657625A (en) * | 2017-09-11 | 2018-02-02 | 南京信息工程大学 | Merge the unsupervised methods of video segmentation that space-time multiple features represent |
Non-Patent Citations (2)
Title |
---|
周倩: "《民族纹饰图案匹配算法与应用》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
陈佳舟等: "《中国科学: 信息科学》", 《基于多层次块匹配的民族图案分割协同优化方法》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Dong et al. | An efficient global energy optimization approach for robust 3D plane segmentation of point clouds | |
CN106709568B (en) | The object detection and semantic segmentation method of RGB-D image based on deep layer convolutional network | |
CN105931295B (en) | A kind of geologic map Extracting Thematic Information method | |
CN106127791B (en) | A kind of contour of building line drawing method of aviation remote sensing image | |
Li et al. | An edge embedded marker-based watershed algorithm for high spatial resolution remote sensing image segmentation | |
Li et al. | Building extraction from remotely sensed images by integrating saliency cue | |
CN108537239B (en) | Method for detecting image saliency target | |
CN107330875B (en) | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image | |
CN102024262A (en) | Method for performing image segmentation by using manifold spectral clustering | |
Miao et al. | Information fusion for urban road extraction from VHR optical satellite images | |
CN109712143A (en) | A kind of Fast image segmentation method based on super-pixel multiple features fusion | |
CN108537816A (en) | A kind of obvious object dividing method connecting priori with background based on super-pixel | |
CN110210418A (en) | A kind of SAR image Aircraft Targets detection method based on information exchange and transfer learning | |
CN110223310A (en) | A kind of line-structured light center line and cabinet edge detection method based on deep learning | |
CN108022245B (en) | Facial line primitive association model-based photovoltaic panel template automatic generation method | |
Paiva et al. | Historical building point cloud segmentation combining hierarchical watershed transform and curvature analysis | |
CN108038458A (en) | Outdoor Scene text automatic obtaining method in the video of feature based summary figure | |
Qian et al. | Spatial contextual noise removal for post classification smoothing of remotely sensed images | |
Wang et al. | A region-line primitive association framework for object-based remote sensing image analysis | |
CN109410233A (en) | A kind of accurate extracting method of high-definition picture road of edge feature constraint | |
CN107194405B (en) | Interactive semi-automatic high-resolution remote sensing image building extraction method | |
Manandhar et al. | Segmentation based building detection in high resolution satellite images | |
CN112861669A (en) | High-resolution DEM topographic feature enhancement extraction method based on earth surface slope constraint | |
CN109785318B (en) | Remote sensing image change detection method based on facial line primitive association constraint | |
CN110047089A (en) | One kind being based on the matched pattern matching method of texture block |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190723 |