CN112037230A - Forest region image segmentation algorithm based on super-pixel and super-metric contour map - Google Patents

Forest region image segmentation algorithm based on super-pixel and super-metric contour map Download PDF

Info

Publication number
CN112037230A
CN112037230A CN201910481471.1A CN201910481471A CN112037230A CN 112037230 A CN112037230 A CN 112037230A CN 201910481471 A CN201910481471 A CN 201910481471A CN 112037230 A CN112037230 A CN 112037230A
Authority
CN
China
Prior art keywords
segmentation
pixel
super
superpixel
map
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
CN201910481471.1A
Other languages
Chinese (zh)
Other versions
CN112037230B (en
Inventor
刘文萍
宗世祥
骆有庆
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.)
Beijing Forestry University
Original Assignee
Beijing Forestry University
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 Beijing Forestry University filed Critical Beijing Forestry University
Priority to CN201910481471.1A priority Critical patent/CN112037230B/en
Publication of CN112037230A publication Critical patent/CN112037230A/en
Application granted granted Critical
Publication of CN112037230B publication Critical patent/CN112037230B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics

Landscapes

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

Abstract

The invention discloses a forest region image segmentation algorithm based on superpixels and a hypermetrological contour map, which comprises the steps of firstly segmenting an image into superpixels by utilizing linear spectral clustering, and calculating dissimilarity between adjacent superpixel regions; then combining the areas with higher similarity from bottom to top, and updating the edge weight in the hypermetric profile map; and finally, outputting the image segmented by using the optimal weight threshold. A smaller weight threshold may result in over-segmentation, with the segmentation result retaining only the more significant edges as the threshold increases. The invention can independently set the segmentation threshold T, and can set the segmentation thresholds T with different sizes according to actual requirements so as to obtain better segmentation results. Compared with other existing image segmentation algorithms, the method has the advantages of remarkably reduced operation complexity, higher algorithm speed, good segmentation effect, small dependence on initial parameters and the like, is very suitable for segmenting high-resolution unmanned aerial vehicle forest area aerial images, and has high popularization and application values.

Description

Forest region image segmentation algorithm based on super-pixel and super-metric contour map
Technical Field
The invention relates to the field of image segmentation, in particular to a forest region image segmentation algorithm based on super pixels and a super-metric contour map.
Background
As a low-altitude remote sensing tool, the unmanned aerial vehicle has the advantages of low flying cost, simplicity and convenience in operation, flexibility, free shooting time and the like, achieves remarkable results in the fields of geological exploration, environmental survey, damage detection and the like, and has immeasurable development prospect. In order to intelligently analyze image information captured by an unmanned aerial vehicle, the research on a rapid and high-quality image segmentation algorithm is always a hot problem in the image field. The image segmentation is to divide an image into a plurality of non-overlapping regions, so that the same region shows similarity and adjacent regions are obviously different, and is a key step in image analysis, pattern recognition and computer vision.
Because the unmanned aerial vehicle aerial image has high resolution, complex scene and various application ranges, a very mature segmentation algorithm can not be applied to various types of unmanned aerial vehicle images at present.
The superpixel is a sub-region which has consistency in an image and can keep local structural characteristics of the image, and the essence of a Linear Spectral Clustering (LSC) superpixel generation algorithm is to map data in an original space to a high-dimensional space for clustering. The algorithm firstly designs a kernel function
Figure BDA0002083977430000011
And mapping pixel features in the image pixel set V to a high-dimensional feature space, and then realizing image segmentation by using a weighted K-means clustering algorithm. The characteristic representation of a pixel point p in the image is (l)p,αp,βp,xp,yp)，l,α,βRespectively representing color components L, a and b of p in a CIELAB color space, L being color brightness, an a-channel representing a position between red and green, and a b-channel representing a position between yellow and blue; x and y are the vertical and horizontal coordinates of p in the image plane. Kernel function mapping from low-dimensional pixel space to high-dimensional feature space
Figure BDA0002083977430000021
Is defined as:
Figure RE-GDA0002154507190000021
wherein the content of the first and second substances,
Figure RE-GDA0002154507190000022
Figure RE-GDA0002154507190000023
p, q are pixel points, parameter CcAnd CsWeights for controlling colour and position information, respectively, Cs＝r×Cc， r＝0.075。
The hypermetrological contour map is derived from Arbelaez and the like and provides an image level segmentation method based on edge detection, namely a global edge probability-direction watershed transformation-hypermetrological contour map (oriented watershed transform and ultrasound contour map, gPb-OWT-UCM).
The bottom-up image level segmentation is constructed by a region merging algorithm. The initial segmentation is taken as a graph, the initial segmentation areas are nodes of the graph, and adjacent areas R are connected1And R2The edge C of (1) is the edge of the graph, the edge set is C ═ C }, and the region R is1And R2The dissimilarity between them is taken as the weight w (c) of the edge, and the set of weights is w (c) ═ w (c). The hierarchical segmentation is a process of sequencing edges through dissimilarity between regions and iteratively merging the most similar regions, and the specific algorithm steps are as follows:
input, edge weight set W (C).
And outputting the segmentation result of each iteration.
Step1. select the edge c with the smallest weight in W (C)*，
c*＝arg minc∈CW(c)。
Step2. if R1，R2Is a quilt c*Two divided regions, thenCombining them into R ═ R1∪R2。
Step3. delete edge C from edge set C*。
Step4, if the C is an empty set, ending the iteration; otherwise, the edge set C and the weight set W (C) of the edge are updated, and the process goes to step1.
The contour generated by the bottom layer of the hierarchical segmentation can keep the boundary with weak strength and can cause over-segmentation at the same time; the contour of the upper layer is only sensitive to the boundary with stronger strength, and partial necessary edge information may be lost. Considering that the weight of the currently deleted edge is minimum in the hierarchical segmentation process, the weight values of all the remaining edges are necessarily greater than the weight values of the previously deleted edges, so that an over-measure contour map with an index hierarchical structure can be constructed, the hierarchical segmentation of the image is understood as a set of different segmentation results under multiple scales, and a suitable contour is obtained as the optimal segmentation result by selecting the scale.
Since the algorithm gPb-OWT-UCM combines multi-dimensional features (3-dimensional color features + 17-dimensional texture features) in different directions of multiple scales when computing the edge strength by using gPb, obtaining an accurate segmentation result inevitably costs huge computation complexity, and the algorithm is obviously inapplicable to unmanned aerial vehicle aerial images with high resolution.
Disclosure of Invention
The invention provides a forest region image segmentation algorithm based on superpixels and a hypermetric contour map, which is used for segmenting an image obtained by unmanned aerial vehicle aerial photography.
In order to achieve the purpose, the invention provides a forest region image segmentation algorithm based on a superpixel and a hypermetrological contour map, which comprises the following steps:
s1: performing superpixel segmentation on an original image to generate a superpixel image comprising a plurality of superpixel regions, wherein the size of the original image is M multiplied by N and is an RGB image;
s2: converting the superpixel graph from an RGB color space to an HSV color space, and averagely dividing hue component intervals, saturation component intervals and lightness component intervals of the converted superpixel graph in the HSV color space into n subintervals;
s3: respectively counting the number of pixels in each sub-interval of each super-pixel region in the hue component interval, the saturation component interval and the brightness component interval and normalizing the number of pixels to obtain a normalized histogram
Figure BDA0002083977430000041
Where m is 3 × n, n is an integer greater than 1, i represents the ith super pixel region,
Figure BDA0002083977430000042
respectively representing the normalized values of the number of pixels of the ith super pixel region in the 1 st to nth sections of the tone component section,
Figure BDA0002083977430000043
respectively representing the pixel number normalized values of the ith super-pixel region in 1 st to nth intervals in the saturation component intervals,
Figure BDA0002083977430000044
respectively representing pixel number normalization values of the ith super pixel region in 1 st to n th sections in the lightness component section;
s4: calculating the dissimilarity D (R) between all adjacent two super-pixel regions according to the following formulai,Rj)：
Figure BDA0002083977430000045
Wherein R isi、RjRepresenting two adjacent super-pixel regions;
s5: according to the calculation result of the step S4, obtaining an dissimilarity sequence S according to the dissimilarity from small to large, wherein each element in the dissimilarity sequence S is the dissimilarity between two adjacent super-pixel regions;
s6: initializing to generate a matrix U, wherein the U is M multiplied by N and is a zero matrix;
s7: corresponding all pixel points between two adjacent super pixel areas in the super pixel image to elements in the matrix U one by one, and respectively assigning the dissimilarity degrees obtained by the calculation in the step S4 to the corresponding elements in the matrix U;
s8: taking out the element U with the minimum value from the dissimilarity sequence S, merging two adjacent superpixel regions corresponding to the element U, wherein the merging rule is that the element in a matrix U corresponding to each pixel point between the two adjacent superpixel regions corresponding to the element U is updated to the element U, and simultaneously, the histogram of the merged superpixel region R is calculated as
Figure BDA0002083977430000046
Wherein a () represents a total number of pixels in the super pixel region;
s9: calculating the dissimilarity degree between the combined super-pixel region R in the step S8 and all the super-pixel regions adjacent to the combined super-pixel region R, and updating a dissimilarity degree sequence S according to the dissimilarity degree sequence;
s10: repeating the steps S8-S9 until the number of elements in the dissimilarity sequence S is zero;
s11: normalizing all elements in the matrix U to be 0-1;
s12: selecting a segmentation threshold value T, removing the adjacent region smaller than the segmentation threshold value T from the superpixel map obtained in the step S1, and reserving the adjacent region larger than the segmentation threshold value T to obtain a segmentation map comprising a plurality of sub-regions;
s13: calculating the average color of each sub-region in the segmentation map obtained in step S12, and filling each average color in the corresponding sub-region to obtain a final segmentation result map.
In one embodiment of the present invention, in step S1, the number of super pixel regions in the super pixel map is initialized to 100.
In an embodiment of the present invention, in step S1, a superpixel map is generated by using a linear spectral clustering method.
In an embodiment of the invention, the value range of the hue component interval is between 0 and 255, the value range of the saturation component interval is between 0 and 255, and the value range of the brightness component interval is between 0 and 255.
In one embodiment of the present invention, n is 25.
In an embodiment of the present invention, the normalization function used in step S11 is a sigmoid function.
The forest region image segmentation algorithm based on the superpixel and the hypermetric profile map can autonomously set the segmentation threshold T, and can set the segmentation thresholds T with different sizes according to actual needs so as to obtain a better segmentation result. Compared with other existing image segmentation algorithms, the method has the advantages of remarkably reduced operation complexity, higher algorithm speed, good segmentation effect, small dependence on initial parameters and the like, is very suitable for segmenting high-resolution unmanned aerial vehicle forest area aerial images, and has high popularization and application values.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is an original image according to an embodiment of the present invention;
FIG. 2 is a diagram of a super pixel in accordance with one embodiment of the present invention;
FIG. 3 is an image corresponding to the normalized matrix U according to an embodiment of the present invention;
FIG. 4a is a graph of the segmentation result obtained when the segmentation threshold T is 0.5;
FIG. 4b is a graph of the segmentation result obtained when the segmentation threshold T is 0.93;
FIG. 4c is a graph of the segmentation result obtained when the segmentation threshold T is 0.99;
FIGS. 5 a-5 d are original images according to a second embodiment of the present invention;
FIGS. 6a to 6d are diagrams illustrating the result of manual segmentation according to the second embodiment of the present invention;
FIGS. 7 a-7 d are graphs showing the result of ISODATA partitioning according to the second embodiment of the present invention;
FIGS. 8a to 8d are graphs showing the result of FCM segmentation according to the second embodiment of the present invention;
FIGS. 9a to 9d are diagrams illustrating the result of gPb-OWT-UCM segmentation according to the second embodiment of the present invention;
FIGS. 10a to 10d are graphs showing the result of the LSC-UCM segmentation according to the second embodiment of the present invention;
FIG. 11 is a schematic diagram of error rate comparison of algorithms according to a second embodiment of the present invention;
FIG. 12 is a diagram illustrating average cross-comparison of algorithms according to a second embodiment of the present invention;
FIG. 13 is a schematic diagram of the gray scale contrast ratio of each algorithm according to the second embodiment of the present invention;
FIG. 14 is a schematic diagram of a comparison of the running times of the algorithms of the second embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without any inventive step, are within the scope of the present invention.
The invention provides a forest region image segmentation algorithm (LSC-UCM) based on superpixels and a hypermetrological contour map, which comprises the steps of firstly segmenting an image into superpixels by utilizing linear spectral clustering, and calculating dissimilarity between adjacent superpixel regions; then combining the areas with higher similarity from bottom to top, and updating the edge weight in the hypermetrological contour map; and finally, outputting the image segmented by using the optimal weight threshold. A smaller weight threshold will result in over-segmentation, with the segmentation result retaining only the more significant edges as the threshold increases.
The superpixel segmentation algorithm can quickly generate a group of initial closed regions, and the superpixel regions have more abundant information than single pixels, so that the invention provides the method for segmenting the unmanned aerial vehicle image by combining linear spectral clustering superpixels and a hypermetrological contour map (LSC-UCM).
The invention provides a forest region image segmentation algorithm based on a superpixel and a hypermetrological contour map, which comprises the following steps of:
s1: performing superpixel segmentation on an original image to generate a superpixel image comprising a plurality of superpixel regions, wherein the size of the original image is M multiplied by N and is an RGB image;
referring to fig. 1 and 2, fig. 1 is an original image according to an embodiment of the invention, and fig. 2 is a super-pixel diagram according to an embodiment of the invention.
In this embodiment, the super-pixel map is generated by using a linear spectral clustering method, and the number of super-pixel regions in the super-pixel map is initialized to 100. If the number of the initial super pixel areas is set to be too small, the weak edge is difficult to segment; if the initial number of the super pixel regions is set to be too large, a plurality of too small super pixel regions can appear, the complexity of region combination is increased, and the number of the super pixel regions is initialized to 100, so that the requirements of clear division and the complexity of combination are considered. It should be noted that, the meaning of initializing the number of the super pixel regions to 100 here is to make the number of the super pixel regions substantially 100, and the number of the super pixel regions in the actually generated super pixel map may be greater than or less than 100, which is an obvious technical means for those skilled in the art and is not described herein again.
S2: converting the superpixel graph from an RGB color space to an HSV color space, and averagely dividing hue component intervals, saturation component intervals and lightness component intervals of the converted superpixel graph in the HSV color space into n subintervals;
because the HSV color space is very close to the human visual system and is suitable for processing and analyzing the color perception characteristic, the color histogram in the HSV space is selected as the color characteristic of the super-pixel.
In this embodiment, the values of the hue component H, the saturation component S, and the value component V in the HSV color space are calculated from the values of the R component, the G component, and the B component in the RGB color space, and specifically, each pixel point in the superpixel map is subjected to the following processing: firstly, normalizing the numerical values of the R component, the G component and the B component (theoretically, the numerical ranges of 3 components are all 0-255) of each pixel point in the RGB color space to be 0-1 (namely, dividing the numerical values of the R component, the G component and the B component by 255), and then calculating the numerical values of the hue component H, the saturation component S and the lightness component V of each pixel point in the HSV color space according to the following formula:
Figure BDA0002083977430000081
Figure BDA0002083977430000082
V＝255×max
the above three formulas are respectively calculated for each pixel, that is, the above 3 formulas are executed for each pixel, max is the maximum value among the R component, G component and B component of the pixel, min is the minimum value among the R component, G component and B component of the pixel, and the value ranges of the hue component H, the saturation component S and the lightness component V obtained by calculation are all between 0 and 255 (theoretical ranges). The range of the hue component interval is a range defined by the minimum value and the maximum value of the hue component H obtained by calculation, the range of the saturation component interval is a range defined by the minimum value and the maximum value of the saturation component S obtained by calculation, the range of the brightness component interval is a range defined by the minimum value and the maximum value of the brightness component V obtained by calculation, and on the basis, the hue component interval, the saturation component interval and the brightness component interval are evenly divided into n subintervals. S3: respectively counting the number of pixels in each sub-interval of each super-pixel region (converted according to step S2) in the hue component interval, the saturation component interval and the brightness component interval and normalizing the number of pixels to obtain a normalized histogram
Figure BDA0002083977430000091
Where m is 3 × n, n is an integer greater than 1, i represents the ith super pixel region,
Figure BDA0002083977430000092
respectively representing the normalized values of the number of pixels of the ith super pixel region in the 1 st to nth sections of the tone component section,
Figure BDA0002083977430000093
respectively representing the pixel number normalized values of the ith super-pixel region in 1 st to nth intervals in the saturation component intervals,
Figure BDA0002083977430000094
respectively representing pixel number normalization values of the ith super pixel region in 1 st to n th sections in the lightness component section;
in the present embodiment, n is 25, so that m is 3 × n is 75, in other embodiments, the value of n may be adjusted according to actual conditions, and the adjustment principle is to be able to extract finer features of the hue component interval, the saturation component interval, and the lightness component interval, and at the same time, reduce the feature dimension, and increase the calculation speed.
S4: calculating the dissimilarity D (R) between all adjacent two super-pixel regions according to the following formulai,Rj)：
Figure BDA0002083977430000095
Wherein R isi、RjRepresenting two adjacent super-pixel regions;
s5: according to the calculation result of the step S4, obtaining an dissimilarity sequence S according to the dissimilarity from small to large, wherein each element in the dissimilarity sequence S is the dissimilarity between two adjacent super-pixel regions;
s6: initializing to generate a matrix U, wherein the U is M multiplied by N and is a zero matrix;
s7: corresponding all pixel points between two adjacent super pixel areas in the super pixel image to elements in the matrix U one by one, and respectively assigning the dissimilarity degrees obtained by the calculation in the step S4 to the corresponding elements in the matrix U;
s8: taking out the element U with the minimum value from the dissimilarity sequence S, merging two adjacent superpixel regions corresponding to the element U, wherein the merging rule is that the element in a matrix U corresponding to each pixel point between the two adjacent superpixel regions corresponding to the element U is updated to the element U, and simultaneously, the histogram of the merged superpixel region R is calculated as
Figure BDA0002083977430000101
Wherein a () represents a total number of pixels in the super pixel region;
the value in the dissimilarity sequence S represents a threshold value when two adjacent superpixel regions are merged, that is, the weight value of the edge to be removed is smaller, which means that the two superpixel regions are more similar and are merged earlier, and the larger the weight value is, which means that the edge between the two superpixel regions disappears later.
It can be found that, when step S8 is executed, two adjacent superpixel regions corresponding to the element U are original regions that are never merged in the superpixel map, only the histogram of the merged superpixel region R is calculated in this step, and it is not necessary to update the matrix U, so that the update of the adjacent region is already performed on the matrix U in step S7.
S9: calculating the dissimilarity degree between the combined super-pixel region R in the step S8 and all the super-pixel regions adjacent to the combined super-pixel region R, and updating a dissimilarity degree sequence S according to the dissimilarity degree sequence;
s10: repeating the steps S8-S9 until the number of elements in the dissimilarity sequence S is zero;
s11: normalizing all elements in the matrix U to be 0-1;
in this embodiment, the normalization function used here is a sigmoid function, and in other embodiments, other normalization functions may be selected. Fig. 3 shows an image corresponding to the normalized matrix U according to an embodiment of the present invention, where the intensity of the edge is from light to dark in fig. 3, which corresponds to the intensity of the edge from strong to weak, indicating that the HSV spatial histogram feature can be used to measure the dissimilarity between the regions well.
S12: selecting a segmentation threshold value T, removing the adjacent region smaller than the segmentation threshold value T from the superpixel map obtained in the step S1, and reserving the adjacent region larger than the segmentation threshold value T to obtain a segmentation map comprising a plurality of sub-regions;
s13: calculating the average color of each sub-region in the segmentation map obtained in step S12, and filling each average color in the corresponding sub-region to obtain a final segmentation result map.
FIGS. 4 a-4 c are graphs of segmentation results obtained with segmentation thresholds T of 0.5, 0.93, and 0.99, respectively, FIG. 4a being more sensitive to weak edges, resulting in over-segmentation; FIG. 4b retains the more prominent edge; fig. 4c shows that the segmentation effect is best when the segmentation threshold T is 0.99.
In step S1 of this embodiment, a linear spectral clustering algorithm is used, so that a group of closed and compact superpixels can be generated while maintaining the edge information of the image, and each pixel point in the output label matrix is labeled as the class label to which it belongs. In the algorithm for generating the superpixel by linear spectral clustering, an image I is input, and a mark matrix L is output, and the method specifically comprises the following steps:
step1, mapping the feature of each pixel p in the image I into a high-dimensional feature vector by using the formula (1)
Figure BDA0002083977430000111
Step2. at horizontal intervals vxAnd a vertical spacing vyThe images are uniformly initialized into K classes.
Step3, setting iteration times T and initializing class center mkIs the mean value of the features of the pixels within the class, K is 1, 2, …, K.
Step4. calculate neighborhood τ v of class centerx×τvy(τ > 0.5) updating Euclidean distance between each pixel and class center to which pixel p belongspNumbering the class center closest to p:
Figure BDA0002083977430000112
step5. update class center mkWeighted mean within class:
Figure BDA0002083977430000121
in the formula pikRepresenting a set of pixels of the kth class.
Step6. up to mkConverge or iterate T times, otherwise go to Step4.
Step7. merge the too small region into its neighborhood, the class labels of all pixels constituting the matrix L. And finally pixels assigned to the same class form a super-pixel.
The invention is further illustrated below in a specific segmentation example (second embodiment):
in this example, the used experimental data are forest region images acquired by unmanned aerial vehicle aerial photography, and the image resolution is 4000 × 3000, 4608 × 2592 and the like. The experimental images were taken from the original aerial photographs for a total of 4 at 512 x 512 pixels. The CPU of the experimental machine is 3.00GHz, the internal memory is 16GB, the operating system is Linux, and the development tool is Matlab (R2016b x64) and C + + mixed programming.
Fig. 5a to 5d are original images of a second embodiment of the present invention, and fig. 5a to 5d are used to compare image segmentation of ISODATA, FCM, gPb-OWT-UCM and the LSC-UCM algorithm proposed by the present invention, fig. 6a to 6d are manual segmentation result graphs of the second embodiment of the present invention, fig. 7a to 7d are ISODATA segmentation result graphs of the second embodiment of the present invention, fig. 8a to 8d are FCM segmentation result graphs of the second embodiment of the present invention, fig. 9a to 9d are gPb-OWT-UCM segmentation result graphs of the second embodiment of the present invention, and fig. 10a to 10d are LSC-UCM segmentation result graphs of the second embodiment of the present invention. The number of categories in segmentation maps of ISODATA and FCM is manually set, the segmentation results of gPb-OWT-UCM and LSC-UCM are also related to selection of UCM threshold values, the optimal segmentation results of algorithms are selected in the embodiment, the optimal threshold values selected by gPb-OWT-UCM and LSC-UCM are different aiming at different images, and the optimal threshold values can be determined to be in a certain interval. The optimal threshold value is selected by adopting a coarse-to-fine search mode, firstly generating all segmentation maps of the threshold value in the UCM within a certain interval, and then finding the optimal segmentation result in the segmentation results. With the increasing threshold value, the weak edge gradually disappears, and the edge with higher significance is reserved. Based on a large number of experiments, the optimal threshold range of the LSC-UCM is 0.5-1, and the optimal threshold range of the gPb-OWT-UCM is 0.1-1. The LSC-UCM sets the number of the superpixels to be 100, and the gPb-OWT-UCM automatically generates about 3000 initial areas, so that the calculation complexity is higher. The colors in the LSC-UCM segmentation graph are represented by the color average value of pixels in the class, different colors represent different classes, and for the convenience of comparison, different classes in the manual segmentation graph and other algorithm segmentation graphs are also represented by corresponding colors in the LSC-UCM segmentation graph.
According to the segmentation results, the number of clusters can be automatically adjusted by the ISODATA through a splitting and merging mechanism, but the number of initial parameters is large, and the phenomenon of over-segmentation can occur due to the large influence of the initial parameter values; the FCM segmentation effect is good overall, but pixels with long space distance or large color difference are classified into the same class, and the clustering number needs to be set manually, so that the segmentation result is greatly influenced; gPb-OWT-UCM and LSC-UCM have small dependence on initial parameters, can generate closed region contours, and have larger difference of similar intervals in the regions. The gPb-OWT-UCM has a large calculation amount when obtaining the edge strength, and is easy to misjudge the target with uneven color into a plurality of categories, so the segmentation effect is worse than that of the LSC-UCM algorithm. The LSC-UCM algorithm provided by the invention is simple to calculate, and the contours of different ground feature regions after segmentation are clear and accurate, and are most similar to the result of manual segmentation.
The following further explains the segmentation evaluation index:
when evaluating the performance of the segmentation algorithm, the quantitative segmentation evaluation criterion is required to be introduced to carry out scientific, objective and accurate comparison only by depending on the subjective judgment to difficultly distinguish the segmentation algorithm performances with similar effects. The four different algorithms are quantitatively analyzed by using three evaluation indexes based on pixels, i.e., error rate (error), mean intersection over unit (mlou), and gray-level contrast (GC), and running time.
The error rate represents the proportion of the error-divided pixel number to the whole image, and is often used to describe the segmentation accuracy of the target object, and the calculation formula is as follows:
Figure BDA0002083977430000131
in the formula NiRepresenting the number of pixels in class i, Ni *And when the number of the classes automatically generated by the algorithm is different from the number of the classes manually segmented, the different part is wrong segmentation. The smaller the error rate, and the better the algorithm segmentation performance.
The intersection-to-union ratio is the ratio of the intersection and union of the algorithm segmentation result and the manual segmentation result, if the image is segmented into k types manually, the average intersection-to-union ratio is the average value of the intersection-to-union ratios of the types, and the calculation formula is as follows:
Figure BDA0002083977430000141
in the formula niiA pixel is classified into the pixel number of the ith class for both manual segmentation and algorithm,
Figure BDA0002083977430000142
for the algorithm to be divided into the number of pixels of the ith class,
Figure BDA0002083977430000143
for the number of pixels of the i-th class of the artificial segmentation (k' is the number of the class of the algorithm segmentation), the numerator and the denominator respectively represent the intersection and the union of the pixels of the i-th class in the artificial segmentation and the algorithm segmentation, and the larger the intersection ratio is, the more similar the result of the algorithm segmentation and the result of the artificial segmentation is.
Gray scale contrast ratio according toJudging the quality of the segmentation map according to the characteristic contrast between the regions, and if the average gray scale of each of two arbitrarily adjacent regions is fiAnd fjThe gray contrast between them can be calculated as follows:
Figure BDA0002083977430000144
the larger the gray scale contrast of the segmentation map is, the larger the difference between the regions segmented by the algorithm is.
Fig. 5a to fig. 10d are quantitatively analyzed by using the error rate, the average cross-correlation ratio, the gray contrast and the running time as evaluation indexes, and the results are shown in fig. 11 to fig. 14, fig. 11 is a schematic diagram of the error rate comparison of each algorithm according to the second embodiment of the present invention, fig. 12 is a schematic diagram of the average cross-correlation comparison of each algorithm according to the second embodiment of the present invention, fig. 13 is a schematic diagram of the gray contrast comparison of each algorithm according to the second embodiment of the present invention, and fig. 14 is a schematic diagram of the running time comparison of each algorithm according to the second embodiment of the present invention.
As shown in fig. 11, the LSC-UCM has the smallest number of wrongly-divided pixels and stable performance; secondly, gPb-OWT-UCM and ISODATA are adopted, but the segmentation performance is relatively unstable; the error rate of FCM is the largest and the segmentation result is the worst.
As shown in fig. 12, the LSC-UCM average cross-over ratio is the largest, which indicates the closest approach to the manual segmentation result; gPb-OWT-UCM segmentation results are also good; the average cross-over ratio of ISODATA and FCM is relatively low, and the difference from the manual segmentation result is large.
As shown in fig. 13, the gray contrast between adjacent regions divided by ISODATA is the highest; the gray scale contrast of the FCM, the gPb-OWT-UCM and the LSC-UCM is relatively close, and the difference among the areas of the segmentation result is slightly poor.
As shown in FIG. 14, FCM and ISODATA run at the shortest time and LSC-UCM runs at the fastest speed, and gPb-OWT-UCM runs at about 60 times the time and at the slowest speed.
The analysis is combined to know that the FCM has the fastest running speed but the highest fault rate; the ISODATA method is fast, the difference between the divided areas is large, but the dividing precision is slightly poor; gPb-OWT-UCM segmentation accuracy is high, but the operation complexity is high when gPb features are extracted and an edge intensity graph is generated, so that the method is not suitable for processing large-size aerial images; the LSC-UCM provided by the invention has high segmentation precision, and the running speed is far higher than gPb-OWT-UCM, so that the algorithm has obvious advantages in both segmentation precision and speed.
The method can rapidly and accurately segment different types of ground objects in the image, the segmentation result not only retains the remarkable edge of the target region, but also combines the regions with higher similarity, the method is superior to other three algorithms in the aspects of error rate and average intersection compared with the two measures reflecting the image segmentation accuracy, and the operation complexity is obviously improved compared with gPb-OWT-UCM.
The forest region image segmentation algorithm based on the superpixel and the hypermetric profile map can autonomously set the segmentation threshold T, and can set the segmentation thresholds T with different sizes according to actual needs so as to obtain a better segmentation result. Compared with other existing image segmentation algorithms, the method has the advantages of remarkably reduced operation complexity, higher algorithm speed, good segmentation effect, small dependence on initial parameters and the like, is very suitable for segmenting high-resolution unmanned aerial vehicle forest area aerial images, and has high popularization and application values.
Those of ordinary skill in the art will understand that: the figures are merely schematic representations of one embodiment, and the blocks or flow diagrams in the figures are not necessarily required to practice the present invention.
Those of ordinary skill in the art will understand that: modules in the devices in the embodiments may be distributed in the devices in the embodiments according to the description of the embodiments, or may be located in one or more devices different from the embodiments with corresponding changes. The modules of the above embodiments may be combined into one module, or further split into multiple sub-modules.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (6)

1. A forest region image segmentation algorithm based on superpixels and a hypermetrological contour map is characterized by comprising the following steps:
s1: performing superpixel segmentation on an original image to generate a superpixel map comprising a plurality of superpixel regions, wherein the size of the original image is M multiplied by N and is an RGB image;
s2: converting the superpixel graph from an RGB color space to an HSV color space, and averagely dividing hue component intervals, saturation component intervals and lightness component intervals of the converted superpixel graph in the HSV color space into n subintervals;
s3: respectively counting the number of pixels in each sub-interval of each super-pixel area in the hue component interval, the saturation component interval and the brightness component interval and normalizing the number of pixels to obtain a normalized histogram
Figure FDA0002083977420000011
Where m is 3 × n, n is an integer greater than 1, i represents the ith super pixel region,
Figure FDA0002083977420000012
respectively representing the normalized values of the number of pixels of the ith super pixel region in the 1 st to nth sections of the tone component section,
Figure FDA0002083977420000013
respectively representing the pixel number normalized values of the ith super pixel region in 1 st to nth intervals in the saturation component interval,
Figure FDA0002083977420000014
1 st to E, respectively representing the ith super pixel region in the lightness component intervalnormalizing the pixel number of the n intervals;
s4: calculating the dissimilarity D (R) between all adjacent two super-pixel regions according to the following formulai,Rj)：
Figure FDA0002083977420000015
Wherein R isi、RjRepresenting two adjacent super-pixel regions;
s5: according to the calculation result of the step S4, obtaining a dissimilarity sequence S according to the dissimilarity from small to large, wherein each element in the dissimilarity sequence S is the dissimilarity between two adjacent superpixel regions;
s6: initializing to generate a matrix U, wherein the U is M multiplied by N and is a zero matrix;
s7: corresponding all pixel points between two adjacent super pixel areas in the super pixel map to elements in the matrix U one by one, and respectively assigning the dissimilarity degrees obtained by the calculation in the step S4 to the corresponding elements in the matrix U;
s8: taking out the element U with the minimum value from the dissimilarity sequence S, merging two adjacent superpixel regions corresponding to the element U, wherein the merging rule is that the element in a matrix U corresponding to each pixel point between the two adjacent superpixel regions corresponding to the element U is updated to the element U, and simultaneously, the histogram of the merged superpixel region R is calculated as
Figure FDA0002083977420000021
Wherein a () represents a total number of pixels in the super pixel region;
s9: calculating the dissimilarity degree between the combined super-pixel region R in the step S8 and all the super-pixel regions adjacent to the combined super-pixel region R, and updating a dissimilarity degree sequence S according to the dissimilarity degree sequence;
s10: repeating the steps S8-S9 until the number of elements in the dissimilarity sequence S is zero;
s11: normalizing all elements in the matrix U to be 0-1;
s12: selecting a segmentation threshold value T, removing the adjacent region smaller than the segmentation threshold value T from the superpixel map obtained in the step S1, and reserving the adjacent region larger than the segmentation threshold value T to obtain a segmentation map comprising a plurality of sub-regions;
in this step, the segmentation threshold T is compared with each element in the matrix U after the completion of the step S11, for each element in the matrix U, all elements greater than the segmentation threshold T and all elements less than the segmentation threshold T are analyzed one by one, the elements are mapped one by one to the superpixel map obtained in the step S1, the elements less than the segmentation threshold T are removed from the corresponding edge in the superpixel map, and the elements greater than the segmentation threshold T are retained from the corresponding edge in the superpixel map. Here, "remove" means to remove the contour line for segmentation drawn at the pixel point, and "retain" means to retain the contour line for segmentation drawn at the corresponding pixel point in the super-pixel map.
S13: calculating the average color of each sub-region in the segmentation map obtained in step S12, and filling each average color in the corresponding sub-region to obtain a final segmentation result map.
2. The forest region image segmentation algorithm based on superpixels and the hypermetrological profile map as claimed in claim 1, wherein in step S1, the number of superpixel regions in the superpixel map is initialized to 100.
3. The forest region image segmentation algorithm based on superpixels and the hypermetrological profile map as claimed in claim 1, wherein in step S1, the superpixel map is generated by using a linear spectral clustering method.
4. The forest zone image segmentation algorithm based on the superpixel and the hypermetrological profile map as claimed in claim 1, wherein the value range of the hue component interval is between 0 and 255, the value range of the saturation component interval is between 0 and 255, and the value range of the lightness component interval is between 0 and 255.
5. The forest region image segmentation algorithm based on superpixels and the hypermetrological profile map as claimed in claim 1, wherein n is 25.
6. The super-pixel and super-metric profile map-based forest region image segmentation algorithm according to claim 1, wherein the normalization function used in the step S11 is a sigmoid function.
CN201910481471.1A 2019-06-04 2019-06-04 Forest image segmentation method based on superpixels and hyper-metric profile map Active CN112037230B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910481471.1A CN112037230B (en) 2019-06-04 2019-06-04 Forest image segmentation method based on superpixels and hyper-metric profile map

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910481471.1A CN112037230B (en) 2019-06-04 2019-06-04 Forest image segmentation method based on superpixels and hyper-metric profile map

Publications (2)

Publication Number Publication Date
CN112037230A true CN112037230A (en) 2020-12-04
CN112037230B CN112037230B (en) 2023-07-18

Family

ID=73575871

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910481471.1A Active CN112037230B (en) 2019-06-04 2019-06-04 Forest image segmentation method based on superpixels and hyper-metric profile map

Country Status (1)

Country Link
CN (1) CN112037230B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115578660A (en) * 2022-11-09 2023-01-06 牧马人(山东)勘察测绘集团有限公司 Land block segmentation method based on remote sensing image

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2682936A1 (en) * 2012-07-02 2014-01-08 TP Vision Holding B.V. Image processing
CN105118049A (en) * 2015-07-22 2015-12-02 东南大学 Image segmentation method based on super pixel clustering
CN105930815A (en) * 2016-05-04 2016-09-07 中国农业大学 Underwater organism detection method and system
WO2019062092A1 (en) * 2017-09-30 2019-04-04 深圳市颐通科技有限公司 Superpixel- and multivariate color space-based body outline extraction method
CN109785329A (en) * 2018-10-29 2019-05-21 重庆师范大学 Based on the purple soil image segmentation extracting method for improving SLIC algorithm

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2682936A1 (en) * 2012-07-02 2014-01-08 TP Vision Holding B.V. Image processing
CN105118049A (en) * 2015-07-22 2015-12-02 东南大学 Image segmentation method based on super pixel clustering
CN105930815A (en) * 2016-05-04 2016-09-07 中国农业大学 Underwater organism detection method and system
WO2019062092A1 (en) * 2017-09-30 2019-04-04 深圳市颐通科技有限公司 Superpixel- and multivariate color space-based body outline extraction method
CN109785329A (en) * 2018-10-29 2019-05-21 重庆师范大学 Based on the purple soil image segmentation extracting method for improving SLIC algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
宋以宁;刘文萍;骆有庆;宗世祥;: "基于线性谱聚类的林地图像中枯死树监测", 林业科学, no. 04, pages 187 - 195 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115578660A (en) * 2022-11-09 2023-01-06 牧马人(山东)勘察测绘集团有限公司 Land block segmentation method based on remote sensing image

Also Published As

Publication number Publication date
CN112037230B (en) 2023-07-18

Similar Documents

Publication Publication Date Title
CN107292339B (en) Unmanned aerial vehicle low-altitude remote sensing image high-resolution landform classification method based on feature fusion
CN109636905B (en) Environment semantic mapping method based on deep convolutional neural network
CN108573276B (en) Change detection method based on high-resolution remote sensing image
CN106909902B (en) Remote sensing target detection method based on improved hierarchical significant model
CN107392925B (en) Remote sensing image ground object classification method based on super-pixel coding and convolutional neural network
CN108537239B (en) Method for detecting image saliency target
CN109146889B (en) Farmland boundary extraction method based on high-resolution remote sensing image
CN106127791B (en) A kind of contour of building line drawing method of aviation remote sensing image
CN107273905B (en) Target active contour tracking method combined with motion information
CN108154157B (en) Fast spectral clustering method based on integration
CN106023257A (en) Target tracking method based on rotor UAV platform
CN110598564B (en) OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method
US20130342694A1 (en) Method and system for use of intrinsic images in an automotive driver-vehicle-assistance device
CN110147780B (en) Real-time field robot terrain identification method and system based on hierarchical terrain
CN104657980A (en) Improved multi-channel image partitioning algorithm based on Meanshift
CN109829507B (en) Aerial high-voltage transmission line environment detection method
CN106960182A (en) A kind of pedestrian integrated based on multiple features recognition methods again
CN112946679B (en) Unmanned aerial vehicle mapping jelly effect detection method and system based on artificial intelligence
CN109829426B (en) Railway construction temporary building monitoring method and system based on high-resolution remote sensing image
CN105825168A (en) Golden snub-nosed monkey face detection and tracking algorithm based on S-TLD
Femiani et al. Shadow-based rooftop segmentation in visible band images
CN114359323A (en) Image target area detection method based on visual attention mechanism
CN109741358B (en) Superpixel segmentation method based on adaptive hypergraph learning
CN110634142A (en) Complex vehicle road image boundary optimization method
Lu et al. Spectral segmentation via midlevel cues integrating geodesic and intensity

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