CN108230340A - A kind of SLIC super-pixel extraction Weighting and super-pixel extracting method based on MMTD - Google Patents
A kind of SLIC super-pixel extraction Weighting and super-pixel extracting method based on MMTD Download PDFInfo
- Publication number
- CN108230340A CN108230340A CN201810109851.8A CN201810109851A CN108230340A CN 108230340 A CN108230340 A CN 108230340A CN 201810109851 A CN201810109851 A CN 201810109851A CN 108230340 A CN108230340 A CN 108230340A
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- value
- weights
- point
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—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
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of SLIC super-pixel extraction Weightings and super-pixel extracting method based on MMTD, and the method increase image superpixels to extract accuracy rate, belongs to the research field of the image segmentation in image procossing.Current SLIC super-pixel extracting methods are to determine the proportion between pixel between Lab distances and coordinate distance with fixed weights.Wherein weights m needs artificially to specify, and for different images, takes same m, the effect of segmentation may be undesirable, influences the extraction of super-pixel block.For this problem, the method of the present invention is according to the similarities of two kinds of distances of image Lab distances and coordinate distance, weights are adaptively determined using the method for iteration, weights of super-pixel range formula are determined with this, reduce the inaccurate situation of super-pixel extraction, the final effect for improving image superpixel extraction.
Description
Technical field
The present invention relates to the technical field of image segmentation, more particularly to a kind of SLIC for being based on MMTD (intermediary's truth scale)
(simple linear iteraction cluster) super-pixel extracting method.
Background technology
Image segmentation is exactly to divide the image into several regions specific, with unique properties and propose interesting target
Technology and process.It is one of computer vision and the particularly important research contents of image processing field, from the seventies in last century
Act the great attention for being constantly subjected to people.The accuracy rate and efficiency of image segmentation directly affect subsequent image classification and identifying processing
Validity.Image partition method is usually classified according to boundary, homogeney, shape knowledge, and thus various dividing methods are general
All attempt the boundary in detection image and homogeney region, and add in shape information to constrain cutting procedure and generate correctly knot
Fruit.According to these three key properties, image partition method be divided into weights dividing method, edge detection method, method for extracting region,
Dividing method based on locally or globally prior shape image partition method and combination Specific Theory Tools.
Super-pixel provides a kind of easily mode to calculate characteristics of image.They are by obtaining the redundancy in image come big
Mitigate to amplitude the complexity of subsequent image processing.Concept is exactly to segment the image into many fritters, then can this is whole
A block is handled as a pixel, and each fritter is exactly super-pixel.Calculating is more had based on the characteristics of image of super-pixel than pixel
Effect.Therefore, in the image processing tasks based on super-pixel, image primitive and redundancy can greatly reduce.In general, it uses
Image superpixel is divided to improve their efficiency and performance.Image surpasses segmentation method and is often used as many computer vision work
Pre-treatment step, pixel partitioning algorithm has studied many years.It is existing in most of papers of super-pixel generation method
The basic thought of method be divided into two classes:Method based on graph theory and the method based on k- mean values.In order to which super-pixel is made to become to have
With they must be quick, easy to use, and generates the piecemeal of high quality.Unfortunately, most of state-of-the-art super-pixel methods
All these requirements cannot all be met, they frequently suffer from high calculating cost, ropy segmentation, inconsistent size and shape
Shape includes multiple parameters for being difficult to tuning.
SLIC generation super-pixel is by the pixel cluster based on color similarity and proximity.The method that SLIC is provided,
Although very simple, it solves the problems, such as these, and produces high quality, compact, almost consistent super pixel.But
Traditional SLIC algorithms are to take coordinate distance knot between pixel Lab distances and pixel when pel spacing is calculated from formula
The method of conjunction wherein the weights m that the two combines needs artificially to specify, for different images, takes same m, the effect of segmentation
Fruit may be undesirable, influences the extraction of super-pixel block.
In conclusion original SLIC algorithms, when calculating pel spacing from formula, very important person is to specify a weights m,
When handling different images, identical weights are taken, in fact it could happen that the deviation of extraction needs to be improved the selection of m, to carry
The extraction effect of high super-pixel.And the present invention can well solve the above problem.
