CN102915541A - Multi-scale image segmenting method - Google Patents

Multi-scale image segmenting method Download PDF

Info

Publication number
CN102915541A
CN102915541A CN2012104250331A CN201210425033A CN102915541A CN 102915541 A CN102915541 A CN 102915541A CN 2012104250331 A CN2012104250331 A CN 2012104250331A CN 201210425033 A CN201210425033 A CN 201210425033A CN 102915541 A CN102915541 A CN 102915541A
Authority
CN
China
Prior art keywords
zone
image
similarity
calculate
merge
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
Application number
CN2012104250331A
Other languages
Chinese (zh)
Other versions
CN102915541B (en
Inventor
刘志
查林
罗书花
沈明华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Shanghai for Science and Technology
Original Assignee
University of Shanghai for Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Shanghai for Science and Technology filed Critical University of Shanghai for Science and Technology
Priority to CN201210425033.1A priority Critical patent/CN102915541B/en
Publication of CN102915541A publication Critical patent/CN102915541A/en
Application granted granted Critical
Publication of CN102915541B publication Critical patent/CN102915541B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention provides a multi-scale image segmenting method. The multi-scale image segmenting method comprises the following specific steps: (1) inputting an original image, pre-segmenting the original image, and establishing a normalized mean offset histogram of each pre-segmented region by a kernel density estimation method; (2) calculating to obtain color similarity values of two adjacent regions; (3) combining the regions to generate a binary tree; and (4) selecting nodes in the binary tree for performing image segmentation. By the multi-scale image segmenting method, the problems of excessive segmentation in the image segmenting process, easy influence of noises in an image on the segmenting method, and high complexity of the segmenting method are solved; and multi-scale segmenting results are helpful to improving working efficiencies of subsequent image analysis, image recognition and other advanced processing stages.

Description

Multiple dimensioned image partition method
Technical field
The present invention relates to the computer image processing technology field, specifically relate to a kind of multiple dimensioned image partition method.
Background technology
Image segmentation is that image is divided into the process that several have the image-region of feature consistency and non-overlapping copies.Desirable image segmentation can extract " image object " corresponding with real world, thereby makes more high-rise image understanding become possibility.Present existing image partition method comprises: based on the dividing method of average drifting (Mean Shift) (referring to document: Comanicu D, Meer P. Mean shift:A robust approach toward feature space analysis. IEEE Trans on Patten Analysis and Machine Intelligence, 2002,24 (5): 603-619.), this method is to come the color cluster in realization character space by the gradient of model space density function, thereby reach the purpose of image segmentation, the shortcoming of the method is that the over-segmentation phenomenon is more serious; Based on the dividing method of normalized cut (Normalized Cuts) (referring to document: F Sun, JP He. A normalized cut based image segmentation method. Information and Computing Science, 2009,2:333-336.), this method be image mapped be one with the non-directed graph of weights, node in the pixel corresponding diagram in the image, the limit of the neighbouring relations corresponding diagram between the pixel, weight in otherness between the pixel characteristic or the similarity corresponding diagram on the limit, then seeking a standardization minimal cut at the figure that sets up divides the node among the figure, and then finish cutting apart image, but this method tends to an isolated point as the target of cutting apart, and therefore easily is subject to the noise in the image; The method that merges (Region Split Merge) based on regional split is (referring to document: Kelkar, S Gupta. Improved Quadtree Method for Split Merge Image Segmentation. Emerging Trends in Engineering and Technology, 2008,44-47.) be to begin constantly to divide to obtain regional from whole image, again the adjacent zone with similarity is merged and obtain segmentation result, the difficult point of regional split merging method is division and the design that merges rule, the shortcoming of this method is that the complexity of algorithm is high, carries out efficient low.
Summary of the invention
The present invention proposes a kind of multiple dimensioned image partition method, solve existing dividing method over-segmentation phenomenon serious, dividing method is subject to the noise in the image, the problem that the dividing method complexity is high; Its multiple dimensioned segmentation result helps to improve follow-up graphical analysis, the work efficiency the processing stage that image recognition etc. being senior.
In order to achieve the above object, the technical solution adopted in the present invention is: above-mentioned multiple dimensioned image partition method, and its concrete steps are as follows:
(1), sets up each regional normalized mean shift histogram step after the pre-segmentation: the input original image, to the original image pre-segmentation, adopt the method for Density Estimator to set up each regional normalized mean shift histogram after the pre-segmentation, its step is as follows:
(1-1), the input original image, with mean shift algorithm original image is pre-segmented into Individual image-region is in the formula
Figure 640711DEST_PATH_IMAGE002
In the presentation video zone
Figure 967787DEST_PATH_IMAGE003
Individual zone, and will
Figure 189821DEST_PATH_IMAGE004
The color space of individual image-region by the RGB color space conversion to the Lab color space;
(1-2), all pixel values of each image-region occur in the statistics Lab color space probability, set up each regional normalized color histogram;
(1-3), select the nucleus vestibularis triangularis function level and smooth to the color histogram convolution that above-mentioned steps (1-2) generates, obtain each regional normalized mean shift histogram;
(2), calculate two neighboring regions of acquisition
Figure 336768DEST_PATH_IMAGE005
With
Figure 551718DEST_PATH_IMAGE006
Color similarity value step: to described each the regional normalized mean shift histogram of above-mentioned steps (1), use formula (1) to calculate and obtain two neighboring regions With The color similarity value:
Figure 88376DEST_PATH_IMAGE007
（1）
In the formula,
Figure 592169DEST_PATH_IMAGE008
With
Figure 628258DEST_PATH_IMAGE009
Represent respectively neighboring region
Figure 254412DEST_PATH_IMAGE005
With
Figure 64367DEST_PATH_IMAGE006
Label,
Figure 434169DEST_PATH_IMAGE010
With
Figure 528027DEST_PATH_IMAGE011
Represent respectively neighboring region
Figure 325081DEST_PATH_IMAGE012
With
Figure 933917DEST_PATH_IMAGE006
Number of pixels,
Figure 28781DEST_PATH_IMAGE013
With
Figure 39462DEST_PATH_IMAGE014
Represent respectively neighboring region With
Figure 103550DEST_PATH_IMAGE006
Normalized mean shift histogram,
Figure 752837DEST_PATH_IMAGE015
It is each pixel
Figure 883604DEST_PATH_IMAGE016
The quantized color value;
(3), the zone merges generation binary tree step: merge neighboring region
Figure 22462DEST_PATH_IMAGE005
With
Figure 25796DEST_PATH_IMAGE006
, generate the merging zone, calculate merging and generate the merging zone Normalized mean shift histogram, its concrete steps are as follows:
(3-1), the size of the color similarity value of neighboring region relatively, iteration is carried out zone merging, its step is as follows:
(3-1-1), merge the most similar neighboring region, use formula (2) to calculate two the most similar neighboring regions
Figure 650999DEST_PATH_IMAGE005
With
Figure 960757DEST_PATH_IMAGE006
, merge With
Figure 756992DEST_PATH_IMAGE006
, and execution in step (3-1-4), otherwise execution in step (3-1-2),
Figure 596772DEST_PATH_IMAGE018
（2）
In the formula,
Figure 77432DEST_PATH_IMAGE003
The expression with
Figure 323868DEST_PATH_IMAGE008
Or
Figure 915386DEST_PATH_IMAGE009
The region labeling of adjacency, Represent current neighboring region number,
Figure 526813DEST_PATH_IMAGE020
It is the label of asking the zone;
(3-1-2), merge the highest neighboring region of similarity in the small size zone, use formula (3) to calculate two the highest neighboring regions of similarity in the small size zone
Figure 244233DEST_PATH_IMAGE005
With , merge
Figure 453815DEST_PATH_IMAGE005
With
Figure 463228DEST_PATH_IMAGE006
, and execution in step (3-1-4), otherwise execution in step,
Figure 730261DEST_PATH_IMAGE021
（3）
In the formula,
Figure 929162DEST_PATH_IMAGE022
Expression small size zone The set that consists of,
Figure 529087DEST_PATH_IMAGE003
Expression and zone
Figure 283417DEST_PATH_IMAGE005
The region labeling of adjacency;
(3-1-3), merge the highest neighboring region of similarity, use formula (3) to calculate two the highest neighboring regions of similarity
Figure 286008DEST_PATH_IMAGE005
With
Figure 494879DEST_PATH_IMAGE006
, merge
Figure 659144DEST_PATH_IMAGE005
With
Figure 900769DEST_PATH_IMAGE006
, and execution in step (3-1-4),
（4）
In the formula,
Figure 757047DEST_PATH_IMAGE024
With
Figure 357792DEST_PATH_IMAGE025
The region labeling that represents respectively adjacency;
(3-1-4), calculate neighboring region
Figure 821135DEST_PATH_IMAGE005
With
Figure 165529DEST_PATH_IMAGE006
Merge to generate and merge the zone
Figure 584877DEST_PATH_IMAGE017
Normalized mean shift histogram, its computing formula is:
Figure 356524DEST_PATH_IMAGE026
In the formula,
Figure 307163DEST_PATH_IMAGE027
It is the zone
Figure 127351DEST_PATH_IMAGE005
With
Figure 479835DEST_PATH_IMAGE006
Merge the label that generates the merging zone,
Figure 422383DEST_PATH_IMAGE028
,
Figure 860318DEST_PATH_IMAGE029
The expression zone
Figure 234930DEST_PATH_IMAGE017
Normalized mean shift histogram;
(3-1-5): if do not satisfy end condition, then proceed the zone and merge execution in step (3-1-1), otherwise execution in step (3-1-6);
If (3-1-6) satisfy end condition, terminator merges and finishes, and the condition of termination is: a surplus next zone after merging;
(3-2), the relation that merges according to the zone sets up a binary tree, its step is as follows:
(3-2-1), with after the original image pre-segmentation
Figure 441920DEST_PATH_IMAGE030
Individual image-region is as the leafy node of binary tree;
(3-2-2), above-mentioned steps (3-1) carried out the merging zone that generates in the regional merging process as father's node of binary tree, and set up thus the data structure of a binary tree;
(4), select the node in the binary tree to finish image segmentation step:
(4-1), the similarity ratio of father and son's node in the definition binary tree
Figure 555370DEST_PATH_IMAGE031
Measure the similarity of father and son's node, its computing formula is:
Figure 480600DEST_PATH_IMAGE032
Figure 642591DEST_PATH_IMAGE033
In the formula,
Figure 969668DEST_PATH_IMAGE034
To calculate father's node
Figure 988439DEST_PATH_IMAGE003
With left child node The similarity ratio,
Figure 615916DEST_PATH_IMAGE036
To calculate father's node
Figure 797498DEST_PATH_IMAGE003
With right child node
Figure 252750DEST_PATH_IMAGE037
The similarity ratio,
Figure 824677DEST_PATH_IMAGE038
To calculate left child node With right child node
Figure 692456DEST_PATH_IMAGE037
Similarity, To calculate left child node
Figure 125635DEST_PATH_IMAGE040
With right child node
Figure 495436DEST_PATH_IMAGE041
Similarity,
Figure 651611DEST_PATH_IMAGE042
To calculate left child node With right child node
Figure 995185DEST_PATH_IMAGE044
Similarity;
(4-2), select in the binary tree
Figure 903098DEST_PATH_IMAGE031
Be worth less (＜=
Figure 881735DEST_PATH_IMAGE031
Maximal value) node is finished multiple dimensioned image segmentation.
Multiple dimensioned image partition method of the present invention compared with prior art has following advantage: the statistical information of describing field color with histogram, effectively reduced the noise in the image, has preferably robustness, reduced the algorithm complex that regional split remerges criterion based on the merging criterion of interregional color similarity, by selecting the node in the binary tree, finish multiple dimensioned image segmentation, its segmentation result can effectively overcome the phenomenon of over-segmentation in the image, is satisfied with different application demands.
Description of drawings
Fig. 1 is the process flow diagram of multiple dimensioned image partition method of the present invention;
Fig. 2 is the original image of input;
Fig. 3 is the result of pre-segmentation;
Fig. 4 (a) is a kind of segmentation result of yardstick;
Fig. 4 (b) is the segmentation result of another kind of yardstick.
Embodiment
Below in conjunction with Figure of description embodiments of the invention are described in further detail.
The emulation experiment that multiple dimensioned image partition method of the present invention carries out be CPU be 2.53GHz, in save as 1.96GB PC test platform programming realize.
Fig. 1 is process flow diagram of the present invention, and at first, the input original image to the original image pre-segmentation, adopts the method for Density Estimator to set up each regional normalized mean shift histogram after the pre-segmentation; Then, calculate the color similarity value that obtains two neighboring regions; Then, the zone merges, and generates binary tree; At last, select the node in the binary tree to finish multiple dimensioned image segmentation, its concrete steps are as follows:
(1), the input original image, as shown in Figure 2, to the original image pre-segmentation, adopt the method for Density Estimator to set up each regional normalized mean shift histogram after the pre-segmentation, its step is as follows:
(1-1), the input original image, as shown in Figure 2, with mean shift algorithm original image is pre-segmented into
Figure 164818DEST_PATH_IMAGE046
Individual image-region wherein is provided with a parameter (＜=1% image area) of controlling the smallest partition area, to obtain the result of a pre-segmentation, as shown in Figure 3, and will To the Lab color space, this conversion more meets the visually-perceptible difference of human eye to the color space of individual image-region by the RGB color space conversion;
(1-2), all pixel values of each image-region occur in the statistics Lab color space probability, set up each regional normalized color histogram;
(1-3), select the nucleus vestibularis triangularis function level and smooth to the color histogram convolution that above-mentioned steps (1-2) generates, obtain each regional normalized mean shift histogram, the color property of selecting histogram to describe the zone here is based on histogram and has preferably robustness;
(2), calculate two neighboring regions of acquisition
Figure 741610DEST_PATH_IMAGE005
With
Figure 818150DEST_PATH_IMAGE006
Color similarity value step: the normalized mean shift histogram that above-mentioned steps (1) is obtained, use formula (1) to calculate and obtain two neighboring regions
Figure 401578DEST_PATH_IMAGE005
With The color similarity value:
Figure 636568DEST_PATH_IMAGE007
（1）
In the formula,
Figure 634742DEST_PATH_IMAGE008
With
Figure 705466DEST_PATH_IMAGE009
Represent respectively neighboring region
Figure 758873DEST_PATH_IMAGE005
With
Figure 598653DEST_PATH_IMAGE006
Label,
Figure 16996DEST_PATH_IMAGE010
With Represent respectively neighboring region
Figure 166534DEST_PATH_IMAGE012
With Number of pixels,
Figure 964912DEST_PATH_IMAGE013
With
Figure 744649DEST_PATH_IMAGE014
Represent respectively neighboring region
Figure 139858DEST_PATH_IMAGE012
With
Figure 891914DEST_PATH_IMAGE006
Normalized mean shift histogram,
Figure 714376DEST_PATH_IMAGE047
It is each pixel
Figure 981410DEST_PATH_IMAGE016
The quantized color value, this formula is zoning respectively
Figure 180310DEST_PATH_IMAGE006
In pixel with the zone
Figure 534674DEST_PATH_IMAGE012
The probable value that obtains of histogram tolerance, and zone
Figure 528038DEST_PATH_IMAGE012
In pixel with the zone
Figure 282368DEST_PATH_IMAGE006
The probable value that obtains of histogram tolerance, this shows,
Figure 19379DEST_PATH_IMAGE048
Value larger, represent two neighboring regions
Figure 480448DEST_PATH_IMAGE005
With
Figure 910292DEST_PATH_IMAGE006
Similarity higher, otherwise then similarity is lower;
(3), the zone merges generation binary tree step: merge neighboring region
Figure 151918DEST_PATH_IMAGE005
With , generate the merging zone, calculate merging and generate the merging zone
Figure 257463DEST_PATH_IMAGE017
Normalized mean shift histogram, its concrete steps are as follows:
(3-1), the size of the color similarity value of neighboring region relatively, iteration is carried out zone merging, its step is as follows:
(3-1-1), merge the most similar neighboring region, use formula (2) to calculate two the most similar neighboring regions
Figure 592629DEST_PATH_IMAGE005
With
Figure 321551DEST_PATH_IMAGE006
, merge
Figure 338048DEST_PATH_IMAGE005
With
Figure 836026DEST_PATH_IMAGE006
, and execution in step (3-1-4), otherwise execution in step (3-1-2),
Figure 607673DEST_PATH_IMAGE049
（2）
In the formula,
Figure 558311DEST_PATH_IMAGE003
The expression with
Figure 129232DEST_PATH_IMAGE008
Or
Figure 481716DEST_PATH_IMAGE009
The region labeling of adjacency,
Figure 424264DEST_PATH_IMAGE019
Represent current neighboring region number,
Figure 862199DEST_PATH_IMAGE020
Be the label of asking the zone, following formula shows and the zone
Figure 220499DEST_PATH_IMAGE012
The similar area of adjacency is
Figure 693068DEST_PATH_IMAGE006
, and with the zone
Figure 806518DEST_PATH_IMAGE006
The similar area of adjacency is
Figure 918699DEST_PATH_IMAGE012
(3-1-2), merge the highest neighboring region of similarity in the small size zone, use formula (3) to calculate two the highest neighboring regions of similarity in the small size zone
Figure 143007DEST_PATH_IMAGE005
With
Figure 204504DEST_PATH_IMAGE006
, merge
Figure 488855DEST_PATH_IMAGE005
With
Figure 839065DEST_PATH_IMAGE006
, and execution in step (3-1-4), otherwise execution in step (3-1-3),
Figure 867064DEST_PATH_IMAGE050
（3）
In the formula, Expression small size zone
Figure 238319DEST_PATH_IMAGE005
Set, namely
Figure 847066DEST_PATH_IMAGE051
, the small size zone refers to
Figure 413176DEST_PATH_IMAGE052
, wherein
Figure 714844DEST_PATH_IMAGE053
,
Figure 75419DEST_PATH_IMAGE054
Be the control parameter, be traditionally arranged to be 0.2,
Figure 400221DEST_PATH_IMAGE055
The area of image, among Fig. 3
Figure 770022DEST_PATH_IMAGE056
, Be current number of regions, because the less zone quantity of information of carrying is less, should be merged into larger zone, the continuity that like this can retaining zone merges;
(3-1-3), merge the highest neighboring region of similarity, use formula (4) to calculate two the highest neighboring regions of similarity
Figure 644623DEST_PATH_IMAGE005
With
Figure 253459DEST_PATH_IMAGE006
, merge
Figure 426951DEST_PATH_IMAGE005
With , and execution in step (3-1-4),
Figure 343272DEST_PATH_IMAGE057
（4）
In the formula,
Figure 439404DEST_PATH_IMAGE024
With
Figure 151008DEST_PATH_IMAGE025
The region labeling that represents respectively adjacency, following formula show, the zone
Figure 704611DEST_PATH_IMAGE012
With
Figure 843469DEST_PATH_IMAGE006
It is the most similar zone in the present all of its neighbor zone;
(3-1-4), calculate neighboring region
Figure 426897DEST_PATH_IMAGE005
With
Figure 676612DEST_PATH_IMAGE006
Merge to generate and merge the zone
Figure 599569DEST_PATH_IMAGE017
Normalized mean shift histogram, its computing formula is:
Figure 909328DEST_PATH_IMAGE058
In the formula, It is the zone
Figure 33458DEST_PATH_IMAGE005
With
Figure 60189DEST_PATH_IMAGE006
Merge the label that generates new region,
Figure 540849DEST_PATH_IMAGE059
,
Figure 833290DEST_PATH_IMAGE029
The expression zone
Figure 690388DEST_PATH_IMAGE017
Normalized mean shift histogram;
(3-1-5): if do not satisfy end condition, then proceed the zone and merge execution in step (3-1-1), otherwise execution in step (3-1-6);
If (3-1-6) satisfy end condition, terminator merges and finishes, and the condition of termination is: a surplus next zone after merging;
(3-2), the relation that merges according to the zone sets up a binary tree, its step is as follows:
(3-2-1), with after the original image pre-segmentation
Figure 322357DEST_PATH_IMAGE046
Individual image-region is as the leafy node of binary tree;
(3-2-2), above-mentioned steps (3-1) is carried out the merging zone that generates in the regional merging process as father's node of binary tree, and set up thus the data structure of a binary tree, in binary tree, what each father's node represented is that two neighboring regions merge the merging zone that generates, and each left child node and right child node represent respectively the neighboring region of two merging;
(4), select the node in the binary tree to finish multiple dimensioned image segmentation step:
(4-1), the similarity ratio of father and son's node in the definition binary tree
Figure 973919DEST_PATH_IMAGE031
Measure the similarity of father and son's node, its computing formula is:
Figure 99930DEST_PATH_IMAGE033
In the formula To calculate father's node
Figure 736765DEST_PATH_IMAGE003
With left child node
Figure 3798DEST_PATH_IMAGE035
The similarity ratio, Be calculate father's node and
Figure 543681DEST_PATH_IMAGE003
Right child node
Figure 802624DEST_PATH_IMAGE037
The similarity ratio,
Figure 556953DEST_PATH_IMAGE038
To calculate left child node
Figure 480916DEST_PATH_IMAGE035
With right child node
Figure 4301DEST_PATH_IMAGE037
Similarity,
Figure 434145DEST_PATH_IMAGE039
To calculate left child node
Figure 410192DEST_PATH_IMAGE040
With right child node
Figure 888577DEST_PATH_IMAGE041
Similarity,
Figure 532048DEST_PATH_IMAGE042
To calculate left child node
Figure 867215DEST_PATH_IMAGE043
With right child node
Figure 596136DEST_PATH_IMAGE044
Similarity,
Figure 363366DEST_PATH_IMAGE031
Less, the similarity of expression father and son node is less, and the zone of such child node representative should be retained in the final segmentation result as a salient region;
(4-2), select in the binary tree
Figure 861344DEST_PATH_IMAGE031
Be worth less (＜= Maximal value) node is finished multiple dimensioned image segmentation, and the result is as follows:
Shown in Fig. 4 (a) is node
Figure 403818DEST_PATH_IMAGE031
The segmentation result of a kind of yardstick of=0.0576 corresponding diagram 2;
Shown in Fig. 4 (b) is node
Figure 756302DEST_PATH_IMAGE031
The segmentation result of the another kind of yardstick of=0.0187 corresponding diagram 2.
Can find out from the result of above-described embodiment, the present invention can obtain multiple dimensioned image segmentation result.

Claims (1)

1. multiple dimensioned image partition method, its concrete steps are as follows:
(1), sets up each regional normalized mean shift histogram step after the pre-segmentation: the input original image, to the original image pre-segmentation, adopt the method for Density Estimator to set up each regional normalized mean shift histogram after the pre-segmentation, its step is as follows:
(1-1), the input original image, with mean shift algorithm original image is pre-segmented into
Figure 2012104250331100001DEST_PATH_IMAGE002
Individual image-region is in the formula
Figure 2012104250331100001DEST_PATH_IMAGE004
In the presentation video zone
Figure 2012104250331100001DEST_PATH_IMAGE006
Individual zone, and will
Figure 2012104250331100001DEST_PATH_IMAGE008
The color space of individual image-region by the RGB color space conversion to the Lab color space;
(1-2), all pixel values of each image-region occur in the statistics Lab color space probability, set up each regional normalized color histogram;
(1-3), select the nucleus vestibularis triangularis function level and smooth to the color histogram convolution that above-mentioned steps (1-2) generates, obtain each regional normalized mean shift histogram;
(2), calculate two neighboring regions of acquisition With Color similarity value step: to described each the regional normalized mean shift histogram of above-mentioned steps (1), use formula (1) to calculate and obtain two neighboring regions With The color similarity value:
（1）
In the formula,
Figure 2012104250331100001DEST_PATH_IMAGE016
With
Figure 2012104250331100001DEST_PATH_IMAGE018
Represent respectively neighboring region With
Figure 696490DEST_PATH_IMAGE012
Label,
Figure 2012104250331100001DEST_PATH_IMAGE020
With
Figure 2012104250331100001DEST_PATH_IMAGE022
Represent respectively neighboring region
Figure 2012104250331100001DEST_PATH_IMAGE024
With
Figure 482044DEST_PATH_IMAGE012
Number of pixels,
Figure 2012104250331100001DEST_PATH_IMAGE026
With
Figure 2012104250331100001DEST_PATH_IMAGE028
Represent respectively neighboring region
Figure 594225DEST_PATH_IMAGE024
With
Figure 818533DEST_PATH_IMAGE012
Normalized mean shift histogram, It is each pixel The quantized color value;
(3), the zone merges generation binary tree step: merge neighboring region With
Figure 367643DEST_PATH_IMAGE012
, generate the merging zone, calculate merging and generate the merging zone
Figure 2012104250331100001DEST_PATH_IMAGE034
Normalized mean shift histogram, its concrete steps are as follows:
(3-1), the size of the color similarity value of neighboring region relatively, iteration is carried out zone merging, its step is as follows:
(3-1-1), merge the most similar neighboring region, use formula (2) to calculate two the most similar neighboring regions With
Figure 231005DEST_PATH_IMAGE012
, merge With
Figure 867840DEST_PATH_IMAGE012
, and execution in step (3-1-4), otherwise execution in step (3-1-2),
Figure 2012104250331100001DEST_PATH_IMAGE036
（2）
In the formula,
Figure 705346DEST_PATH_IMAGE006
The expression with
Figure 271456DEST_PATH_IMAGE016
Or The region labeling of adjacency, Represent current neighboring region number,
Figure 2012104250331100001DEST_PATH_IMAGE040
It is the label of asking the zone;
(3-1-2), merge the highest neighboring region of similarity in the small size zone, use formula (3) to calculate two the highest neighboring regions of similarity in the small size zone
Figure 120649DEST_PATH_IMAGE010
With
Figure 507768DEST_PATH_IMAGE012
, merge
Figure 611991DEST_PATH_IMAGE010
With
Figure 705849DEST_PATH_IMAGE012
, and execution in step (3-1-4), otherwise execution in step,
Figure 2012104250331100001DEST_PATH_IMAGE042
（3）
In the formula, Expression small size zone
Figure 502903DEST_PATH_IMAGE010
The set that consists of,
Figure 797225DEST_PATH_IMAGE006
Expression and zone
Figure 970717DEST_PATH_IMAGE010
The region labeling of adjacency;
(3-1-3), merge the highest neighboring region of similarity, use formula (3) to calculate two the highest neighboring regions of similarity
Figure 981399DEST_PATH_IMAGE010
With
Figure 949355DEST_PATH_IMAGE012
, merge
Figure 983170DEST_PATH_IMAGE010
With
Figure 429195DEST_PATH_IMAGE012
, and execution in step (3-1-4),
Figure 2012104250331100001DEST_PATH_IMAGE046
（4）
In the formula, With The region labeling that represents respectively adjacency;
(3-1-4), calculate neighboring region
Figure 746912DEST_PATH_IMAGE010
With
Figure 885770DEST_PATH_IMAGE012
Merge to generate and merge the zone
Figure 469198DEST_PATH_IMAGE034
Normalized mean shift histogram, its computing formula is:
Figure 2012104250331100001DEST_PATH_IMAGE052
In the formula,
Figure 2012104250331100001DEST_PATH_IMAGE054
It is the zone
Figure 656597DEST_PATH_IMAGE010
With
Figure 641870DEST_PATH_IMAGE012
Merge the label that generates the merging zone, ,
Figure 2012104250331100001DEST_PATH_IMAGE058
The expression zone
Figure 640044DEST_PATH_IMAGE034
Normalized mean shift histogram;
(3-1-5): if do not satisfy end condition, then proceed the zone and merge execution in step (3-1-1), otherwise execution in step (3-1-6);
If (3-1-6) satisfy end condition, terminator merges and finishes, and the condition of termination is: a surplus next zone after merging;
(3-2), the relation that merges according to the zone sets up a binary tree, its step is as follows:
(3-2-1), with after the original image pre-segmentation
Figure 2012104250331100001DEST_PATH_IMAGE060
Individual image-region is as the leafy node of binary tree;
(3-2-2), above-mentioned steps (3-1) carried out the merging zone that generates in the regional merging process as father's node of binary tree, and set up thus the data structure of a binary tree;
(4), select the node in the binary tree to finish image segmentation step:
(4-1), the similarity ratio of father and son's node in the definition binary tree
Figure 2012104250331100001DEST_PATH_IMAGE062
Measure the similarity of father and son's node, its computing formula is:
Figure 2012104250331100001DEST_PATH_IMAGE064
Figure DEST_PATH_IMAGE066
In the formula,
Figure DEST_PATH_IMAGE068
To calculate father's node
Figure 835402DEST_PATH_IMAGE006
With left child node
Figure DEST_PATH_IMAGE070
The similarity ratio,
Figure DEST_PATH_IMAGE072
To calculate father's node With right child node The similarity ratio,
Figure DEST_PATH_IMAGE076
To calculate left child node
Figure 666272DEST_PATH_IMAGE070
With right child node
Figure 832418DEST_PATH_IMAGE074
Similarity,
Figure DEST_PATH_IMAGE078
To calculate left child node
Figure DEST_PATH_IMAGE080
With right child node
Figure DEST_PATH_IMAGE082
Similarity,
Figure DEST_PATH_IMAGE084
To calculate left child node
Figure DEST_PATH_IMAGE086
With right child node
Figure DEST_PATH_IMAGE088
Similarity;
(4-2), select in the binary tree
Figure 62542DEST_PATH_IMAGE062
Be worth less (＜=
Figure 106590DEST_PATH_IMAGE062
Maximal value) node is finished multiple dimensioned image segmentation.
CN201210425033.1A 2012-10-31 2012-10-31 Multi-scale image segmenting method Expired - Fee Related CN102915541B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210425033.1A CN102915541B (en) 2012-10-31 2012-10-31 Multi-scale image segmenting method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210425033.1A CN102915541B (en) 2012-10-31 2012-10-31 Multi-scale image segmenting method

Publications (2)

Publication Number Publication Date
CN102915541A true CN102915541A (en) 2013-02-06
CN102915541B CN102915541B (en) 2015-07-01

Family

ID=47613894

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210425033.1A Expired - Fee Related CN102915541B (en) 2012-10-31 2012-10-31 Multi-scale image segmenting method

Country Status (1)

Country Link
CN (1) CN102915541B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103871050A (en) * 2014-02-19 2014-06-18 小米科技有限责任公司 Image partition method, device and terminal
CN104112007A (en) * 2014-07-16 2014-10-22 深圳大学 Data storage, organization and retrieval methods of image gradation segmentation result
CN105488818A (en) * 2015-11-27 2016-04-13 优渊商贸（上海）有限公司 Vector diagram colour identification method
CN108124168A (en) * 2017-12-28 2018-06-05 洛阳师范学院 A kind of video sharing method perceived under mobile radio network based on user interest
CN108596918A (en) * 2018-04-23 2018-09-28 内蒙古农业大学 A kind of remote sensing image segmentation method merging tree based on level
CN108986175A (en) * 2018-07-24 2018-12-11 中国航发沈阳发动机研究所 Temperature indicating paint regional temperature interpretation method
CN109087311A (en) * 2018-07-24 2018-12-25 中国航发沈阳发动机研究所 Temperature indicating paint temperature interpretation method
CN109741337A (en) * 2018-12-11 2019-05-10 太原理工大学 Region merging technique watershed RS Color Image dividing method based on Lab color space
CN112070090A (en) * 2020-07-31 2020-12-11 南京卫泽科技信息有限公司 Mixed target extraction method based on BPT node analysis in natural scene
CN113168510A (en) * 2018-11-16 2021-07-23 谷歌有限责任公司 Segmenting objects a priori by refining shape

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030174889A1 (en) * 2002-01-08 2003-09-18 Dorin Comaniciu Image segmentation using statistical clustering with saddle point detection
CN102096816A (en) * 2011-01-28 2011-06-15 武汉大学 Multi-scale multi-level image segmentation method based on minimum spanning tree
CN102136145A (en) * 2011-04-27 2011-07-27 中国科学院遥感应用研究所 Region merging method for threshold-restrained minimum spanning tree algorithm

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030174889A1 (en) * 2002-01-08 2003-09-18 Dorin Comaniciu Image segmentation using statistical clustering with saddle point detection
CN102096816A (en) * 2011-01-28 2011-06-15 武汉大学 Multi-scale multi-level image segmentation method based on minimum spanning tree
CN102136145A (en) * 2011-04-27 2011-07-27 中国科学院遥感应用研究所 Region merging method for threshold-restrained minimum spanning tree algorithm

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIN ZHA ET AL.: "A Novel Region Merging Based Image Segmentation Approach for Automatic Object Extraction", 《IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS(ISCAS)》 *
刘志等: "基于二叉划分树的多尺度图像分割算法", 《计算机辅助设计与图形学学报》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103871050A (en) * 2014-02-19 2014-06-18 小米科技有限责任公司 Image partition method, device and terminal
CN103871050B (en) * 2014-02-19 2017-12-29 小米科技有限责任公司 icon dividing method, device and terminal
CN104112007A (en) * 2014-07-16 2014-10-22 深圳大学 Data storage, organization and retrieval methods of image gradation segmentation result
CN104112007B (en) * 2014-07-16 2017-07-25 深圳大学 A kind of data storage, tissue and the search method of image level segmentation result
CN105488818A (en) * 2015-11-27 2016-04-13 优渊商贸（上海）有限公司 Vector diagram colour identification method
CN108124168A (en) * 2017-12-28 2018-06-05 洛阳师范学院 A kind of video sharing method perceived under mobile radio network based on user interest
CN108596918A (en) * 2018-04-23 2018-09-28 内蒙古农业大学 A kind of remote sensing image segmentation method merging tree based on level
CN108986175A (en) * 2018-07-24 2018-12-11 中国航发沈阳发动机研究所 Temperature indicating paint regional temperature interpretation method
CN109087311A (en) * 2018-07-24 2018-12-25 中国航发沈阳发动机研究所 Temperature indicating paint temperature interpretation method
CN109087311B (en) * 2018-07-24 2021-05-28 中国航发沈阳发动机研究所 Temperature judging and reading method for temperature indicating paint
CN108986175B (en) * 2018-07-24 2021-05-28 中国航发沈阳发动机研究所 Temperature interpretation method for temperature indicating paint area
CN113168510A (en) * 2018-11-16 2021-07-23 谷歌有限责任公司 Segmenting objects a priori by refining shape
CN109741337A (en) * 2018-12-11 2019-05-10 太原理工大学 Region merging technique watershed RS Color Image dividing method based on Lab color space
CN109741337B (en) * 2018-12-11 2022-11-29 太原理工大学 Region merging watershed color remote sensing image segmentation method based on Lab color space
CN112070090A (en) * 2020-07-31 2020-12-11 南京卫泽科技信息有限公司 Mixed target extraction method based on BPT node analysis in natural scene

Also Published As

Publication number Publication date
CN102915541B (en) 2015-07-01

Similar Documents

Publication Publication Date Title
CN102915541B (en) Multi-scale image segmenting method
Dornaika et al. Building detection from orthophotos using a machine learning approach: An empirical study on image segmentation and descriptors
CN110866896B (en) Image saliency target detection method based on k-means and level set super-pixel segmentation
CN108629783B (en) Image segmentation method, system and medium based on image feature density peak search
CN108537239B (en) Method for detecting image saliency target
Tian et al. Review of object instance segmentation based on deep learning
CN106815842B (en) improved super-pixel-based image saliency detection method
Kang et al. Dynamic random walk for superpixel segmentation
CN111583279A (en) Super-pixel image segmentation method based on PCBA
CN103093470A (en) Rapid multi-modal image synergy segmentation method with unrelated scale feature
Boiangiu et al. Voting-based image segmentation
Zhou et al. An efficient two-stage region merging method for interactive image segmentation
Scheuermann et al. Slimcuts: Graphcuts for high resolution images using graph reduction
CN110264483B (en) Semantic image segmentation method based on deep learning
CN101702236B (en) Multi-target foreground segmentation method
Li et al. Saliency detection via alternative optimization adaptive influence matrix model
CN113408524A (en) Crop image segmentation and extraction algorithm based on MASK RCNN
Li et al. Edge-based split-and-merge superpixel segmentation
CN102270338B (en) Method for effectively segmenting repeated object based on image representation improvement
CN114627139A (en) Unsupervised image segmentation method, unsupervised image segmentation device and unsupervised image segmentation equipment based on pixel feature learning
CN108765428A (en) A kind of target object extracting method based on click interaction
Liu et al. Saliency detection using two-stage scoring
CN114821059A (en) Salient object detection method and system based on boundary enhancement
Dong et al. Interactive image segmentation with color and texture information by region merging
Li et al. Region proposal ranking via fusion feature for object detection

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20150701

Termination date: 20201031