CN113052859A - Super-pixel segmentation method based on self-adaptive seed point density clustering - Google Patents

Super-pixel segmentation method based on self-adaptive seed point density clustering Download PDF

Info

Publication number
CN113052859A
CN113052859A CN202110426049.3A CN202110426049A CN113052859A CN 113052859 A CN113052859 A CN 113052859A CN 202110426049 A CN202110426049 A CN 202110426049A CN 113052859 A CN113052859 A CN 113052859A
Authority
CN
China
Prior art keywords
seed
pixel
point
points
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.)
Pending
Application number
CN202110426049.3A
Other languages
Chinese (zh)
Inventor
王卫兵
何金喜
张晓琢
郑岩
权霜霜
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.)
Harbin University of Science and Technology
Original Assignee
Harbin University of 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 Harbin University of Science and Technology filed Critical Harbin University of Science and Technology
Priority to CN202110426049.3A priority Critical patent/CN113052859A/en
Publication of CN113052859A publication Critical patent/CN113052859A/en
Pending legal-status Critical Current

Links

Images

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • 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
    • G06T2207/20021—Dividing image into blocks, subimages or windows

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a super-pixel segmentation method based on self-adaptive seed point density clustering, which mainly comprises the following steps: firstly, preprocessing an input image, dividing the image into a plurality of sub image blocks which are approximately same, then searching the maximum inscribed circle center on the sub image blocks by using a region growing method, adaptively obtaining the position of a seed point, clustering outwards in sequence by taking the selected seed point as the center according to the density of the adaptive seed point, and taking the seed point fully distributed on the whole image as an end condition, thereby generating an initial superpixel. Secondly, further clustering the sub-image blocks by using a space-limited k-medoids method, and further updating the seed points. And finally, cleaning the unprocessed abnormal pixel points. The invention firstly combines the self-adaptive seed points and the density clustering and introduces the self-adaptive seed points and the density clustering into the superpixel segmentation so as to generate the superpixel with more regular shape and better boundary fitting degree.

