CN109712143A - A kind of Fast image segmentation method based on super-pixel multiple features fusion - Google Patents
A kind of Fast image segmentation method based on super-pixel multiple features fusion Download PDFInfo
- Publication number
- CN109712143A CN109712143A CN201811619273.9A CN201811619273A CN109712143A CN 109712143 A CN109712143 A CN 109712143A CN 201811619273 A CN201811619273 A CN 201811619273A CN 109712143 A CN109712143 A CN 109712143A
- Authority
- CN
- China
- Prior art keywords
- super
- pixel
- hog
- hog feature
- similarity measurement
- 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.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of Fast image segmentation methods based on super-pixel multiple features fusion, by the way that image is carried out super-pixel pre-segmentation, calculate HOG feature similarity measurement, strength characteristic similarity measurement and the spatial position feature similarity measurement between each super-pixel and other super-pixel, then the multiple features similarity measurement between super-pixel is obtained, according to the multiple features similarity measurement, the super-pixel pre-segmentation image is converted to similarity relationships non-directed graph, and discussed according to figure hugger and similarity relationships non-directed graph is divided, it is partitioned into the cutting object.The present invention compares the image partition method based on Pixel-level, when being split to high-definition picture, has many advantages, such as that splitting speed is fast, robustness is good.
Description
Technical field
This application involves technical field of image processing more particularly to a kind of rapid images based on super-pixel multiple features fusion
Dividing method.
Background technique
Image segmentation is great challenge and highly important bottom visual problem in Computer Vision Task, is had extensive
Application prospect.With the continuous development of digital technology, the resolution ratio of image is greatly improved, and forms all kinds of high definition figures
Picture, but this to current pixel grade image segmentation algorithm either time cost and be that space cost all brings larger challenge,
Even it is also possible to cause some algorithms that can not work.It includes pixels to reach millions for a current common photo, general to scheme
As partitioning algorithm has been difficult to its directly progress operation, even if on the high-performance processor using the image segmentation being simple and efficient
Algorithm is split it, as long as also needing several minutes, and segmentation effect be not it is very ideal, segmentation result boundary and target are true
There are certain gaps for real edge.
There are 2 kinds of methods to can solve problems to a certain extent: 1) by down-sampling, the height of width M × N size
Image in different resolution, carry out s times of down-sampling, the low-resolution image of corresponding (M/s) × (N/s) size can be obtained, then into
Row subsequent processing.But such methods can lose the detailed information of great amount of images, and down-sampling degree is bigger, and it is more serious to lose information；
2) by generating super-pixel (superpixel), thousands of more than one hundred million a pixels be classified as hundreds of homogeneity, have it is good
The subregion of edge abutting degree, this strategy can be good at keeping the space structure and edge detail information of image object, together
When greatly simplify subsequent processing, naturally also just reduce corresponding computation burden.Therefore super-pixel processing becomes as in recent years
A kind of fast-developing Preprocessing Technique, super-pixel are not only more advantageous to the extraction of local feature and the table of structural information
Reach, capture various multilayer visual patterns in image, extracted feature have more robustness with it is representational, and can significantly drop
The computation complexity of low subsequent processing is widely used in computer vision field especially image segmentation.
Image segmentation has been widely studied decades, and based in super-pixel image partition method, different researchers are mentioned
Gone out different methods, as the method based on graph theory, the method based on region growing, the method based on threshold value and based on boundary
Method etc., wherein being wherein most representational method based on graph theory and based on region growing.
Method based on graph theory is pixel each in image to be regarded as the vertex of figure, and the side in figure is each pixel and its
4 fields or 8 field similarities.If normalized cut is by finding optimal segmentation in an iterative manner in Laplce's figure to obtain
The objective function of global minimization, although achieving preferable effect, computation complexity is higher.
Region growing method generates super-pixel since predefined seed point set, using different technologies.Wherein divide
Water ridge algorithm is based on one of most classic algorithm in region, and the algorithm is by creation gradient image, then in gradient plane simulation
Water flow is to form super-pixel.Mean shift algorithm is another form based on zone algorithm, is developed on its basis later
Quick Shift algorithm, all achieves good segmentation performance, has a preferable segmentation precision, however generation time of algorithm
Valence is higher.