Invention content
Goal of the invention:In order to overcome the deficiencies in the prior art, the present invention provides a kind of SLIC based on MMTD and surpasses
Pixel extraction Weighting and extracting method are solved handling different images, be needed in manual selected distance formula
The problem of weights.So for different images, the weights how chosen in suitable range formula surpass picture so as to improve image
Element extraction quality is the Important Problems of super-pixel extraction.
Technical solution:To achieve the above object, the technical solution adopted by the present invention is:
A kind of SLIC super-pixel extraction Weighting based on MMTD, current SLIC super-pixel extracting methods are,
The proportion between pixel between Lab distances and coordinate distance is determined with fixed weights.Wherein weights m needs artificially to specify, right
In different images, same m is taken, the effect of segmentation may be undesirable, influences the extraction of super-pixel block.It is asked for such
Topic, method of the invention are adaptive using the method for iteration according to the similarities of two kinds of distances of image Lab distances and coordinate distance
It determines weights with answering, weights of super-pixel range formula is determined with this, reduce the inaccurate situation of super-pixel extraction, most
The effect of image superpixel extraction is improved eventually, specifically includes following steps:
Step 1, brightness value similarity L (x, y) between pixel and super-pixel block central pixel point, red value of green phase are determined
Like degree A (x, y), champac value similarity B (x, y) and coordinate Euclidean distance D (x, y).Then according to brightness value similarity L (x,
Y), red value of green similarity A (x, y), champac value similarity B (x, y) and coordinate Euclidean distance D (x, y) determine distance proportion
Function h (x, y).Super-pixel block central pixel point center=x (i, j) and its place are calculated by distance proportion function h (x, y)
The phase of the similarity, the then each pixel of the super-pixel block and two kinds of distances of central point of each two kinds of distances of pixel of super-pixel block
It is respectively d1 like degree, d2, d3...dn composition D, wherein n represent super-pixel block except central pixel point shares pixel number.
Step 2, weights m=(max (D)+min (D))/2 is determined.
Step 3, super-pixel pixel in the block is divided by two parts according to weights m, two kinds of pixel and central point away from
M1 is denoted as from Distance conformability degree set of the similarity more than weights m, the Distance conformability degree set less than weights m is denoted as m2.
Step 4, m1, the mean value P1, P2 of m2 are calculated respectively.
Step 5, new weights mm=(P1+P2)/2 is calculated.
Step 6, replace original weights m with new weights mm, step 3), step 4), step 5) and step 6) are repeated, until two
Part weights difference d=m-mm, less than permissible value allow, then iteration terminates, and the new weights mm that iteration is terminated is as the block of pixels
Weights.
Preferably:Pixel and the brightness value similarity L (x, y) of super-pixel block central pixel point are determined according to formula (2).
Wherein, L (x, y) represents the brightness value similarity of pixel and super-pixel block central pixel point, and (x, y) represents pixel
Point and super-pixel block central pixel point, Lx is the brightness value of x, and Ly is the brightness value of y.
Preferably:The red value of green similarity A (x, y) of pixel and super-pixel block central pixel point is determined according to formula (3).
Wherein, A (x, y) represents the red value of green similarity of pixel and super-pixel block central pixel point, and Ax is the red green of x
Value, Ay is the red value of green of y.
Preferably:Pixel and the champac value similarity B (x, y) of super-pixel block central pixel point are determined according to formula (4).
Wherein, B (x, y) represents the champac value similarity of pixel and super-pixel block central pixel point, and Bx is the blue yellow values of x,
Bx, By are the yellow values of indigo plant of y.
Preferably:Distance proportion function h (x, y) is determined according to formula (5).
Wherein, D (x, y) represents pixel and the European coordinate distance of super-pixel block central pixel point.
A kind of SLIC super-pixel extracting methods based on MMTD, include the following steps:
Step 1 generates super-pixel block using SLIC algorithms, and the distance metric formula that wherein SLIC is used is as follows:
Wherein, d_Lab is the Lab distances of pixel x and pixel y, and Lx, Ly are the brightness of pixel x and y respectively, and Ax, Ay distinguish
The red value of green of x and y, Bx, By are the yellow value of indigo plant of x and y respectively, and D (x, y) is the coordinate distance of x and y, Xx and Xy be x respectively with
The abscissa value of y, Yx and Yy are the ordinate value of x and y respectively, and Ds is the summation of distance, and K is super-pixel block number, for one
For the image of N number of pixel, each super-pixel block size is about N/K pixel, between the adjacent super-pixel block of each two
Distance is S=sqrt (N/K), when algorithm starts, the center Ck=[Lk, ak, bk, xk, yk] of cluster, k is selected to belong to [1, K];
The area of each super-pixel is about square of S.