Description

Super-pixel segmentation method based on self-adaptive seed point density clustering
The technical field is as follows:
the invention belongs to a preprocessing technology of image segmentation in the field of computer vision, and particularly relates to a super-pixel segmentation method based on self-adaptive seed point density clustering.
Background art:
in the field of computer vision, the problem of digital image processing has been a significant challenge facing people. With the continuous development of image technology, a large amount of images are generated every day, and image segmentation plays an extremely important role in digital image processing. In the conventional image segmentation, the processing of pixel points is taken as a basic unit, however, under the rapid development of modern technologies, the size of an image to be processed is continuously increased, and in terms of efficiency, the current requirement is difficult to meet by a conventional pixel point processing mode. In 2003, superpixels were introduced by REN et al as a new idea, which refers to a pixel region composed of pixels with similar texture and color characteristics.
Many superpixel segmentation methods have been proposed at present, which include two major categories, one is the idea based on graph theory, and the other is the idea based on gradient.
The segmentation idea based on the graph theory is that pixel block segmentation is regarded as an optimization function problem, pixel points of an image are regarded as vertexes of the image, edges connected with the vertexes represent the mutual relations of the pixel points, then segmentation processing is carried out on the nodes in the image through a series of segmentation standards, and then superpixel segmentation is completed. The regularization segmentation algorithm adopts iterative graph segmentation of image ripple and outline characteristics. The method has the advantages that the generated superpixels are regular, but the boundary is not tightly attached and is not high in calculation efficiency, so that the method is not suitable for segmenting large-size images.
The idea of gradient-based segmentation is to first initialize a pixel cluster and then re-check the result of the cluster until the requirement is met, thus obtaining superpixels. The watershed method with shape constraint is to improve the traditional watershed algorithm and add space limitation, so that more compact superpixels are generated, the shape of the superpixels is more regular, and when the image with complex environment is solved, the boundary information cannot be accurately processed. The mean shift algorithm is to divide the pixels with similar characteristics into superpixels by using the local maximum density function. Its advantages are good adhesion of boundary and noise resistance, and poor rule of super-pixel. The simple linear iterative clustering method is characterized in that the total number of superpixels is regulated by the number of pixel points in the current superpixel, a quantitative superpixel set is distributed in an image, and the local pixels are subjected to iterative clustering processing based on a given color similarity and spatial distance formula, so that the target superpixel is generated.
The invention content is as follows:
the invention discloses a super-pixel segmentation method based on self-adaptive seed point density clustering, which aims to solve the problems that the conventional super-pixel segmentation method based on density clustering cannot obtain a regular super-pixel shape, has low image boundary fitting degree, cannot process abnormal pixel points in time and the like. FIG. 1 is a flow chart of a super-pixel segmentation method of the present invention.
Therefore, the invention provides the following technical scheme:
step 1: dividing an input image into a plurality of sub image blocks which are approximately same, and then searching the center coordinate of the maximum inscribed circle on the sub image blocks by using a region growing method so as to self-adaptively obtain the positions of the seed points.
Step 2: and uniformly spreading seed points on the preprocessed image, and sequentially clustering outwards by using the selected seed points as centers according to the self-adaptive seed point density cluster until the seed points are fully distributed in the whole image, thereby finishing the initial superpixel segmentation.
And step 3: and clustering the seed points on all the obtained sub-image blocks by combining a space-limited k-medoids method, further updating the seed points, and solving the influence of noise on the method.
And 4, step 4: and traversing the seed points in sequence according to the idea of combining the top-down and bottom-up, and cleaning abnormal pixel points which cannot be processed in time.
The technical effect of the technical scheme is as follows: the invention firstly combines the self-adaptive seed points and the density clustering together and introduces the combination into the super-pixel segmentation so as to generate the super-pixel with more regular shape and better boundary fitting degree. On the basis, a k-medoids clustering method is combined, the automatically obtained seed center point is updated, and the image segmentation precision and the boundary recall rate are effectively improved. And finally, traversing the seed points by using a new idea to clear the abnormal pixel points.
Further, in the step 1, the specific steps are as follows:
step 1-1, acquiring all pixel information of a region to be segmented as input, and performing preprocessing operation on an image;
step 1-2, filtering and screening according to the gray value range between the pixel points to be segmented, and dividing the image into a plurality of sub image blocks with similar sizes and similar colors, wherein the color range of the sub image block is within the gray value range of the area to be segmented;
step 1-3, calculating and searching the center coordinates of the maximum inscribed circle on the four neighborhoods of the sub-image block by using a region growing method, and determining the positions of the center coordinates (X, Y) by using a standard deviation method. And after the coordinates are determined, automatically selecting the seed points. The specific calculation method of the circle center coordinate is as follows:
Figure BDA0003029588150000031
Figure BDA0003029588150000032
wherein the content of the first and second substances,
Figure BDA0003029588150000033
in order to count the number of seed points in the horizontal search,
Figure BDA0003029588150000034
the number of the seed points in the longitudinal search is shown, and n is the number of the pixel points.
The technical effect of the technical scheme is as follows: the error caused by artificially and subjectively selecting the seed points is avoided, and the post-processing operation is convenient.
Further, in the step 2, the specific steps are as follows:
step 2-1 converts the rgb color space of the image to be processed to the lab color space.
Step 2-2, respectively defining the two sets as a mark set and a candidate set, then locking an area to be segmented, adaptively selecting a seed point position according to the calculated circle center coordinate, adding the seed point position into the mark set, and if the mark set exists at the point, turning to step 2, and restarting the search of the next target area;
and 2-3, counting four pixel sets including a seed pixel set, a marked pixel set, an unmarked pixel set and a candidate pixel set. Wherein the seed pixel set stores all the seed information, and the marker pixel set is reset when the seed point changes.
Step 2-4, measuring the effect after clustering by adopting a standard function of absolute deviation, wherein the specific calculation method comprises the following steps:
Figure BDA0003029588150000035
wherein: p is cluster deltajObject of δjAs cluster center, p and δjIs multi-dimensional.
The technical effect of the technical scheme is as follows: and the pixel information is effectively classified, so that the subsequent processing is facilitated.
Further, in the step 3, the specific steps are as follows:
and 3-1, combining a space-limited k-medoids method to ensure that the number of pixels in the clustered cluster is larger than a given threshold value s/n, wherein s represents the size of the input image, and n represents the number of super-pixels required by a user. The space limitation is to limit the search range of the pixel points starting from the seed points to the range of 2s multiplied by 2s in order to make the algorithm converge as soon as possible;
step 3-2 changes the marker set to an empty set, which indicates that the neighboring pixels of the previous marker set will be located in the boundary region, and the current condition helps the algorithm to stop at the boundary;
3-3, automatically selecting new seed points from the unmarked pixel set according to a clustering algorithm, and repeatedly executing the process until all the seed points are marked, otherwise, returning to the step 3-1 and marking again;
and 3-4, generating initial superpixels by the obtained clusters of different labels.
The specific calculation method is as follows:
Figure BDA0003029588150000041
Figure BDA0003029588150000042
Figure BDA0003029588150000043
Figure BDA0003029588150000044
wherein lq,aq,bqRepresenting the color characteristic of the pixel point q; lu,au,buThe color characteristics of the marker point μ are shown. lt,at,btThe color characteristics of the seed point t are shown. oq,cqThe spatial characteristics of the pixel point q are represented; ot,utThe spatial characteristics of the seed point t are represented. Theta is a parameter and is used for adjusting the size of an inscribed circle of the region to be segmented; delta is a boundary limiting parameter of the seed point and is used for adjusting the effective position of the seed point q on the boundary.
Figure BDA0003029588150000045
Representing the color difference between the mark point u and the pixel point q,
Figure BDA0003029588150000046
representing the color distance, d, between the seed point t and the pixel point qsThe distance between the seed point t and the pixel point q in space, and the distance between the marking point and the seed point to the pixel point.
The technical effect of the technical scheme is as follows: in different clusters, seed points are automatically selected according to the method to generate initial superpixels.
Further, in the step 4, the specific process is as follows:
step 4-1, sequentially traversing the pixel points which are not visited according to the idea of combining top-down and bottom-up;
and 4-2, calculating and finding the minimum difference between the minimum difference and all the seed points, wherein the seed points near the pixel point accord with the interval condition, and the cleaning is finished. The specific calculation method is as follows:
Figure BDA0003029588150000051
Figure BDA0003029588150000052
Figure BDA0003029588150000053
wherein d is1Representing the interval between a pixel point i and a clustering central point upsilon space; d2And representing the color distance between the pixel point i and the seed central point upsilon. α is a parameter for adjusting the color distance. τ denotes the ratio of the parameters of the results after and before segmentation.
Description of the drawings:
FIG. 1 is a flow chart of a superpixel segmentation method of the present invention;
fig. 2(a) shows a super pixel generated when k is 100 on the left side and a partial enlarged view on the right side;
fig. 2(b) shows a super pixel generated when k is 200 on the left side and a partial enlarged view on the right side;
fig. 2(c) shows a super pixel generated when k is 300 on the left side and a partial enlarged view on the right side;
fig. 2(d) shows a super pixel generated when k is 400 on the left side and a partial enlarged view on the right side;
fig. 2(e) shows a super pixel generated when k is 500 on the left side and a partial enlarged view on the right side;
FIG. 3(a) is a super pixel generated by the DBSCAN method on the left side and a partial enlarged view on the right side;
FIG. 3(b) shows the super-pixel generated by the SLIC method on the left and a partial enlarged view on the right;
FIG. 3(c) shows a super pixel generated by the present invention on the left and a partial enlarged view on the right;
FIG. 4(a) is a graph of segmentation accuracy;
FIG. 4(b) is a graph of edge recall;
FIG. 4(c) is a compactness chart;
fig. 4(d) is an under-segmentation error rate map.
The specific implementation mode is as follows:
the embodiment of the invention provides a super-pixel segmentation method based on self-adaptive seed point density clustering, which comprises the following steps.
Step 1, dividing an input image into a plurality of sub image blocks which are approximately same, and then searching the center of a maximum inscribed circle on the sub image blocks by using a region growing method so as to adaptively obtain the positions of seed points. The method comprises the following specific steps:
firstly, preprocessing an image, namely dividing the image into a plurality of sub-image blocks with the same size and the similar shape according to the similarity of color features and spatial distances among pixel points, and then calculating and searching the center coordinates of the maximum inscribed circle in four adjacent region regions on the sub-image blocks by using a region growing method to serve as the basis for self-adaptively acquiring the seed points.
In the process of seed point selection, the coordinates of the largest inscribed circle and the center of the circle are found, so that the seed point is automatically selected and the correct position of the seed point is selected. The basic requirements are as follows: first, a circle having the same area as the connection area is defined, and its radius R is calculated and obtained. The radius R is used as a human operator and the linking area is recorded. And judging whether the connection areas are completely connected. If it is completely connected, the radius of the circle is reduced by 1 unit and the above steps are repeated. Otherwise, the operator obtained in the previous step is taken as the maximum inscribed circle, and the final pixel point obtained after connection is taken as the center of the circle.
Step 2: and uniformly spreading seed points on the preprocessed image, and clustering outward by using the selected seed points as centers according to the improved density cluster until the seed points are fully distributed in the whole image, thereby finishing the initial superpixel segmentation.
The conversion of the rgb color space of the image to be processed into the lab color space requires the conversion from the rgb to the xyz space and from the xyz to the lab space. The specific process is as follows:
regularization treatment:
Figure BDA0003029588150000071
wherein r, g, b are original color components
Correcting r, g, b components:
Figure BDA0003029588150000072
r, g, b to x, y, z:
Figure BDA0003029588150000073
wherein k is a constant.
x, y, z to l, a, b:
Figure BDA0003029588150000074
wherein, l is the brightness from black to white, the scale is [0, 100], a represents the color range from green to red, the scale is [ -128, 127], b is the color range from yellow to blue, and the scale is [ -128, 127 ]. Compared with the rgb color space, the lab space has a larger color range, and is closer to the color that the human eye can accept, so the lab color space has better effect on the experimental calculation.
And then generating two sets which are respectively a mark set and a candidate set, locking the region to be segmented by using a maximum area measurement method, calculating the maximum inscribed circle area in the region, calculating the residual area of the region to be segmented according to the area, continuously finding the maximum inscribed circle in the residual region, and calculating the center coordinate of each circle, thereby adaptively selecting the position of the seed point and adding the position into the mark set.
Four sets of pixels are generated, seed, labeled, unlabeled, and candidate. Wherein the seed pixel set stores all the seed information, and the marker pixel set is reset when the seed point changes. The basic principle is as follows: first, a set of four neighboring pixel markers for all markers is found, then the combined distance between the center pixel of each marker pixel is calculated, and if the distance is less than a given threshold, it is deposited into the candidate set. Second, the existing label is updated with the candidate set and replaces it with the label that gives them the same as the seed. These two steps are repeated until a termination condition is satisfied.
And step 3: and further clustering the seed points on the sub-image blocks by using a space-limited k-medoids method through the automatically obtained seed point positions, and updating the clustering center point.
Firstly, randomly selecting k objects as representative points of initial k clusters, distributing the rest objects to the nearest clusters according to the distance between the rest objects and the seed point, and then repeatedly replacing the clustering center point with the non-clustering center point to improve the clustering quality. So that the current conditions help to control the size of the cluster. Inputting a database of n objects, and expecting k cluster clusters. And outputting k clusters, and minimizing the sum of deviations of all the objects from the center points of the clusters to which the objects belong. The method comprises the following steps: k objects are selected as initial cluster centers. (1) Each remaining object is assigned to the cluster represented by the nearest center point. (2) And randomly selecting a non-central point. (3) The total cost of forming a new cluster with non-centered replacement of the center point is calculated. (4) If the total cost is less than 0, the center points are replaced by non-center points to form a new set of k center points. Until the set is no longer changed. In addition, the space limitation is to make the algorithm converge as soon as possible, and the search range of the pixel point from the seed point is limited to be within the range of 2s × 2 s.
And 4, step 4: and traversing the seed points in sequence according to the idea of combining the top-down and bottom-up, and cleaning abnormal pixel points which cannot be processed in time.
And traversing the pixel points which are not visited in sequence according to the top-down idea, and comparing the result to obtain the minimum distance, wherein the pixel point belongs to the seed point corresponding to the distance. The method comprises the following specific steps: firstly, n pixel points which are uniformly distributed in the lab color space of the region to be segmented are extracted, and the boundary value is calculated according to the values of the pixel points. And then calculating the sum and difference of each pixel point and the boundary value. And if the sum is smaller than a preset threshold value or the difference is larger than the threshold value, determining the abnormal pixel point. The specific calculation method is as follows:
Figure BDA0003029588150000091
Figure BDA0003029588150000092
Figure BDA0003029588150000093
wherein d is1Representing the interval between a pixel point i and a clustering central point upsilon space; d2And representing the color distance between the pixel point i and the seed central point upsilon. α is a parameter for adjusting the color distance. τ denotes the ratio of the parameters of the results after and before segmentation.
In order to clearly and completely describe the technical solutions in the embodiments of the present invention, the following experiments of the present invention are further described in detail with reference to the drawings in the embodiments.
The method is experimentally verified, and obvious effects are achieved. The invention, the DBSCAN method and the SLIC method are compared in visual effect and quantification effect, and experimental results are all from codes and data provided by an original author and are realized on a BSD (Berkeley Segmentation dataset) natural image Segmentation data set. After multiple experiments, the results of finally determining various parameters are as follows: θ is 0.0003, δ is 1.28, and β is 1.9. Fig. 2(a) shows a super pixel generated when k is 100 on the left side and a partial enlarged view on the right side. Fig. 2(b) shows a super pixel generated when k is 200 on the left side and a partial enlarged view on the right side. Fig. 2(c) shows a super pixel generated when k is 300 on the left side and a partial enlarged view on the right side. Fig. 2(d) shows a super pixel generated when k is 400 on the left side and a partial enlarged view on the right side. Fig. 2(e) shows a super pixel generated when k is 500 on the left side and a partial enlarged view on the right side. It can be seen that when the k value of the number of superpixels is gradually increased, the shape of the obtained superpixels becomes more regular, and the image boundary segmentation is clearer.
(1) Visual effect comparison
As the number of the segmented superpixels is preset in 3 methods in the experiment, through experimental analysis, the number of the superpixels of each method is uniformly specified as 500 for carrying out the experiment. Fig. 3(a) shows the super pixel generated by the DBSCAN method on the left and a partially enlarged view on the right. Fig. 3(b) shows the super pixel generated by the SLIC method on the left side and a partial enlarged view on the right side. FIG. 3(c) shows a superpixel generated by the present invention on the left side and a partial enlarged view on the right side. It can be observed from the figure that the super-pixel segmentation case of the present invention is significantly more regular than the DBSCAN method. Because the traditional super-pixel generation method based on density clustering only considers the color problem between pixel points and seed points and does not consider that the artificial selection of the seed points is easily influenced by subjective factors, and the seed center points at wrong positions are selected, the super-pixel segmentation shape is irregular. The invention adds the condition of automatically selecting the seed points on the basis of the DBSCAN method, can well avoid the problem that the boundary position of the segmentation area selects the seed points, avoids the subjective factor selecting the seed points at the wrong position, and solves the problem that the super-pixel shape is irregular because no proper seed points are arranged near the boundary. Although both the DBSCAN method and the SLIC method can produce compact and uniform superpixels, they do not consider the problem of automatically acquiring the seed center point, and it is difficult to generate regular superpixel shapes.
(2) Quantization effect comparison
The invention adopts four quantitative standards to evaluate the performance of the algorithm, including Segmentation accuracy ASA (adaptive Segmentation accuracy), edge recall BR (boundary recall), compactness COM (compact) and under-Segmentation error rate USE (under-Segmentation error). All methods operate with a berkeley BSD image segmentation dataset.
Segmentation accuracy ASA: segmentation accuracy is a trade-off between whether a target in an image can be accurately identified. I.e. the best degree of similarity to the evaluation reference sample, i.e. the greater the value the better the segmentation accuracy. Fig. 4(a) is a division accuracy map. It is observed that the segmentation accuracy of the present invention is still much higher compared to DBSCAN.
Edge recall BR: edge recall is a criterion to weigh how well the image and superpixel boundaries fit. The proportion of pixels at the edges of the superpixel that fall within the pixel width region around the edges of the standard segmented image is calculated. Fig. 4(b) is an edge recall graph. The larger the value of the edge recall rate, the higher the degree of coincidence with the evaluation reference sample.
Compactness COM: fig. 4(c) is a compactness chart. It can be seen from the figure that the difference between compactness and SLIC of the method is small, but obviously, the method is much higher than the clustering method of the DBSCAN, which shows that the partitioned superpixel blocks of the present invention are more regular than the method of the DBSCAN.
Under-segmentation error rate USE: the under-segmentation error rate is the fault tolerance rate that exists with the standard segmentation result when the segmentation method is used for processing the segmented image. The smaller the under-segmentation error rate, the better the accurate segmentation of the image. Fig. 4(d) is an under-segmentation error rate map. It is observed that the under-segmentation error rate of the present invention is much lower than the DBSCAN method.
The invention firstly finds the maximum inscribed circle and the circle center coordinates from the region to be segmented, which can solve the problem that the segmented region selects the seed points at the boundary and avoid selecting the seed points of the boundary region due to subjective factors, thereby ensuring the self-adaptability of the seed points to select the center point. And then, further clustering the selected seed points through a k-medoids algorithm, and updating the clustering center point. And finally, cleaning the unaccessed abnormal pixel points. The problem of irregular super-pixel shapes caused by the fact that no proper seed points exist near the boundary is solved to a certain extent, and the segmentation precision of the super-pixels is further improved.
The foregoing is a detailed description of embodiments of the invention, taken in conjunction with the accompanying drawings, wherein the specific embodiments are merely provided to assist in understanding the method of the invention. For those skilled in the art, the invention can be modified and adapted within the scope of the embodiments and applications according to the spirit of the present invention, and therefore the present invention should not be construed as being limited thereto.

Claims (5)

1. The super-pixel segmentation method based on self-adaptive seed point density clustering is characterized by comprising the following steps of:
step 1: dividing an input image into a plurality of sub image blocks which are approximately the same, and then searching the center coordinate of the maximum inscribed circle on the sub image blocks by using a region growing method so as to self-adaptively obtain the positions of seed points;
step 2: uniformly spreading seed points on the preprocessed image, and sequentially clustering the selected seed points outwards as a center according to the density clustering of the self-adaptive seed points until the seed points are fully distributed in the whole image, thereby completing the initial superpixel segmentation;
and step 3: clustering the seed points on all the obtained sub-image blocks by combining a space-limited k-medoids method, further updating the seed points, and solving the influence of noise on the method;
and 4, step 4: and traversing the seed points in sequence according to the idea of combining the top-down and bottom-up, and cleaning abnormal pixel points which cannot be processed in time.
2. The adaptive seed point density clustering-based superpixel segmentation method according to claim 1, wherein in the step 1, the image is divided, and the coordinates of the center of the largest inscribed circle are calculated and found in the sub-image block by using a region growing process, and the specific steps are as follows:
step 1-1, acquiring all pixel information of a region to be segmented as input, and performing preprocessing operation on an image;
step 1-2, filtering and screening according to the gray value range between the pixel points to be segmented, and dividing the image into a plurality of sub image blocks with similar sizes and similar colors, wherein the color range of the sub image block is within the gray value range of the area to be segmented;
step 1-3, calculating and searching the center coordinates of the maximum inscribed circle on the four neighborhoods of the sub-image block by using a region growing method, determining the positions of center coordinates (X, Y) by using a standard deviation method, automatically selecting seed points after determining the coordinates, wherein the specific calculation method of the center coordinates is as follows:
Figure FDA0003029588140000021
Figure FDA0003029588140000022
wherein the content of the first and second substances,
Figure FDA0003029588140000023
in order to count the number of seed points in the horizontal search,
Figure FDA0003029588140000024
the number of the seed points in the longitudinal search is shown, and n is the number of the pixel points.
3. The adaptive seed point density clustering-based superpixel segmentation method according to claim 1, wherein in step 2, the seed points are sequentially clustered outwards according to adaptive seed point density clustering with the seed points as clustering centers, and the specific steps are as follows:
step 2-1, converting the rgb color space of the image to be processed into a lab color space;
step 2-2, respectively defining the two sets as a mark set and a candidate set, then locking an area to be segmented, adaptively selecting a seed point position according to the calculated circle center coordinate, adding the seed point position into the mark set, and if the mark set exists at the point, turning to step 2, and restarting the search of the next target area;
step 2-3, four pixel sets are counted, wherein the four pixel sets comprise a seed pixel set, a marked pixel set, an unmarked pixel set and a candidate pixel set, the seed pixel set stores all seed information, and the marked pixel set is reset when the seed points change;
step 2-4, measuring the effect after clustering by adopting a standard function of absolute deviation, wherein the specific calculation method comprises the following steps:
Figure FDA0003029588140000025
wherein: p is cluster deltajObject of δjAs cluster center, p and δjIs multi-dimensional.
4. The adaptive seed point density clustering-based superpixel segmentation method according to claim 1, wherein in step 3, the sub-image blocks are further clustered, and the seed point center point is updated, specifically comprising the steps of:
step 3-1, combining a space-limited k-medoids method to ensure that the number of pixels in the clustered cluster is greater than a given threshold value s/n, wherein s represents the size of an input image, and n represents the number of super-pixels needed by a user; the space limitation is to limit the search range of the pixel points starting from the seed points to the range of 2s multiplied by 2s in order to make the algorithm converge as soon as possible;
step 3-2 changes the marker set to an empty set, which indicates that the neighboring pixels of the previous marker set will be located in the boundary region, and the current condition helps the algorithm to stop at the boundary;
3-3, automatically selecting new seed points from the unmarked pixel set according to a clustering algorithm, and repeatedly executing the process until all the seed points are marked, otherwise, returning to the step 3-1 and marking again;
step 3-4, generating initial superpixels by the obtained clusters of different labels, wherein the specific calculation method comprises the following steps:
Figure FDA0003029588140000031
Figure FDA0003029588140000032
Figure FDA0003029588140000033
Figure FDA0003029588140000034
wherein lq,aq,bqRepresenting the color characteristics of the pixel point q, lu,au,buThe color characteristics of the marking point mu are indicated, lt,at,btRepresenting the color characteristic of the seed point t, oq,cqRepresenting the spatial characteristics of the pixel point q, ot,utThe spatial characteristics of the seed point t are shown, theta is a parameter used for adjusting the size of an inscribed circle of the region to be segmented, delta is a boundary limiting parameter of the seed point used for adjusting the effective position of the seed point q on the boundary,
Figure FDA0003029588140000035
representing the color difference between the mark point u and the pixel point q,
Figure FDA0003029588140000036
representing the color distance, d, between the seed point t and the pixel point qsThe distance between the seed point t and the pixel point q in space, and the distance between the marking point and the seed point to the pixel point.
5. The adaptive seed point density clustering-based superpixel segmentation method according to claim 1, wherein in said step 4, abnormal pixel points that cannot be processed in time are cleaned, and the specific process is as follows:
step 4-1, sequentially traversing the pixel points which are not visited according to the idea of combining top-down and bottom-up;
step 4-2, calculating and finding the minimum difference between the minimum difference and all the seed points, and then the seed points near the pixel point accord with the interval condition, namely the cleaning is finished, and the specific calculation method is as follows:
Figure FDA0003029588140000041
Figure FDA0003029588140000042
Figure FDA0003029588140000043
wherein d is1Representing the interval between a pixel point i and a clustering central point upsilon space, d2And representing the color distance between the pixel point i and the seed central point upsilon, wherein alpha is a parameter for adjusting the color distance, and tau represents the parameter ratio of the result after segmentation and the result before segmentation.
CN202110426049.3A 2021-04-20 2021-04-20 Super-pixel segmentation method based on self-adaptive seed point density clustering Pending CN113052859A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110426049.3A CN113052859A (en) 2021-04-20 2021-04-20 Super-pixel segmentation method based on self-adaptive seed point density clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110426049.3A CN113052859A (en) 2021-04-20 2021-04-20 Super-pixel segmentation method based on self-adaptive seed point density clustering

Publications (1)

Publication Number Publication Date
CN113052859A true CN113052859A (en) 2021-06-29

Family

ID=76520713

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110426049.3A Pending CN113052859A (en) 2021-04-20 2021-04-20 Super-pixel segmentation method based on self-adaptive seed point density clustering

Country Status (1)

Country Link
CN (1) CN113052859A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113553966A (en) * 2021-07-28 2021-10-26 中国科学院微小卫星创新研究院 Method for extracting effective starry sky area of single star map
CN115457050A (en) * 2022-08-29 2022-12-09 武汉纺织大学 Super-pixel-based ancient mural image optimization segmentation method and system
CN115601630A (en) * 2022-12-15 2023-01-13 山东天意装配式建筑装备研究院有限公司（Cn） Stain recognition method for automatic wallboard mold cleaning machine
CN116761023A (en) * 2023-07-04 2023-09-15 唯科终端技术(东莞)有限公司 Video transmission method between different devices
CN118053081A (en) * 2024-04-15 2024-05-17 西安城市发展资源信息有限公司 Urban and rural planning data updating method based on image processing

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040130546A1 (en) * 2003-01-06 2004-07-08 Porikli Fatih M. Region growing with adaptive thresholds and distance function parameters
US20090238457A1 (en) * 2008-03-21 2009-09-24 General Electric Company Methods and systems for automated segmentation of dense cell populations
US20120201460A1 (en) * 2011-02-08 2012-08-09 Kyungpook National University Industry-Academic Cooperation Foundation Image segmentation devices and methods based on sequential frame image of static scene
CN106919950A (en) * 2017-01-22 2017-07-04 山东大学 Probability density weights the brain MR image segmentation of geodesic distance
CN109389601A (en) * 2018-10-19 2019-02-26 山东大学 Color image superpixel segmentation method based on similitude between pixel
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
CN109598726A (en) * 2018-10-26 2019-04-09 哈尔滨理工大学 A kind of adapting to image target area dividing method based on SLIC
CN111932575A (en) * 2020-06-24 2020-11-13 山东师范大学 Image segmentation method and system based on fuzzy C-means and probability label fusion
CN112561919A (en) * 2019-09-10 2021-03-26 中科星图股份有限公司 Image segmentation method, device and computer readable storage medium

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040130546A1 (en) * 2003-01-06 2004-07-08 Porikli Fatih M. Region growing with adaptive thresholds and distance function parameters
US20090238457A1 (en) * 2008-03-21 2009-09-24 General Electric Company Methods and systems for automated segmentation of dense cell populations
US20120201460A1 (en) * 2011-02-08 2012-08-09 Kyungpook National University Industry-Academic Cooperation Foundation Image segmentation devices and methods based on sequential frame image of static scene
CN106919950A (en) * 2017-01-22 2017-07-04 山东大学 Probability density weights the brain MR image segmentation of geodesic distance
CN109389601A (en) * 2018-10-19 2019-02-26 山东大学 Color image superpixel segmentation method based on similitude between pixel
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
CN109598726A (en) * 2018-10-26 2019-04-09 哈尔滨理工大学 A kind of adapting to image target area dividing method based on SLIC
CN112561919A (en) * 2019-09-10 2021-03-26 中科星图股份有限公司 Image segmentation method, device and computer readable storage medium
CN111932575A (en) * 2020-06-24 2020-11-13 山东师范大学 Image segmentation method and system based on fuzzy C-means and probability label fusion

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
CHENG LI等: "NICE:Superpixel Segmentation Using Non-Iterative Clustering with Efficiency", 《APPLIED SCIENCES》 *
JIANBING SHEN等: "Real-time superpixel segmentation by DBSCAN clustering algorithm", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *
许晓丽等: "一种基于超像素的快速聚类图像分割算法", 《电脑知识与技术》 *
韩剑辉等: "空间约束密度聚类的超像素分割算法", 《哈尔滨理工大学学报》 *
魏津瑜等: "基于改进算法的自动种子区域生长图像分割", 《2013年中国智能自动化学术会议论文集（第五分册）》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113553966A (en) * 2021-07-28 2021-10-26 中国科学院微小卫星创新研究院 Method for extracting effective starry sky area of single star map
CN113553966B (en) * 2021-07-28 2024-03-26 中国科学院微小卫星创新研究院 Method for extracting effective starry sky area of single star map
CN115457050A (en) * 2022-08-29 2022-12-09 武汉纺织大学 Super-pixel-based ancient mural image optimization segmentation method and system
CN115601630A (en) * 2022-12-15 2023-01-13 山东天意装配式建筑装备研究院有限公司（Cn） Stain recognition method for automatic wallboard mold cleaning machine
CN116761023A (en) * 2023-07-04 2023-09-15 唯科终端技术(东莞)有限公司 Video transmission method between different devices
CN118053081A (en) * 2024-04-15 2024-05-17 西安城市发展资源信息有限公司 Urban and rural planning data updating method based on image processing
CN118053081B (en) * 2024-04-15 2024-06-11 西安城市发展资源信息有限公司 Urban and rural planning data updating method based on image processing

Similar Documents

Publication Publication Date Title
CN113052859A (en) Super-pixel segmentation method based on self-adaptive seed point density clustering
CN109658381B (en) Method for detecting copper surface defects of flexible IC packaging substrate based on super-pixels
CN108537239B (en) Method for detecting image saliency target
CN109636784B (en) Image saliency target detection method based on maximum neighborhood and super-pixel segmentation
CN108629783B (en) Image segmentation method, system and medium based on image feature density peak search
CN110866896B (en) Image saliency target detection method based on k-means and level set super-pixel segmentation
Rathore et al. Colour based image segmentation using L* a* b* colour space based on genetic algorithm
CN108230340B (en) Method for determining SLIC superpixel extraction weight based on MMTD and superpixel extraction method
CN115131375B (en) Automatic ore cutting method
CN114913138A (en) Method and system for detecting defects of pad printing machine product based on artificial intelligence
EP1459258B1 (en) Segmentation of images using the watershed method
CN111339924A (en) Polarized SAR image classification method based on superpixel and full convolution network
CN114745532A (en) White balance processing method and device for mixed color temperature scene, storage medium and terminal
CN109741358B (en) Superpixel segmentation method based on adaptive hypergraph learning
CN109272539A (en) The decomposition method of image texture and structure based on guidance figure Total Variation
CN110634142B (en) Complex vehicle road image boundary optimization method
CN115457551A (en) Leaf damage identification method suitable for small sample condition
CN105913451B (en) A kind of natural image superpixel segmentation method based on graph model
CN113436206B (en) Pathological tissue section scanning area positioning method based on cluster segmentation
CN109191482B (en) Image merging and segmenting method based on regional adaptive spectral angle threshold
CN114862902A (en) Illumination self-adaptive ORB feature extraction and matching method based on quadtree
CN111539966A (en) Colorimetric sensor array image segmentation method based on fuzzy c-means clustering
CN108537798B (en) Rapid super-pixel segmentation method
CN116543414A (en) Tongue color classification and tongue redness and purple quantification method based on multi-model fusion
CN113223098B (en) Preprocessing optimization method for image color classification

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210629