In recent years, many researchs realize image segmentation with the method that other algorithms combine by super-pixel algorithm,
Achieve good segmentation effect.However different algorithms has different application backgrounds, there is presently no a kind of general methods
All image segmentation problems are able to solve, this research is for prospect and background near object edge to be split in high-definition picture
The incomplete pattern of the segmentation object that discrimination is small and be easy to cause proposes a kind of based on the quick of super-pixel multiple features fusion
Image segmentation algorithm.
The present invention proposes a kind of rapid image based on super-pixel multiple features fusion by introducing super-pixel HOG feature
Partitioning algorithm.Segmented image is treated using current most effective super-pixel algorithm first and carries out super-pixel processing, then extracts base
In the HOG feature of super-pixel, Lab color characteristic and spatial position feature, super-pixel multiple features fusion is based on by design
The Fast image segmentation algorithm of (superpixel multi-feature fusion, SMFF), realizes how special based on super-pixel
Levy the Fast image segmentation of fusion.
Summary of the invention
In view of this, the present disclosure proposes a kind of image partition method, it is time-consuming to solve high-definition picture Target Segmentation
Long, the classical low problem of segmenting edge.
In order to reach the object of the invention, the Fast image segmentation side provided by the invention based on super-pixel multiple features fusion
Method, it is characterised in that: the following steps are included:
Step 1 obtains image to be split, and the image to be split includes object to be split；
Image to be split is carried out super-pixel pre-segmentation formation super-pixel pre-segmentation image by step 2, and the super-pixel is divided in advance
Image is cut to be made of several super-pixel；
Step 3 extracts HOG feature, strength characteristic and space bit to each super-pixel in the super-pixel pre-segmentation image
Set feature, and calculate the HOG feature similarity measurement between each super-pixel and other super-pixel, strength characteristic similarity measurement,
Spatial position feature similarity measurement；
Step 4, to the HOG feature similarity measurement, strength characteristic similarity measurement and spatial position feature being calculated
Similarity measurement is merged, and the multiple features similarity measurement between each super-pixel and other super-pixel is obtained；
Step 5, according to the multiple features similarity measurement, the super-pixel pre-segmentation image is converted to similarity relationships
Non-directed graph, the super-pixel are mapped as the vertex in non-directed graph, and similarity measurement relationship map is each super-pixel to each other
The weight size of corresponding sides in non-directed graph；
Step 6, foundation figure hugger are discussed and are divided to the similarity relationships non-directed graph, and the cutting object is partitioned into.
Further, the present invention also further characterized in that
1, in step 3, using 9 directions of n × n grid computing 0 °, 20 °, 40 °, 60 °, 80 °, 100 °, 120 °, 140 °,
160 ° } on gradient, and gradient magnitude is normalized, obtains 9 × 1 HOG feature；The center of n × n grid is located at super picture
The boundary of element, then it is boundary HOG feature that HOG feature should be obtained by 8 × 8 grid computing, is denoted as HOGon；It otherwise is interior
Portion's HOG feature, is denoted as HOGin, and the HOG feature of super-pixel is corresponded to internal HOG character representation, more when existing in same super-pixel
When a internal HOG feature, then HOG feature of the mean value as the super-pixel is taken to multiple internal HOG feature；With boundary HOG spy
Sign indicates the HOG feature on boundary, when the boundary of same super-pixel is there are when multiple boundary HOG features, then to multiple boundary
HOG feature takes HOG feature of the mean value as the boundary；Calculate the HOG feature similarity measurement Sim_ of super-pixel i and super-pixel j
HOG (i, j):
In formula, Sip_HOGiniIndicate the HOG feature of super-pixel i, Sup_HOGinjIndicate the HOG feature of super-pixel j, E ()
Indicate expectation, μiIndicate the mean value of each element in the HOG feature of super-pixel i, μjIndicate each element in the HOG feature of super-pixel j
Mean value, ∑ onijIndicate that the average of all directions amplitude in the HOG feature on the shared boundary of super-pixel i and super-pixel j is poor.
2, in step 3, the strength characteristic similarity measurement between super-pixel i and super-pixel j, which is calculated by the following formula, to be obtained
:
It is wherein ηiAnd ηjThe internal color intensity average for respectively indicating super-pixel i and super-pixel j is poor, Li, Ai, BiPoint
Be not the channel L of super-pixel i, A channel, channel B average value, Lj, Aj, BjIt is the channel L of super-pixel j, A channel, B logical respectively
The average value in road.
3, in step 3, the spatial position feature similarity measurement between super-pixel i and super-pixel j passes through following formula meter
It calculates and obtains:
Wherein, xi, yiIndicate the abscissa and ordinate of seed point in super-pixel i generating process；xj, yjIndicate super-pixel j
The abscissa and ordinate of seed point in generating process.
4, in step 4, the multiple features similarity measurement Sim_ between super-pixel I and super-pixel j is calculated according to the following formula
Sup(i,j)
Sim_Sup (i, j)=α Sim_HOF (i, j)+β Sim_I (i, j)+γ Sim_D (i, j)
Wherein, α, β, γ are parameter preset.
Image by being carried out super-pixel pre-segmentation by the present invention, the multiple features similarity measurement between extraction super-pixel, according to
According to the multiple features similarity measurement, the super-pixel pre-segmentation image is converted to similarity relationships non-directed graph, and according to figure
Hugger is discussed and is divided to similarity relationships non-directed graph, and the cutting object is partitioned into.The present invention compares the figure based on Pixel-level
As dividing method, when being split to high-definition picture, have many advantages, such as that splitting speed is fast, robustness is good.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the Fast image segmentation method flow diagram the present invention is based on super-pixel multiple features fusion.
Fig. 2 is super-pixel HOG feature extraction schematic diagram.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Whole description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Super-pixel segmentation algorithm can be divided into the high-definition picture comprising thousands of more than one hundred million a pixels hundreds and thousands of
Homogeneity, subregion with good edge abutting degree, claim such subregion to be known as super-pixel, super-pixel is not only able to greatly
Subsequent processing is simplified, segmentation efficiency is improved, there is certain anti-noise ability, moreover it is possible to keep the space structure of image object to be split
And edge detail information.Super-pixel image segmentation algorithm is usually that simple local shape factor reality is carried out on image pixel-class
The segmentation of existing image, being formed has certain rule, the super-pixel of certain same characteristic features.Super-pixel be image over-segmentation as a result,
Often do not have complete visual meaningaaa, to realize the higher level image segmentation with complete visual effect, needs further
Higher level super-pixel feature is extracted, to realize subsequent combination and segmentation.
Image gradient direction histogram (Histogram of Oriented Gradient, HOG) is based on shape feature
The important innovations of extraction are the description of image information in a certain direction, the presentation and shape of image local target
(appearance and shape) can be described well by the direction Density Distribution at gradient or edge.It is in position and direction
Translation and rotational invariance are realized to a certain degree in the quantization in space；It takes simultaneously in regional area normalization histogram, it can
To overcome illumination variation bring to influence.The embodiment of the present application is designed by the way that HOG feature is introduced into super-pixel image segmentation
The extraction algorithm of HOG feature based on super-pixel extracts feature of the HOG feature of each super-pixel as measuring similarity to each other
One of, this is one of key innovations of the present embodiment.Based on the introducing of super-pixel HOG feature, solve in complex pattern to
Segmentation object adjacent edges prospect and background discrimination it is small or as noise effect discrimination it is lower and caused by segmentation it is imperfect
Problem.
For this purpose, the embodiment of the present application provides a kind of SMFF Fast image segmentation method, algorithm improvement traditional images point
Algorithm is cut to the segmentation efficiency of high-definition picture, while also having the characteristics that object edge abutting is kept to spend.Mesh is used first
The preceding highest super-pixel algorithm of segmentation efficiency treats segmented image and carries out super-pixel processing, and image to be split is divided into certain
The super-pixel of kind same characteristic features extracts HOG feature, Lab color characteristic and sky based on super-pixel using super-pixel as basic unit
Between position feature, by design the multiple features metric algorithm based on super-pixel, using be based on super-pixel multiple features fusion
The method of (superpixel multi-feature fusion, SMFF), realizes Fast image segmentation.This research achieves
Good segmentation effect, arithmetic accuracy is close to most classic image segmentation algorithm, and the time performance of this research will be substantially better than
Experimental comparison's algorithm.
As shown in Figure 1, be Fast image segmentation method flow diagram of the embodiment of the present invention based on super-pixel multiple features fusion,
The following steps are included:
S101, image to be split is obtained, the image to be split includes object to be split.
Original image untreated for one, if in the original image include user wish to be partitioned into it is to be split
The original image can be integrally used as the image to be split by object, can also be by the partial region (ratio in the original image
Such as can use drawing tools and manually intercept the region) it is used as the image to be split, as long as including in the image to be split
The object to be split.It should also be noted that, the image to be split can be high-definition picture, it is also possible to it
The image of low resolution, the present embodiment are not construed as limiting this.For example, when the image high-resolution image to be split, this Shen
Please the effect of embodiment can become apparent from.
S102, image to be split is carried out to super-pixel pre-segmentation formation super-pixel pre-segmentation image, the super-pixel is divided in advance
Image is cut to be made of several super-pixel.This step super-pixel pre-segmentation technology belongs to prior art scope, and specific dividing method can
Referring to paper " Zhao J X, Ren B, Hou Q B, et al.FLIC:Fast Linear Iterative Clustering
With Active Search [C] //AAAI.2018. ", the embodiment of the present invention are not set forth in detail.Main innovation of the invention
It is how to carry out the extraction of HOG feature for hyperfractionated image, high-definition picture is realized while to guarantee segmentation precision
Fast Segmentation.
S103, HOG feature, strength characteristic and space bit are extracted to each super-pixel in the super-pixel pre-segmentation image
Set feature, and calculate the HOG feature similarity measurement between each super-pixel and other super-pixel, strength characteristic similarity measurement,
Spatial position feature similarity measurement.
HOG feature similarity measurement Sim_HOG (i, j) calculation method of super-pixel i and super-pixel j is as follows:
Using the ladder on 8 × 8 grid computing, 9 directions { 0 °, 20 °, 40 °, 60 °, 80 °, 100 °, 120 °, 140 °, 160 ° }
Degree, and gradient magnitude is normalized, obtain 9 × 1 HOG feature；The center of 8 × 8 grids is located at the boundary of super-pixel,
It is boundary HOG feature that HOG feature should be then obtained by 8 × 8 grid computing, is denoted as HOGon；It otherwise is internal HOG feature, note
For HOGin, the HOG feature of super-pixel is corresponded to internal HOG character representation, when there are multiple internal HOG are special in same super-pixel
When sign, then HOG feature of the mean value as the super-pixel is taken to multiple internal HOG feature；With boundary HOG character representation boundary
HOG feature then takes mean value to multiple boundary HOG feature when the boundary of same super-pixel is there are when multiple boundary HOG features
HOG feature as the boundary；Calculate the HOG feature similarity measurement Sim_HOG (i, j) of super-pixel i and super-pixel j:
In formula, Sup_HOGiniIndicate the HOG feature of super-pixel i, Sup_HOGinjIndicate the HOG feature of super-pixel j, E ()
Indicate expectation, μiIndicate the mean value of each element in the HOG feature of super-pixel i, μjIndicate each element in the HOG feature of super-pixel j
Mean value, ∑ onijIndicate that the average of all directions amplitude in the HOG feature on the shared boundary of super-pixel i and super-pixel j is poor.
When calculating HOG feature, 5 × 5 grids, 6 × 6 grids, 7 × 7 grids or 9 × 9 grid computings can also be used.
In order to more easily carry out the HOG feature similarity measurement based on super-pixel, it is illustrated by taking Fig. 2 as an example.In Fig. 2
In, it (b) is (a) small rectangle frame range partial enlargement effect, HOGon that (a), which is super-pixel and HOG feature Overlay Local map,
The boundary characteristic and provincial characteristics of super-pixel HOG, Sup_HOGini, Sup_HOGinj and Sup_HOGink are indicated with HOGin, point
Not Biao Shi where super-pixel region HOG feature mean value, be (c) Sup_HOGini, Sup_HOGinj and Sup_HOGink in (b)
The histogram display effect of feature, segmentation result vision is imitated before and after HOG feature (d) is added for range shown in rectangle frame big in (a)
Fruit, (1) are the shown range part original image for scheming big rectangle frame in (a), and (2) (3) are respectively the segmentation effect before and after HOG feature is added
Fruit.
Super-pixel algorithm is the super-pixel of homogeneity image pre-segmentation, and each super-pixel boundary can be divided into two classes: the first kind is
The interim partitioning boundary formed to divide super-pixel need to merge disappearance in the subsequent image segmentation based on super-pixel, this
Class super-pixel is normally at image smoothing area；Second class is the true super-pixel segmentation boundary formed to portray object boundary,
The boundary that need to retain in subsequent singulation, forms final partitioning boundary, and this kind of super-pixel is normally at the variable region of image.This
Invention utilizes the good gradient direction Expressive Features of HOG, and HOG feature is effectively combined with super-pixel piecemeal, and design is special based on HOG
The super-pixel similarity measurements quantity algorithm of sign, improves the accuracy of each super-pixel similarity measurement.
Strength characteristic similarity measurement between super-pixel i and super-pixel j is calculated by the following formula acquisition:
It is wherein ηiAnd ηjThe internal color intensity average for respectively indicating super-pixel i and super-pixel j is poor, Li, Ai, BiPoint
Be not the channel L of super-pixel i, A channel, channel B average value, Lj, Aj, BjIt is the channel L of super-pixel j, A channel, B logical respectively
The average value in road.
Spatial position feature similarity measurement between super-pixel i and super-pixel j is calculated by the following formula acquisition:
Wherein, xi, yiIndicate the abscissa and ordinate of seed point in super-pixel i generating process；xj, yjIndicate super-pixel j
The abscissa and ordinate of seed point in generating process.
S104, the HOG feature similarity measurement to being calculated, strength characteristic similarity measurement and spatial position feature phase
It is merged like property measurement, obtains the multiple features similarity measurement between each super-pixel and other super-pixel.
The multiple features similarity measurement Sim_Sup (i, j) between super-pixel i and super-pixel j is calculated according to the following formula
Sim_Sup (i, j)=α Sim_HOG (i, j)+β Sim_I (i, j)+γ Sim_D (i, j)
Wherein, α, β, γ are parameter preset.Value range [0.391~0.472], the value range of β of the parameter alpha be
[0.341~0.372], the value range of γ are [0.187~0.353], in this example the value of α, β, γ be respectively 0.471,
0.342、0.813。
S105, according to the multiple features similarity measurement, the super-pixel pre-segmentation image is converted to similarity relationships
Non-directed graph, wherein super-pixel is mapped as the vertex in non-directed graph, and similarity measurement relationship map is each super-pixel to each other
The weight size of corresponding sides in non-directed graph.
S106, foundation figure hugger are discussed and are divided to the similarity relationships non-directed graph, and the cutting object is partitioned into.This
Step belongs to prior art scope, and specific method can be found in paper " Felzenszwalb P F, Huttenlocher D
P.Efficient Graph-Based Image Segmentation[J].International Journal of
Computer Vision,2004,59(2):167-181.”。
In addition to the implementation, the present invention can also have other embodiments.It is all to use equivalent substitution or equivalent transformation shape
At technical solution, fall within the scope of protection required by the present invention.