Step 2 determines formula (1) using Weighting is extracted based on the SLIC super-pixel of MMTD as described above
In weights m.
Step 3, the weights for the image that step is calculated are brought into formula (1), SLIC algorithms generation super-pixel block
When, K seed point is firstly generated, then the nearest several pictures of the detection range seed point in the surrounding space of each seed point
Element, by they be classified as with the seed point one kind, all sort out until all pixels point and finish.Then institute in this K super-pixel is calculated
Have the average vector value of pixel, retrieve K cluster centre, then again with this K center removal search around it with it most
For similar several pixels, all pixels retrieve K super-pixel after all having sorted out, update cluster centre, again iteration, such as
This is repeatedly until convergence.
The present invention compared with prior art, has the advantages that:
Current SLIC super-pixel extracting methods are to determine Lab distances and coordinate distance between pixel with fixed weights
Between proportion.But this mode there are it is many shortcomings that, in some special images, such as large-scale image but Lab colors
The similar image in space and the small-scale image image that still Lab color spaces differ greatly, if weighed using identical distance
Value m, then super-pixel extraction effect may the larger difference of bad student.In order to solve these problems, the present invention is proposed according to every width
The different size of image and Lab color space characteristics, the weights m of metric space is calculated using the method for MMTD, improves biography
The accuracy rate of system SLIC super-pixel extractions.During MMTD is applied to iterative calculation weights by the present invention simultaneously, for image
The Similarity measures of two kinds of distances of Lab distances and coordinate go out suitable weights m so that the effect divided in image superpixel extraction
The phenomenon that fruit may be undesirable, the extraction for influencing super-pixel block is reduced, and improves the effect of image superpixel block extraction.
Description of the drawings
SLIC super-pixel extracting method flow charts of the Fig. 1 based on MMTD.
The different correspondence with pixel brightness value section of Fig. 2 predicates.
Fig. 3 predicates are different with red value of green and the correspondence in champac value section;
Fig. 4 determines weights m flow charts using MMTD.
Specific embodiment
In the following with reference to the drawings and specific embodiments, the present invention is furture elucidated, it should be understood that these examples are merely to illustrate this
It invents rather than limits the scope of the invention, after the present invention has been read, those skilled in the art are various to the present invention's
The modification of equivalent form falls within the application range as defined in the appended claims.
A kind of SLIC super-pixel extracting methods based on MMTD, as shown in figure 4, it includes the following steps:
First, SLIC range formulas
SLIC generation super-pixel is by the pixel cluster based on color similarity and proximity.SLIC is by L, a, b value
In the 5-d spaces defined with x, y pixel coordinate, the pixel cluster of a part is generated.SLIC uses a kind of distance metric, super
Compactedness and regularity are realized in primitive shape, and seamlessly adapts to gray scale and coloured image.SLIC is the face based on image
Color similitude and distance on the image plane generate super-pixel.This is completed in the Labxy spaces of five dimensions, Lab face
Pixel color vector in the colour space, is widely regarded as consistent to the perception of small color distance.One element is brightness L, a and
B is two Color Channels.The color that a includes is again to bright pink from bottle green (low brightness values) to grey (middle brightness value)
(high luminance values);B is again to yellow (high luminance values) from sapphirine (low brightness values) to grey (middle brightness value).And x, y are pictures
Plain position.Input the parameter K of a super-pixel number.So for the image of a N number of pixel, each super-pixel size
About N/K pixel.So, the distance between adjacent super-pixel block of each two is S=sqrt (N/K).When algorithm starts, choosing
The center Ck=[Lk, ak, bk, xk, yk] of cluster is selected, k belongs to [1, K].The area of each super-pixel is about the square (near of S
It is similar to the area of super-pixel).Can safely it assume:Pixel is in the range of the 2S*2S of cluster centre.This range is exactly each
The search area of cluster centre.
Euclidean distance in Lab color spaces is meaningful for short-range perception.If space pixel away from
It is limited from more than this perceived color distance, then the similitude of space pixel begins to the similitude more than pixel color
(super-pixel of generation disrespects zone boundary, only approaches on the image plane).Therefore, it is not using simple in 5D spaces
Euclidean specification, but be defined as follows using a kind of distance scale:
Ds is the summation for testing distance, and x/y plane distance is normalized by grid interval S.One is introduced in Ds
Variable m enables us to control the compactness of a super-pixel.The value of m is bigger, and the distance in space is with regard to closer, and cluster is also
It is compacter.Original SLIC is that m is manually specified, and when different images are handled, inevitably will appear super-pixel block extraction effect
Undesirable situation.
To solve the above-mentioned problems, the present invention is based on the method for MMTD using determining weights, for different image, by
Formula calculates suitable weights, solves the disadvantage that fixed weights.
2nd, MMTD determines weights m
If I is the nonempty set of image pixel, to the pixel value a and b (a, b ∈ I) of any two points in image, there is unique reality
Number L (a, b), A (a, b), B (a, b), D (a, b) are corresponding to it, wherein L (a, b), A (a, b), and B (a, b) is respectively for two pixels
Point brightness value similarity, red green value range similarity, champac value range absolute value of the difference.D (a, b) is European between two pixels
Coordinate distance.Assuming that:There are two pixels in image:X is regarded as point to be investigated by x and y, its brightness value may be 0~100
Between any value, its yellowish green value and blue yellow value may be any value between -128~127.
Remember that predicate P (x, y) represents to treat that investigation point x is similar to y, ╕ P (x, y) represent x, and y is different, and~P (x, y) represents x and y
Between it is similar it is different between, predicate is different as shown in Figure 2 with the correspondence in image brightness values section.
Lx, Ly are the brightness value of x and y respectively, work as Lx>During Ly, corresponding value region is as shown in Fig. 2 left-hand components;
Work as Lx<During Ly as shown in Fig. 2 right-hand components.
It can be obtained by Fig. 2:
The correspondence in the different value of green red with image of predicate and champac value section is as shown in Figure 3.Ax, Ay are x and y respectively
Red value of green, Bx, By are the yellow value of indigo plant of x and y respectively, work as Ax>Ay and Bx>During By, corresponding value region such as Fig. 3 left laterals
Shown in point;Work as Ax<Ay and Bx<During By, as shown in Fig. 3 right-hand components.
It can be obtained by Fig. 3:
The size of L (x, y), A (x, y) and B (x, y) value (truth scale) reflects the similarity of the Lab of pixel x and y,
As L (x, y), when A (x, y) and B (x, y) are equal to 1, the brightness value of x and y, the complete phase of similitude of red value of green and champac value are represented
Together;The value of L (x, y), A (x, y) and B (x, y) are smaller or bigger, represent the similitude of two kinds of distances of x and y with regard to smaller;Work as L
(x, y), A (x, y) and B (x, y) equal to 0 or it is infinitely great when, represent that the similitude of two kinds of distances of x and y is completely different.
By formula (2), (3), (4) determine the similarity L (x, y), A (x, y) and B of brightness value, red value of green and champac value
(x, y) and coordinate Euclidean distance D (x, y), to determine that the proportion function h (x, y) of two kinds of distances of Lab and coordinate is as follows:
Distance proportion function L (x, y), A (x, y) and B (x, y) are calculated by MMTD, picture is then determined according to formula (5)
The similitude h (x, y) of two kinds of distances between vegetarian refreshments.The number of image superpixel block is determined first, is then proposed by formula (5)
The formula of distance rates function calculate each pixel of each super-pixel block and two kinds of distances of center pixel center
Similitude, i.e., each super-pixel block central pixel point and each pixel Lab distances of this super-pixel block are calculated according to formula (5)
With the similarity of two kinds of distances of coordinate distance.Using the half of similarity maximin sum as initial weight, then using changing
The method of generationization calculates weights m of the last weights as each super-pixel block, calculates the weights m of all super-pixel block,
The final weights m of this image is used as with their mean value.
The unstability of extraction super-pixel block is reduced with improved weights m, accurate super-pixel is obtained and gathers soon, specifically
Step is as follows:
As shown in Figure 2,3, according to formula (2), (3), (4) determine pixel and super-pixel block central pixel point to step 1)
Brightness value, the similarity L (x, y), A (x, y) and B (x, y) and coordinate Euclidean distance D (x, y) of red value of green and champac value,
Then distance proportion function h (x, y) is determined according to formula (5), calculate super-pixel block central pixel point center=x (i, j) with
The similarity of each pixel lab distances of super-pixel block where it and two kinds of distances of coordinate distance, if super-pixel block removes middle imago
Vegetarian refreshments one shares n pixel, then the similarity of each pixel of the super-pixel block and two kinds of distances of central point is respectively d1, d2,
D3...dn forms D;
Step 2) determines initial weight m=(max (D)+min (D))/2;
Super-pixel pixel in the block is divided into two parts by step 3) according to weights m, two kinds of pixel and central point away from
M1 is denoted as from Distance conformability degree set of the similarity more than weights m, the Distance conformability degree set less than weights m is denoted as m2;
Step 4) calculates m1, the mean value P1, P2 of m2 respectively;
Step 5) calculates new weights mm=(P1+P2)/2;
Step 6) replaces original weights m to repeat step 3), step 4), step 5) and step 6) with new weights mm, until two
Part weights difference d=m-mm, less than permissible value allow, then iteration terminates, weights of the mm as the block of pixels.
Step 7) calculates the weights m of all super-pixel block, takes weights m of its average value as image.
By determining weights m, the inaccurate influence that fixed weights generate the extraction of image superpixel block is removed, is improved
The accuracy of super-pixel block extraction.
3rd, start to divide
Algorithm firstly generates K seed point, and then the detection range seed point is most in the surrounding space of each seed point
Near several pixels, by they be classified as with seed point one kind, all sort out until all pixels point and finish.Then this K are calculated
The average vector value of all pixels point, retrieves K cluster centre in super-pixel, then again with this K center removal search its
Surrounding and its most similar several pixel, all pixels retrieve K super-pixel after all having sorted out, update cluster centre,
Iteration again, so repeatedly until convergence.
The above is only the preferred embodiment of the present invention, it should be pointed out that:For the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (6)
1. a kind of SLIC super-pixel extraction Weighting based on MMTD, which is characterized in that include the following steps:
Step 1, brightness value similarity L (x, y) between pixel and super-pixel block central pixel point, red value of green similarity are determined
A (x, y), champac value similarity B (x, y) and coordinate Euclidean distance D (x, y);Then according to brightness value similarity L (x, y),
Red value of green similarity A (x, y), champac value similarity B (x, y) and coordinate Euclidean distance D (x, y) determine distance proportion function h
(x,y);Super-pixel block central pixel point center=x (i, j) and picture super where it are calculated by distance proportion function h (x, y)
The similarity of the plain each two kinds of distances of pixel of block, the then similarity of each pixel of the super-pixel block and two kinds of distances of central point
Respectively d1, d2, d3...dn form D, and wherein n represents super-pixel block except central pixel point shares pixel number;
Step 2, weights m=(max (D)+min (D))/2 is determined;
Step 3, super-pixel pixel in the block is divided by two parts according to weights m, two kinds of pixel and central point apart from phase
M1 is denoted as like Distance conformability degree set of the degree more than weights m, the Distance conformability degree set less than weights m is denoted as m2;
Step 4, m1, the mean value P1, P2 of m2 are calculated respectively;
Step 5, new weights mm=(P1+P2)/2 is calculated;
Step 6, replace original weights m with new weights mm, step 3), step 4), step 5) and step 6) are repeated, until two parts
Weights difference d=m-mm, less than permissible value allow, then iteration terminates, using the new weights mm that iteration terminates as the power of the block of pixels
Value;
Step 7, the weights of all super-pixel block are calculated, take weights of its average value as image.
2. the SLIC super-pixel extraction Weighting based on MMTD according to claim 1, it is characterised in that:According to public affairs
Formula (2) determines pixel and the brightness value similarity L (x, y) of super-pixel block central pixel point;
Wherein, L (x, y) represents the brightness value similarity of pixel and super-pixel block central pixel point, (x, y) represent pixel and
Super-pixel block central pixel point, Lx are the brightness values of x, and Ly is the brightness value of y.
3. the SLIC super-pixel extraction Weighting based on MMTD according to claim 2, it is characterised in that:According to public affairs
Formula (3) determines the red value of green similarity A (x, y) of pixel and super-pixel block central pixel point;
Wherein, A (x, y) represents the red value of green similarity of pixel and super-pixel block central pixel point, and Ax is the red value of green of x, Ay
It is the red value of green of y.
4. the SLIC super-pixel extraction Weighting based on MMTD according to claim 3, it is characterised in that:According to public affairs
Formula (4) determines pixel and the champac value similarity B (x, y) of super-pixel block central pixel point;
Wherein, B (x, y) represents the champac value similarity of pixel and super-pixel block central pixel point, and Bx is the blue yellow values of x, Bx, By
It is the yellow value of indigo plant of y.
5. the SLIC super-pixel extraction Weighting based on MMTD according to claim 4, it is characterised in that:According to public affairs
Formula (5) determines distance proportion function h (x, y);
Wherein, D (x, y) represents pixel and the European coordinate distance of super-pixel block central pixel point.
6. a kind of SLIC super-pixel extracting methods based on MMTD, which is characterized in that include the following steps:
Step 1 generates super-pixel block using SLIC algorithms, and the distance metric formula that wherein SLIC is used is as follows:
Wherein, d_Lab is the Lab distances of pixel x and pixel y, and Lx, Ly are the brightness of pixel x and y respectively, and Ax, Ay are x respectively
With the red value of green of y, Bx, By are the yellow value of indigo plant of x and y respectively, and D (x, y) is the coordinate distance of x and y, and Xx and Xy are x and y respectively
Abscissa value, Yx and Yy are the ordinate value of x and y respectively, and Ds is the summation of distance, and K is super-pixel block number, for a N
For the image of a pixel, each super-pixel block size is about the distance between N/K pixel, the adjacent super-pixel block of each two
For S=sqrt (N/K), when algorithm starts, the center Ck=[Lk, ak, bk, xk, yk] of cluster, k is selected to belong to [1, K];Each
The area of super-pixel is square of S;
Step 2, it is true using the extraction Weighting of the SLIC super-pixel based on MMTD as described in claim 1-5 is any
Determine the weights m in formula (1);
Step 3, the weights for the image that step is calculated are brought into formula (1), first when SLIC algorithms generate super-pixel block
K seed point is first generated, then the nearest several pixels of the detection range seed point in the surrounding space of each seed point, it will
They be classified as with the seed point one kind, all sort out until all pixels point and finish;Then all pixels in this K super-pixel are calculated
The average vector value of point, retrieves K cluster centre, then the most similar to it around it with this K center removal search again
Several pixels, all pixels retrieve K super-pixel, update cluster centre, again iteration, so repeatedly after all having sorted out
Until convergence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810109851.8A CN108230340B (en) | 2018-02-05 | 2018-02-05 | Method for determining SLIC superpixel extraction weight based on MMTD and superpixel extraction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810109851.8A CN108230340B (en) | 2018-02-05 | 2018-02-05 | Method for determining SLIC superpixel extraction weight based on MMTD and superpixel extraction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108230340A true CN108230340A (en) | 2018-06-29 |
CN108230340B CN108230340B (en) | 2021-11-26 |
Family
ID=62670578
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810109851.8A Active CN108230340B (en) | 2018-02-05 | 2018-02-05 | Method for determining SLIC superpixel extraction weight based on MMTD and superpixel extraction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108230340B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108881465A (en) * | 2018-07-03 | 2018-11-23 | 肖鑫茹 | A kind of intelligent monitor system based on big data |
CN109472794A (en) * | 2018-10-26 | 2019-03-15 | 北京中科晶上超媒体信息技术有限公司 | A kind of pair of image carries out the method and system of super-pixel segmentation |
CN109523127A (en) * | 2018-10-17 | 2019-03-26 | 平安科技(深圳)有限公司 | Staffs training evaluating method and relevant device based on big data analysis |
CN109977767A (en) * | 2019-02-18 | 2019-07-05 | 浙江大华技术股份有限公司 | Object detection method, device and storage device based on super-pixel segmentation algorithm |
CN110443809A (en) * | 2019-07-12 | 2019-11-12 | 太原科技大学 | Structure sensitive property color images super-pixel method with boundary constraint |
CN113095286A (en) * | 2021-04-30 | 2021-07-09 | 汪知礼 | Big data image processing algorithm and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104123554A (en) * | 2013-04-23 | 2014-10-29 | 南京邮电大学 | SIFT image characteristic extraction method based on MMTD |
CN105809651A (en) * | 2014-12-16 | 2016-07-27 | 吉林大学 | Image saliency detection method based on edge non-similarity comparison |
CN106991683A (en) * | 2017-03-27 | 2017-07-28 | 西安电子科技大学 | Local active contour image segmentation method based on intermediate truth degree measurement |
-
2018
- 2018-02-05 CN CN201810109851.8A patent/CN108230340B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104123554A (en) * | 2013-04-23 | 2014-10-29 | 南京邮电大学 | SIFT image characteristic extraction method based on MMTD |
CN105809651A (en) * | 2014-12-16 | 2016-07-27 | 吉林大学 | Image saliency detection method based on edge non-similarity comparison |
CN106991683A (en) * | 2017-03-27 | 2017-07-28 | 西安电子科技大学 | Local active contour image segmentation method based on intermediate truth degree measurement |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108881465A (en) * | 2018-07-03 | 2018-11-23 | 肖鑫茹 | A kind of intelligent monitor system based on big data |
CN109523127A (en) * | 2018-10-17 | 2019-03-26 | 平安科技(深圳)有限公司 | Staffs training evaluating method and relevant device based on big data analysis |
CN109472794A (en) * | 2018-10-26 | 2019-03-15 | 北京中科晶上超媒体信息技术有限公司 | A kind of pair of image carries out the method and system of super-pixel segmentation |
CN109472794B (en) * | 2018-10-26 | 2021-03-09 | 北京中科晶上超媒体信息技术有限公司 | Method and system for performing superpixel segmentation on image |
CN109977767A (en) * | 2019-02-18 | 2019-07-05 | 浙江大华技术股份有限公司 | Object detection method, device and storage device based on super-pixel segmentation algorithm |
CN109977767B (en) * | 2019-02-18 | 2021-02-19 | 浙江大华技术股份有限公司 | Target detection method and device based on superpixel segmentation algorithm and storage device |
CN110443809A (en) * | 2019-07-12 | 2019-11-12 | 太原科技大学 | Structure sensitive property color images super-pixel method with boundary constraint |
CN113095286A (en) * | 2021-04-30 | 2021-07-09 | 汪知礼 | Big data image processing algorithm and system |
Also Published As
Publication number | Publication date |
---|---|
CN108230340B (en) | 2021-11-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108230340A (en) | A kind of SLIC super-pixel extraction Weighting and super-pixel extracting method based on MMTD | |
CN106778604B (en) | Pedestrian re-identification method based on matching convolutional neural network | |
CN101371274B (en) | Edge comparison in video sequence partition | |
CN111798400A (en) | Non-reference low-illumination image enhancement method and system based on generation countermeasure network | |
CN106056155A (en) | Super-pixel segmentation method based on boundary information fusion | |
CN106295656A (en) | Image outline characteristic extraction method based on image color lump content and device | |
CN107392880A (en) | A kind of imitative pattern painting automatic generation method | |
CN109710791A (en) | A kind of multi-source color image color moving method based on significant filter | |
CN106296620B (en) | A kind of color rendition method based on histogram translation | |
CN108319966A (en) | The method for identifying and classifying of equipment in a kind of substation's complex background infrared image | |
CN110032925A (en) | A kind of images of gestures segmentation and recognition methods based on improvement capsule network and algorithm | |
CN110675462A (en) | Gray level image colorizing method based on convolutional neural network | |
CN107657619A (en) | A kind of low-light (level) Forest fire image dividing method | |
CN105761202B (en) | A kind of color image color moving method | |
CN110276764A (en) | K-Means underwater picture background segment innovatory algorithm based on the estimation of K value | |
CN111583279A (en) | Super-pixel image segmentation method based on PCBA | |
CN113052859A (en) | Super-pixel segmentation method based on self-adaptive seed point density clustering | |
CN107154044A (en) | A kind of dividing method of Chinese meal food image | |
CN105913451B (en) | A kind of natural image superpixel segmentation method based on graph model | |
CN110427868A (en) | A kind of pedestrian identify again in feature extracting method | |
US9613294B2 (en) | Control of computer vision pre-processing based on image matching using structural similarity | |
CN102930289B (en) | Method for generating mosaic picture | |
Mandal et al. | Image segmentation using local thresholding and Ycbcr color space | |
CN113139899A (en) | Design method of high-quality light-weight super-resolution reconstruction network model | |
CN108268533A (en) | A kind of Image Feature Matching method for image retrieval |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |