CN110443817A - A method of improving image segmentation precision - Google Patents
A method of improving image segmentation precision Download PDFInfo
- Publication number
- CN110443817A CN110443817A CN201910535813.3A CN201910535813A CN110443817A CN 110443817 A CN110443817 A CN 110443817A CN 201910535813 A CN201910535813 A CN 201910535813A CN 110443817 A CN110443817 A CN 110443817A
- Authority
- CN
- China
- Prior art keywords
- merge
- pixel
- point
- common
- image
- 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
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T2207/20076—Probabilistic image processing
-
- 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
- G06T2207/20081—Training; Learning
Abstract
A method of improving image segmentation precision, comprising: step 1, the model of setting condition random field；Step 2 designs the parameter based on condition random field；Step 3 designs connection regional search algorithm；Step 4 accurately confines minimum external quadrangle.The present invention uses a kind of method of fining image segmentation for sample problem to improve the precision of image segmentation, its most prominent feature is for segmentation problem of rough, condition random field optimization processing is introduced, it can satisfy the object edge segmentation more refined and interior void filled up, improve segmentation locating effect；For segmented image without obvious frame problem, minimum external quadrangle algorithm is used, the connection region after binaryzation search and minimum frame limits, perfect frame is obtained and limits result.The present invention can be widely applied to framing identification field, such as Logistics Park vehicle identification etc..
Description
Technical field
The present invention relates to a kind of methods for improving image segmentation precision.
Technical background
In recent years with the rapid development of computer science and technology, image procossing, image object based on computer technology
Detection etc. also obtains unprecedented fast development, and wherein deep learning is extracted crucial by the digital picture feature of study magnanimity
Target signature has been more than the mankind in target detection, is brought to industry one and another pleasantly surprised.With neuroid
It rises once again, the video image method based on convolutional Neural metanetwork becomes the mainstream technology of image segmentation and identification, using template
The means such as matching, Edge Gradient Feature, histogram of gradients, realization accurately identify image.Although figure neural network based
Effective feature identification can be carried out for the target of complex scene as feature detects, and its effect is much better than traditional side
Method, but there is also shortcomings: (1) it is to noise anti-interference weaker；(2) over-fitting is solved by using Dropout method
Problem improves convolutional neural networks model and parameter, but precision is but declined slightly；(3) introduce changeable type convolution with can
Convolutional coding structure is separated, the generalization of model is improved, enhances network model ability in feature extraction, but to the target of complex scene
Identification performance is not good enough；(4) newer a kind of image partition method, i.e. End-to-End, direct forecast image pixel classifications information,
The pixel positioning of target object is accomplished, but model the problems such as that there are parameter amounts is big, efficiency is slow, segmentation is coarse.In short, traditional
There is cumbersome, accuracy of identification is not high, recognition efficiency is slow and divides the problems such as coarse for detection method and video image method.
Summary of the invention
In order to overcome the above-mentioned deficiency of the prior art, the present invention provides a kind of fining image segmentation for sample problem
Method the method for the related informations such as pixel spacing, color similarity is arranged using condition random field, meet finer
The object edge segmentation of change and interior void are filled up, and are finally confined using minimum external quadrangle, solve characteristics of image
The problem of boundary survey, is conducive to the further extraction of information.
To achieve the above object, the invention adopts the following technical scheme:
A method of image segmentation precision is improved, is included the following steps:
Step 1, the model of setting condition random field；
Traditional image partition method has " shift-and-stitch " intensively output and method progress using interpolation
Up-sampling operation, but the result that these methods obtain is relatively rough, even if using traditional expansion, corrosion treatment, pixel
Classification results it is still inaccurate.To solve this problem, using a kind of post-processing means of condition random field to pixel classifications
The training stage is intervened afterwards, its classification is made to obtain more accurate pixels probability value, thus reach to object pixel classify into
Row precision positioning.
Condition random field is a kind of undirected graph model of discriminate, for multiple variables or observation sequence x={ x1,
x2,...,xn, i.e., given object pixel value sequence, in given observation or flag sequence y={ y1,y2,...,yn, i.e. classification
Label, building conditional probability model P (y | x).Enable G=<V, E>expression node and the label one-to-one non-directed graph of y, yvIt indicates
Token variable corresponding with node v, n (v) indicate the adjacent node of node v, each variable yvAll meet Markov property, i.e.,
P(yv|x,yv)=P (yv|x,yn(v)) (1)
Then (y, x) constitutes a condition random field, models to it, defines conditional probability P using potential function and group
(y | x), so that token variable { yiAnd adjacent token variable { yi-1,yiComposed by group's potential function it is maximum, pass through selection
Exponential Potential Function, objective function are defined as
In formula: tj(yi+1,yi, x, i) be two adjacent variable mark positions transfer characteristic function, for portraying adjacent mark
Remember the influence of the correlativity and observation sequence of variable to it；sk(yi, x, i) and it is state of the observation sequence on mark position i
Characteristic function, for portraying influence of the observation sequence to token variable；λjAnd μkFor parameter；Z is standardizing factor, for accurate
Define probability.
Step 2 designs the parameter based on condition random field；
For above-mentioned condition random field, in conjunction with the universal model that characteristics of image is classified, the energy potential function used for
In formula: θi(xi) it is unitary potential function；xiFor the tag along sort of the pixel i in observation sequence, then there is class probability P
(xi), convert θi(xi)=- logP (xi).And the pairs of potential function θ of Section 2ij(xi,xj) be extended to
In formula: μ (xi,xj) it is label contrast function, work as xi≠xjWhen, μ (xi,xj)=1, otherwise μ (xi,xj)=0, is used for
Judge the distance between neighbor pixel；wm·km(fi,fj) it is Gaussian convolution core characteristic function, use wmWeigh adjacent pixel point feature
Relationship, physical relationship function are
In formula: piWith pjFor adjacent position pixel coordinate；IiWith IjFor the two colouring information；σαFor location factor；σβFor
The color similarity factor；σγFor the additional Location Scale factor；w1, w2The respectively weight of linear combination.Pass through mean field approximation
FunctionThe K-L divergence that iteration Q (x) minimizes P (x) and Q (x) is updated, the optimal solution of model is obtained.
Step 3 designs connection regional search algorithm；
Image connection regional search method is more, there is pixel point mark method, line segment labelling method etc..Wherein pixel point mark method
It is divided into region growth method, sequential scan method, recursion marking method again.Line segment labelling method is mainly distance of swimming labelling method.And pixel mark
Notation is most common, converts binary map for the prediction result of the logistics vehicles of each classification, is looked by connection region labeling
It looks for.If coordinate is respectively f (x-1, y), f (x+1, y), f (x, y-1) up and down for its left and right pixel f (x, y), f (x, y+1) then joins
Logical region labeling merge (x, y) is scanned in 4 fields, and left, upper position f (x-1, y) and f (x, y- have been scanned when putting by f (x, y)
1), therefore the connectivity of f (x, y), specific discriminate can be determined by judging merge (x-1, y) and merge (x, y-1)
For
1) show the Rule of judgment being connected with left collar domain: as f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1)
When, merge (x, y)=merge (x-1, y).
2) show the Rule of judgment being connected with upper field: as f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y)
When, merge (x, y)=merge (x, y-1).
3) show the Rule of judgment being connected with left, upper field: as f (x, y)=f (x-1, y) and f (x, y)=f (x, y-
1) when, merge (x, y)=merge (x-1, y)=merge (x, y-1).
4) show the Rule of judgment with left, upper field not connection: as f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-
1) when, merge (x, y)=NewLabel new connection label.
Set up an one-dimension array common, under be designated as the value of interim connection region labeling merge (x, y), merge
The value of (x, y) represents some common connection region labeling, the i.e. common connection region labeling common of pixel f (x, y)
(merge(x,y)).Binary map classification image is scanned, detailed process is
1) when there is current coordinate point f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-1), show pixel f
(x, y) belongs to new connection region, and array common is one newly-increased, and record common (merge (x, y))=merge (x,
y)。
2) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y)=f (x-1, y), it is also necessary to relatively more interim
The value of connection region labeling merge (x-1, y) and merge (x, y-1).
If there is merge (x-1, y)=merge (x, y-1) situation then merge (x, y)=merge (x, y-1)；
If occur merge (x-1, y) ≠ merge (x, y-1) situation then when common (i)=common (merge (x-1,
When y)), there is common (i)=common (merge (x, y-1)).
3) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y), then show and upper field
Connection records merge (x, y)=merge (x, y-1).
4) when there is current coordinate point f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1), then show and left collar domain
Connection records merge (x, y)=merge (x-1, y).
After above step, merge all connection regions, obtain the connection region of each classification, picture can be made to target image
Vegetarian refreshments segmentation positioning.
Step 4 accurately confines minimum external quadrangle；
On the basis of segmentation, target is confined using the method for minimum external quadrangle, is conducive to calculate in this way
The high Pixel Information of the width of target.The external quadrangle calculation process of minimum wherein positioned is
1) each classification of step 3 segmented image is switched into bianry image, finds its approximate polygon profile.
2) polygonal profile is made of every series of points, is found y-coordinate maximum, the smallest point of x coordinate in discrete point and is denoted as A
Point.
3) using A as origin, the forward and reverse ray of x-axisScanning clockwise, finds scanning element when rotation angle minimum, is denoted as B
Point.
4) using B point as origin, AB oriented radialScanning, point when finding rotation angle minimum are denoted as C point clockwise.
5) and so on, until finding A point, to obtain polygon P.
6) area rotated each time is calculated with rotary process using P as chimb, obtains minimum area, i.e., minimum external four side
Shape, the height and width of the minimum external quadrangle of record.
The invention has the advantages that
The present invention improves the precision of image segmentation for sample problem using a kind of method for refining image segmentation,
Its most prominent feature is to have introduced condition random field optimization processing for segmentation problem of rough, can satisfy and more refine
Object edge segmentation and interior void are filled up, and segmentation locating effect is improved；For segmented image without obvious frame problem, use
Minimum external quadrangle algorithm search to the connection region after binaryzation and minimum frame limits, obtains perfect side
Frame limits result.The present invention can be widely applied to framing identification field, such as Logistics Park vehicle identification etc..
Detailed description of the invention
Fig. 1 a~Fig. 1 c is the defect schematic diagram of traditional images segmentation, wherein Fig. 1 a is original image, and Fig. 1 b is label, Fig. 1 c
It is prediction；
Fig. 2 a~Fig. 2 f is the front and back comparison of use condition random field of the invention.Before Fig. 2 is use condition random field
Original image, Fig. 2 b label, Fig. 2 c are predictions；Fig. 2 d is the original image after use condition random field, and Fig. 2 e is label, Fig. 2 f prediction；
Fig. 3 a~Fig. 3 b is the external quadrangle positioning of minimum of the invention, and Fig. 3 a is lateral register, and Fig. 3 b is positive positioning.
Specific embodiment
In order to overcome the above-mentioned deficiency of the prior art, the present invention provides a kind of fining image segmentation for sample problem
Method the method for the related informations such as pixel spacing, color similarity is arranged using condition random field, meet finer
The object edge segmentation of change and interior void are filled up, and are finally confined using minimum external quadrangle, solve characteristics of image
The problem of boundary survey, is conducive to the further extraction of information.
To achieve the above object, the invention adopts the following technical scheme:
A method of image segmentation precision is improved, is included the following steps:
Step 1, the model of setting condition random field；
Traditional image partition method has " shift-and-stitch " intensively output and method progress using interpolation
Up-sampling operation, but the result that these methods obtain is relatively rough, even if using traditional expansion, corrosion treatment, pixel
Classification results it is still inaccurate.To solve this problem, using a kind of post-processing means of condition random field to pixel classifications
The training stage is intervened afterwards, its classification is made to obtain more accurate pixels probability value, thus reach to object pixel classify into
Row precision positioning.
Condition random field is a kind of undirected graph model of discriminate, for multiple variables or observation sequence x={ x1,
x2,...,xn, i.e., given object pixel value sequence, in given observation or flag sequence y={ y1,y2,...,yn, i.e. classification
Label, building conditional probability model P (y | x).Enable G=<V, E>expression node and the label one-to-one non-directed graph of y, yvIt indicates
Token variable corresponding with node v, n (v) indicate the adjacent node of node v, each variable yvAll meet Markov property, i.e.,
P(yv|x,yv)=P (yv|x,yn(v)) (1)
Then (y, x) constitutes a condition random field, models to it, defines conditional probability P using potential function and group
(yx), so that token variable { yiAnd adjacent token variable { yi-1,yiComposed by group's potential function it is maximum, referred to by selecting
Number potential function, objective function are defined as
In formula: tj(yi+1,yi, x, i) be two adjacent variable mark positions transfer characteristic function, for portraying adjacent mark
Remember the influence of the correlativity and observation sequence of variable to it；sk(yi, x, i) and it is state of the observation sequence on mark position i
Characteristic function, for portraying influence of the observation sequence to token variable；λjAnd μkFor parameter；Z is standardizing factor, for accurate
Define probability.
Step 2 designs the parameter based on condition random field；
For above-mentioned condition random field, in conjunction with the universal model that characteristics of image is classified, the energy potential function used for
In formula: θi(xi) it is unitary potential function；xiFor the tag along sort of the pixel i in observation sequence, then there is class probability P
(xi), convert θi(xi)=- logP (xi).And the pairs of potential function θ of Section 2ij(xi,xj) be extended to
In formula: μ (xi,xj) it is label contrast function, work as xi≠xjWhen, μ (xi,xj)=1, otherwise μ (xi,xj)=0, is used for
Judge the distance between neighbor pixel；wm·km(fi,fj) it is Gaussian convolution core characteristic function, use wmWeigh adjacent pixel point feature
Relationship, physical relationship function are
In formula: piWith pjFor adjacent position pixel coordinate；IiWith IjFor the two colouring information；σαFor location factor；σβFor
The color similarity factor；σγFor the additional Location Scale factor；w1, w2The respectively weight of linear combination.Pass through mean field approximation
FunctionThe K-L divergence that iteration Q (x) minimizes P (x) and Q (x) is updated, the optimal solution of model is obtained.
Step 3 designs connection regional search algorithm；
Image connection regional search method is more, there is pixel point mark method, line segment labelling method etc..Wherein pixel point mark method
It is divided into region growth method, sequential scan method, recursion marking method again.Line segment labelling method is mainly distance of swimming labelling method.And pixel mark
Notation is most common, converts binary map for the prediction result of the logistics vehicles of each classification, is looked by connection region labeling
It looks for.If coordinate is respectively f (x-1, y), f (x+1, y), f (x, y-1) up and down for its left and right pixel f (x, y), f (x, y+1) then joins
Logical region labeling merge (x, y) is scanned in 4 fields, and left, upper position f (x-1, y) and f (x, y- have been scanned when putting by f (x, y)
1), therefore the connectivity of f (x, y), specific discriminate can be determined by judging merge (x-1, y) and merge (x, y-1)
For
1) show the Rule of judgment being connected with left collar domain: as f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1)
When, merge (x, y)=merge (x-1, y).
2) show the Rule of judgment being connected with upper field: as f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y)
When, merge (x, y)=merge (x, y-1).
3) show the Rule of judgment being connected with left, upper field: as f (x, y)=f (x-1, y) and f (x, y)=f (x, y-
1) when, merge (x, y)=merge (x-1, y)=merge (x, y-1).
4) show the Rule of judgment with left, upper field not connection: as f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-
1) when, merge (x, y)=NewLabel new connection label.
Set up an one-dimension array common, under be designated as the value of interim connection region labeling merge (x, y), merge
The value of (x, y) represents some common connection region labeling, the i.e. common connection region labeling common of pixel f (x, y)
(merge(x,y)).Binary map classification image is scanned, detailed process is
1) when there is current coordinate point f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-1), show pixel f
(x, y) belongs to new connection region, and array common is one newly-increased, and record common (merge (x, y))=merge (x,
y)。
2) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y)=f (x-1, y), it is also necessary to relatively more interim
The value of connection region labeling merge (x-1, y) and merge (x, y-1).
If there is merge (x-1, y)=merge (x, y-1) situation then merge (x, y)=merge (x, y-1)；
If occur merge (x-1, y) ≠ merge (x, y-1) situation then when common (i)=common (merge (x-1,
When y)), there is common (i)=common (merge (x, y-1)).
3) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y), then show and upper field
Connection records merge (x, y)=merge (x, y-1).
4) when there is current coordinate point f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1), then show and left collar domain
Connection records merge (x, y)=merge (x-1, y).
After above step, merge all connection regions, obtain the connection region of each classification, picture can be made to target image
Vegetarian refreshments segmentation positioning.
Step 4 accurately confines minimum external quadrangle；
On the basis of segmentation, target is confined using the method for minimum external quadrangle, is conducive to calculate in this way
The high Pixel Information of the width of target.The external quadrangle calculation process of minimum wherein positioned is
1) each classification of step 3 segmented image is switched into bianry image, finds its approximate polygon profile.
2) polygonal profile is made of every series of points, is found y-coordinate maximum, the smallest point of x coordinate in discrete point and is denoted as A
Point.
3) using A as origin, the forward and reverse ray of x-axisScanning clockwise, finds scanning element when rotation angle minimum, is denoted as B
Point.
4) using B point as origin, AB oriented radialScanning, point when finding rotation angle minimum are denoted as C point clockwise.
5) and so on, until finding A point, to obtain polygon P.
6) area rotated each time is calculated with rotary process using P as chimb, obtains minimum area, i.e., minimum external four side
Shape, the height and width of the minimum external quadrangle of record.
In order to verify the superiority of the invention, using Logistics Park vehicle as example, following network model is constructed, is compareed
Experiment:
It builds lightweight and conditional random field models network structure: acquiring cargo, towed goods from Logistics Park
Vehicle, dumper, four seed type of tank truck logistics vehicles, be divided into training set 8 000, each classification 2 000 is surveyed
Examination collection 4 000, each classification 1 000.Each parameter configuration of the network architecture built is as shown in table 1 below.
In table 1: k is convolution kernel size；S is step-length；P is the size of filling；DW is channel convolution group, indicates channel convolution
The regular collocation of core composition；Residual error summation has been used to be conducive to the gradient transmitting of big network；The activation of each layer and batch standardize
Operation (Batch Normalization, BN) is conducive to accelerate the training of network；ReLU is amendment linear unit, is one and swashs
Function living.
Each parameter designing of 1 network architecture of table
After having introduced condition random field optimization, segmentation locating effect is improved.Use condition random field segmentation before with make
Comparative result after being divided with condition random field is as shown in Figure of description 2.
For segmented image without obvious frame problem, minimum external quadrangle algorithm is used, to the connection after binaryzation
Region search and minimum frame limits.Perfect frame is obtained to limit as a result, as shown in Figure of description 3.No matter logistics
How is vehicle direction, and the bezel locations after segmentation positioning can be limited in minimum rectangle frame.
The invention has the advantages that
The present invention improves the precision of image segmentation for sample problem using a kind of method for refining image segmentation,
Its most prominent feature is to have introduced condition random field optimization processing for segmentation problem of rough, can satisfy and more refine
Object edge segmentation and interior void are filled up, and segmentation locating effect is improved；For segmented image without obvious frame problem, use
Minimum external quadrangle algorithm search to the connection region after binaryzation and minimum frame limits, obtains perfect side
Frame limits result.The present invention can be widely applied to framing identification field, such as Logistics Park vehicle identification etc..
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. a kind of method for improving image segmentation precision, includes the following steps:
Step 1, the model of setting condition random field；
The rear training stage of pixel classifications is intervened using a kind of post-processing means of condition random field, obtains its classification
More accurate pixels probability value carries out precision positioning to object pixel classification to reach；
Condition random field is a kind of undirected graph model of discriminate, for multiple variables or observation sequence x={ x1,x2,...,
xn, i.e., given object pixel value sequence, in given observation or flag sequence y={ y1,y2,...,yn, i.e. class label, structure
Build conditional probability model P (y | x)；Enable G=<V, E>expression node and the label one-to-one non-directed graph of y, yvIt indicates and node v
Corresponding token variable, n (v) indicate the adjacent node of node v, each variable yvAll meet Markov property, i.e.,
P(yv|x,yv)=P (yv|x,yn(v)) (1)
Then (y, x) constitute a condition random field, it is modeled, using potential function and group come define conditional probability P (y |
X), so that token variable { yiAnd adjacent token variable { yi-1,yiComposed by group's potential function it is maximum, pass through and select index
Potential function, objective function are defined as
In formula: tj(yi+1,yi, x, i) be two adjacent variable mark positions transfer characteristic function, for portray adjacent marker become
Influence of the correlativity and observation sequence of amount to it；sk(yi, x, i) and it is state feature of the observation sequence on mark position i
Function, for portraying influence of the observation sequence to token variable；λjAnd μkFor parameter；Z is standardizing factor, is used for accurate definition
Probability；
Step 2 designs the parameter based on condition random field；
For above-mentioned condition random field, in conjunction with the universal model that characteristics of image is classified, the energy potential function used for
In formula: θi(xi) it is unitary potential function；xiFor the tag along sort of the pixel i in observation sequence, then there is class probability P (xi),
Convert θi(xi)=- logP (xi)；And the pairs of potential function θ of Section 2ij(xi,xj) be extended to
In formula: μ (xi,xj) it is label contrast function, work as xi≠xjWhen, μ (xi,xj)=1, otherwise μ (xi,xj)=0, for judging
Distance between neighbor pixel；wm·km(fi,fj) it is Gaussian convolution core characteristic function, use wmWeigh adjacent pixel point feature to close
System, physical relationship function are
In formula: piWith pjFor adjacent position pixel coordinate；IiWith IjFor the two colouring information；σαFor location factor；σβFor color
The similarity factor；σγFor the additional Location Scale factor；w1, w2The respectively weight of linear combination；Pass through mean field approximation functionThe K-L divergence that iteration Q (x) minimizes P (x) and Q (x) is updated, the optimal solution of model is obtained；
Step 3 designs connection regional search algorithm；
Image connection regional search method is more, there is pixel point mark method, line segment labelling method etc.；Wherein pixel point mark method divides again
For region growth method, sequential scan method, recursion marking method；Line segment labelling method is mainly distance of swimming labelling method；And pixel point mark method
It is most common, binary map is converted by the prediction result of the logistics vehicles of each classification, is searched by connection region labeling；If
Coordinate is respectively f (x-1, y), f (x+1, y), f (x, y-1), f (x, y+1) up and down for its left and right pixel f (x, y), then connection area
Domain label merge (x, y) is scanned in 4 fields, and left, upper position f (x-1, y) and f (x, y-1) have been scanned when putting by f (x, y),
Therefore the connectivity of f (x, y) can be determined by judging merge (x-1, y) and merge (x, y-1), specific discriminate is
S1) show the Rule of judgment being connected with left collar domain: as f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1),
Merge (x, y)=merge (x-1, y)；
S2) show the Rule of judgment being connected with upper field: as f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y),
Merge (x, y)=merge (x, y-1)；
S3) show the Rule of judgment being connected with left, upper field: as f (x, y)=f (x-1, y) and f (x, y)=f (x, y-1)
When, merge (x, y)=merge (x-1, y)=merge (x, y-1)；
S4) show the Rule of judgment with left, upper field not connection: as f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-1)
When, merge (x, y)=NewLabel new connection label；
Set up an one-dimension array common, under be designated as the value of interim connection region labeling merge (x, y), merge (x, y)
Value represent some common connection region labeling, i.e. the common connection region labeling common of pixel f (x, y) (merge (x,
y))；Binary map classification image is scanned, detailed process is
T1) when there is current coordinate point f (x, y) ≠ f (x-1, y) and f (x, y) ≠ f (x, y-1), show pixel f (x,
Y) belong to new connection region, array common is one newly-increased, and records common (merge (x, y))=merge (x, y)；
T2) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y)=f (x-1, y), it is also necessary to relatively more interim connection
The value of logical region labeling merge (x-1, y) and merge (x, y-1)；
If there is merge (x-1, y)=merge (x, y-1) situation then merge (x, y)=merge (x, y-1)；
If there is merge (x-1, y) ≠ merge (x, y-1) situation then as common (i)=common (merge (x-1, y))
When, there is common (i)=common (merge (x, y-1))；
T3) when there is current coordinate point f (x, y)=f (x, y-1) and f (x, y) ≠ f (x-1, y), then show to join with upper field
It is logical, it records merge (x, y)=merge (x, y-1)；
T4) when there is current coordinate point f (x, y)=f (x-1, y) and f (x, y) ≠ f (x, y-1), then show to join with left collar domain
It is logical, it records merge (x, y)=merge (x-1, y)；
After above step, merge all connection regions, obtain the connection region of each classification, pixel can be done to target image
Segmentation positioning；
Step 4 accurately confines minimum external quadrangle；
On the basis of segmentation, target is confined using the method for minimum external quadrangle, is conducive to calculate target in this way
The high Pixel Information of width；The external quadrangle calculation process of minimum wherein positioned are as follows:
1) each classification of step 3 segmented image is switched into bianry image, finds its approximate polygon profile；
2) polygonal profile is made of every series of points, is found y-coordinate maximum, the smallest point of x coordinate in discrete point and is denoted as A point；
3) using A as origin, the forward and reverse ray of x-axisScanning clockwise, finds scanning element when rotation angle minimum, is denoted as B point；
4) using B point as origin, AB oriented radialScanning, point when finding rotation angle minimum are denoted as C point clockwise；
5) and so on, until finding A point, to obtain polygon P；
6) area rotated each time is calculated with rotary process using P as chimb, obtain minimum area, i.e., minimum external quadrangle, note
The height and width of the minimum external quadrangle of record.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910535813.3A CN110443817B (en) | 2019-06-20 | 2019-06-20 | Method for improving image segmentation precision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910535813.3A CN110443817B (en) | 2019-06-20 | 2019-06-20 | Method for improving image segmentation precision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110443817A true CN110443817A (en) | 2019-11-12 |
| CN110443817B CN110443817B (en) | 2021-02-02 |
Family
ID=68428313
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910535813.3A Active CN110443817B (en) | 2019-06-20 | 2019-06-20 | Method for improving image segmentation precision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443817B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113129323A (en) * | 2021-04-27 | 2021-07-16 | 西安微电子技术研究所 | Remote sensing ridge boundary detection method and system based on artificial intelligence, computer equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105321176A (en) * | 2015-09-30 | 2016-02-10 | 西安交通大学 | Image segmentation method based on hierarchical higher order conditional random field |
| CN106997466A (en) * | 2017-04-12 | 2017-08-01 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detecting road |
| US20180181842A1 (en) * | 2016-12-22 | 2018-06-28 | TCL Research America Inc. | Method and device for quasi-gibbs structure sampling by deep permutation for person identity inference |
| CN108876795A (en) * | 2018-06-07 | 2018-11-23 | 四川斐讯信息技术有限公司 | A kind of dividing method and system of objects in images |
| CN109285162A (en) * | 2018-08-30 | 2019-01-29 | 杭州电子科技大学 | A kind of image, semantic dividing method based on regional area conditional random field models |
-
2019
- 2019-06-20 CN CN201910535813.3A patent/CN110443817B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105321176A (en) * | 2015-09-30 | 2016-02-10 | 西安交通大学 | Image segmentation method based on hierarchical higher order conditional random field |
| US20180181842A1 (en) * | 2016-12-22 | 2018-06-28 | TCL Research America Inc. | Method and device for quasi-gibbs structure sampling by deep permutation for person identity inference |
| CN106997466A (en) * | 2017-04-12 | 2017-08-01 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detecting road |
| CN108876795A (en) * | 2018-06-07 | 2018-11-23 | 四川斐讯信息技术有限公司 | A kind of dividing method and system of objects in images |
| CN109285162A (en) * | 2018-08-30 | 2019-01-29 | 杭州电子科技大学 | A kind of image, semantic dividing method based on regional area conditional random field models |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113129323A (en) * | 2021-04-27 | 2021-07-16 | 西安微电子技术研究所 | Remote sensing ridge boundary detection method and system based on artificial intelligence, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110443817B (en) | 2021-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738207B (en) | Character detection method for fusing character area edge information in character image | |
| CN114120102A (en) | Boundary-optimized remote sensing image semantic segmentation method, device, equipment and medium | |
| Liu et al. | Panoptic feature fusion net: a novel instance segmentation paradigm for biomedical and biological images | |
| CN111523553B (en) | Central point network multi-target detection method based on similarity matrix | |
| CN111738055B (en) | Multi-category text detection system and bill form detection method based on same | |
| CN113223068B (en) | Multi-mode image registration method and system based on depth global features | |
| US20170076448A1 (en) | Identification of inflammation in tissue images | |
| CN106815323B (en) | Cross-domain visual retrieval method based on significance detection | |
| Mosinska et al. | Joint segmentation and path classification of curvilinear structures | |
| CN112949338A (en) | Two-dimensional bar code accurate positioning method combining deep learning and Hough transformation | |
| Chen et al. | Vectorization of historical maps using deep edge filtering and closed shape extraction | |
| Lv et al. | Nuclei R-CNN: improve mask R-CNN for nuclei segmentation | |
| CN113158895A (en) | Bill identification method and device, electronic equipment and storage medium | |
| CN111507337A (en) | License plate recognition method based on hybrid neural network | |
| CN106845458A (en) | A kind of rapid transit label detection method of the learning machine that transfinited based on core | |
| Zhou et al. | Attention transfer network for nature image matting | |
| CN116645592A (en) | Crack detection method based on image processing and storage medium | |
| Du et al. | Improved detection method for traffic signs in real scenes applied in intelligent and connected vehicles | |
| CN111210447A (en) | Method and terminal for hierarchical segmentation of hematoxylin-eosin staining pathological image | |
| Naiemi et al. | Scene text detection using enhanced extremal region and convolutional neural network | |
| CN110443817A (en) | A method of improving image segmentation precision | |
| CN105844299B (en) | A kind of image classification method based on bag of words | |
| Yao et al. | Encoder–decoder with pyramid region attention for pixel‐level pavement crack recognition | |
| Zuo et al. | A pixel-level segmentation convolutional neural network based on global and local feature fusion for surface defect detection | |
| Patil et al. | Road segmentation in high-resolution images using deep residual networks |
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 |