Claims (7)
1. a kind of Fast image segmentation method based on super-pixel multiple features fusion, it is characterised in that: the following steps are included:
Step 1 obtains image to be split, and the image to be split includes object to be split；
Image to be split is carried out super-pixel pre-segmentation formation super-pixel pre-segmentation image, the super-pixel pre-segmentation figure by step 2
As being made of several super-pixel；
Step 3 extracts HOG feature, strength characteristic and spatial position spy to each super-pixel in the super-pixel pre-segmentation image
Sign, and calculate the HOG feature similarity measurement between each super-pixel and other super-pixel, strength characteristic similarity measurement, space
Position feature similarity measurement；
It is step 4, similar with spatial position feature to the HOG feature similarity measurement, strength characteristic similarity measurement that are calculated
Property measurement merged, obtain the multiple features similarity measurement between each super-pixel and other super-pixel；
Step 5, according to the multiple features similarity measurement, it is undirected that the super-pixel pre-segmentation image is converted to similarity relationships
Figure, the super-pixel are mapped as the vertex in non-directed graph, and similarity measurement relationship map is undirected to each super-pixel to each other
The weight size of corresponding sides in figure；
Step 6, foundation figure hugger are discussed and are divided to the similarity relationships non-directed graph, and the cutting object is partitioned into.
2. the Fast image segmentation method according to claim 1 based on super-pixel multiple features fusion, it is characterised in that: step
In rapid 3, using the ladder on 9 directions of n × n grid computing { 0 °, 20 °, 40 °, 60 °, 80 °, 100 °, 120 °, 140 °, 160 ° }
Degree, and gradient magnitude is normalized, obtain 9 × 1 HOG feature；The center of n × n grid is located at the boundary of super-pixel,
It is boundary HOG feature that HOG feature should be then obtained by 8 × 8 grid computing, is denoted as HOGon；It otherwise is internal HOG feature, note
For HOGin, the HOG feature of super-pixel is corresponded to internal HOG character representation, when there are multiple internal HOG are special in same super-pixel
When sign, then HOG feature of the mean value as the super-pixel is taken to multiple internal HOG feature；With boundary HOG character representation boundary
HOG feature then takes mean value to multiple boundary HOG feature when the boundary of same super-pixel is there are when multiple boundary HOG features
HOG feature as the boundary；Calculate the HOG feature similarity measurement Sim_HOG (i, j) of super-pixel i and super-pixel j:
In formula, Sup-HOGiniIndicate the HOG feature of super-pixel i, Sup_HOGinjIndicate the HOG feature of super-pixel j, E () is indicated
It is expected that μiIndicate the mean value of each element in the HOG feature of super-pixel i, μiEach element is equal in the HOG feature of expression super-pixel j
Value, ∑ onijIndicate that the average of all directions amplitude in the HOG feature on the shared boundary of super-pixel i and super-pixel j is poor.
3. the Fast image segmentation method according to claim 2 based on super-pixel multiple features fusion, it is characterised in that: step
In rapid 3, the strength characteristic similarity measurement between super-pixel i and super-pixel j is calculated by the following formula acquisition:
It is wherein ηiAnd ηjThe internal color intensity average for respectively indicating super-pixel i and super-pixel j is poor, Li, Ai, BiIt is respectively
The channel L of super-pixel i, A channel, channel B average value, Lj, Aj, BjIt is the channel L, the A channel, channel B of super-pixel j respectively
Average value.
4. the Fast image segmentation method according to claim 3 based on super-pixel multiple features fusion, it is characterised in that: step
In rapid 3, the spatial position feature similarity measurement between super-pixel i and super-pixel j is calculated by the following formula acquisition:
Wherein, xi, yiIndicate the abscissa and ordinate of seed point in super-pixel i generating process；xj, yjIndicate that super-pixel j is generated
The abscissa and ordinate of seed point in the process.
5. the Fast image segmentation method according to claim 4 based on super-pixel multiple features fusion, it is characterised in that: step
In rapid 4, the multiple features similarity measurement Sim_Sup (i, j) between super-pixel i and super-pixel j is calculated according to the following formula
Sim_Sup (i, j)=α Sim_HOG (i, j)+β Sim_I (i, j)+ySim_D (i, j)
Wherein, α, β, γ are parameter preset.
6. the Fast image segmentation method according to claim 5 based on super-pixel multiple features fusion, it is characterised in that: institute
Value range [0.391~0.472], the value range of β for stating parameter alpha are [0.341~0.372], and the value range of γ is
[0.187~0.353].
7. the Fast image segmentation method according to claim 5 based on super-pixel multiple features fusion, it is characterised in that: n
=5 or 6 or 7 or 8 or 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811619273.9A CN109712143B (en) | 2018-12-27 | 2018-12-27 | Rapid image segmentation method based on superpixel multi-feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811619273.9A CN109712143B (en) | 2018-12-27 | 2018-12-27 | Rapid image segmentation method based on superpixel multi-feature fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109712143A true CN109712143A (en) | 2019-05-03 |
| CN109712143B CN109712143B (en) | 2021-01-26 |
Family
ID=66257900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811619273.9A Active CN109712143B (en) | 2018-12-27 | 2018-12-27 | Rapid image segmentation method based on superpixel multi-feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109712143B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353546A (en) * | 2020-03-09 | 2020-06-30 | 腾讯科技（深圳）有限公司 | Training method and device of image processing model, computer equipment and storage medium |
| CN111833362A (en) * | 2020-06-17 | 2020-10-27 | 北京科技大学 | Unstructured road segmentation method and system based on superpixel and region growing |
| CN111931811A (en) * | 2020-06-29 | 2020-11-13 | 南京巨鲨显示科技有限公司 | Calculation method based on super-pixel image similarity |
| CN112330697A (en) * | 2020-12-04 | 2021-02-05 | 中信银行股份有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
| CN113696454A (en) * | 2021-10-28 | 2021-11-26 | 南通三信塑胶装备科技股份有限公司 | Artificial intelligence-based extrusion molding equipment fault early warning method and system |
| CN115776410A (en) * | 2023-01-29 | 2023-03-10 | 深圳汉德霍尔科技有限公司 | Face data encryption transmission method for terminal identity authentication |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103413307A (en) * | 2013-08-02 | 2013-11-27 | 北京理工大学 | Method for image co-segmentation based on hypergraph |
| CN104537623A (en) * | 2014-12-31 | 2015-04-22 | 深圳先进技术研究院 | Image fog-removing method and device based on image segmentation |
| CN104899875A (en) * | 2015-05-20 | 2015-09-09 | 天津大学 | Rapid image cooperation salient region monitoring method based on integration matching |
| CN105427296A (en) * | 2015-11-11 | 2016-03-23 | 北京航空航天大学 | Ultrasonic image low-rank analysis based thyroid lesion image identification method |
| CN105761266A (en) * | 2016-02-26 | 2016-07-13 | 民政部国家减灾中心 | Method of extracting rectangular building from remote sensing image |
| CN106530303A (en) * | 2016-11-01 | 2017-03-22 | 陕西科技大学 | Spectral clustering-based color image fast segmentation method |
| CN107025672A (en) * | 2017-03-30 | 2017-08-08 | 上海理工大学 | A kind of conspicuousness detection method based on improvement convex closure |
-
2018
- 2018-12-27 CN CN201811619273.9A patent/CN109712143B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103413307A (en) * | 2013-08-02 | 2013-11-27 | 北京理工大学 | Method for image co-segmentation based on hypergraph |
| CN104537623A (en) * | 2014-12-31 | 2015-04-22 | 深圳先进技术研究院 | Image fog-removing method and device based on image segmentation |
| CN104899875A (en) * | 2015-05-20 | 2015-09-09 | 天津大学 | Rapid image cooperation salient region monitoring method based on integration matching |
| CN105427296A (en) * | 2015-11-11 | 2016-03-23 | 北京航空航天大学 | Ultrasonic image low-rank analysis based thyroid lesion image identification method |
| CN105761266A (en) * | 2016-02-26 | 2016-07-13 | 民政部国家减灾中心 | Method of extracting rectangular building from remote sensing image |
| CN106530303A (en) * | 2016-11-01 | 2017-03-22 | 陕西科技大学 | Spectral clustering-based color image fast segmentation method |
| CN107025672A (en) * | 2017-03-30 | 2017-08-08 | 上海理工大学 | A kind of conspicuousness detection method based on improvement convex closure |
Non-Patent Citations (3)
| Title |
|---|
| DANIEL P. HUTTENLOCHER 等: "Efficient Graph-Based Image Segmentation", 《INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
| EDWARD KIM 等: "A Hierarchical Image Clustering Cosegmentation Framework", 《IEEE》 * |
| 李学龙 等: "场景图像分类技术综述", 《中国科学：信息科学》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353546A (en) * | 2020-03-09 | 2020-06-30 | 腾讯科技（深圳）有限公司 | Training method and device of image processing model, computer equipment and storage medium |
| CN111353546B (en) * | 2020-03-09 | 2022-12-23 | 腾讯科技（深圳）有限公司 | Training method and device of image processing model, computer equipment and storage medium |
| CN111833362A (en) * | 2020-06-17 | 2020-10-27 | 北京科技大学 | Unstructured road segmentation method and system based on superpixel and region growing |
| CN111931811A (en) * | 2020-06-29 | 2020-11-13 | 南京巨鲨显示科技有限公司 | Calculation method based on super-pixel image similarity |
| WO2022001571A1 (en) * | 2020-06-29 | 2022-01-06 | 南京巨鲨显示科技有限公司 | Computing method based on super-pixel image similarity |
| CN111931811B (en) * | 2020-06-29 | 2024-03-29 | 南京巨鲨显示科技有限公司 | Calculation method based on super-pixel image similarity |
| CN112330697A (en) * | 2020-12-04 | 2021-02-05 | 中信银行股份有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
| CN112330697B (en) * | 2020-12-04 | 2024-04-30 | 中信银行股份有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
| CN113696454A (en) * | 2021-10-28 | 2021-11-26 | 南通三信塑胶装备科技股份有限公司 | Artificial intelligence-based extrusion molding equipment fault early warning method and system |
| CN115776410A (en) * | 2023-01-29 | 2023-03-10 | 深圳汉德霍尔科技有限公司 | Face data encryption transmission method for terminal identity authentication |
| CN115776410B (en) * | 2023-01-29 | 2023-05-02 | 深圳汉德霍尔科技有限公司 | Face data encryption transmission method for terminal identity authentication |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109712143B (en) | 2021-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109712143A (en) | A kind of Fast image segmentation method based on super-pixel multiple features fusion | |
| Zaitoun et al. | Survey on image segmentation techniques | |
| Li et al. | An edge embedded marker-based watershed algorithm for high spatial resolution remote sensing image segmentation | |
| Lu et al. | Salient object detection using concavity context | |
| Vanhamel et al. | Multiscale gradient watersheds of color images | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN101706950A (en) | High-performance implementation method for multi-scale segmentation of remote sensing images | |
| CN102024262A (en) | Method for performing image segmentation by using manifold spectral clustering | |
| Yuan et al. | Volume cutout | |
| 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 | |
| Najman et al. | A short tour of mathematical morphology on edge and vertex weighted graphs | |
| Serna et al. | Attribute controlled reconstruction and adaptive mathematical morphology | |
| Li et al. | Edge-based split-and-merge superpixel segmentation | |
| CN112884884A (en) | Candidate region generation method and system | |
| Bhadauria et al. | Building extraction from satellite images | |
| Geng et al. | A novel color image segmentation algorithm based on JSEG and Normalized Cuts | |
| Liu et al. | Graph-based image segmentation using directional nearest neighbor graph | |
| CN115409954A (en) | Dense point cloud map construction method based on ORB feature points | |
| Li et al. | Automated methods for indoor point cloud preprocessing: Coordinate frame reorientation and building exterior removal | |
| CN111242960A (en) | Image segmentation method based on complex network theory | |
| Garro et al. | Label transfer exploiting three-dimensional structure for semantic segmentation | |
| Aravinda et al. | Simple linear iterative clustering based tumor segmentation in liver region of abdominal CT-scan | |
| Kuo et al. | Watershed segmentation with automatic altitude selection and region merging based on the Markov random field model | |
| Li et al. | Stereo Image Matching Algorithm Based on Texture Segmentation and Color Segmentation |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Hou Xiaogang Inventor after: Zhao Haiying Inventor after: Chen Hong Inventor before: Chen Hong Inventor before: Zhao Haiying Inventor before: Hou Xiaogang |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |