CN110796181B - Cultural relic disease high-precision automatic extraction method based on texture - Google Patents

Cultural relic disease high-precision automatic extraction method based on texture Download PDF

Info

Publication number
CN110796181B
CN110796181B CN201910973573.5A CN201910973573A CN110796181B CN 110796181 B CN110796181 B CN 110796181B CN 201910973573 A CN201910973573 A CN 201910973573A CN 110796181 B CN110796181 B CN 110796181B
Authority
CN
China
Prior art keywords
clustering
edge
cultural relic
disease
segmentation
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.)
Active
Application number
CN201910973573.5A
Other languages
Chinese (zh)
Other versions
CN110796181A (en
Inventor
胡春梅
王瑜
黄浩雯
崔哲
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 University of Civil Engineering and Architecture
Original Assignee
Beijing University of Civil Engineering and Architecture
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 University of Civil Engineering and Architecture filed Critical Beijing University of Civil Engineering and Architecture
Publication of CN110796181A publication Critical patent/CN110796181A/en
Application granted granted Critical
Publication of CN110796181B publication Critical patent/CN110796181B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a texture-based cultural relic disease high-precision automatic extraction method, which comprises the following steps: step one, acquiring an orthoscopic image of a cultural relic; taking the orthoimage of the cultural relic as a data processing basis, carrying out SLIC clustering on the orthoimage data of the cultural relic by using an improved simple linear iteration superpixel segmentation method, and segmenting the diseased region of the cultural relic into a plurality of pixel sets with similar sizes and same attributes; and step three, fusing the disease areas of the cultural relics together by taking the pixel set as a clustering center through a clustering mode based on information transmission, and counting the disease conditions of the cultural relics. The invention also discloses a texture-based cultural relic disease high-precision automatic extraction system, which comprises an image segmentation module, a clustering module and a disease statistical module. The method greatly reduces the human intervention interference in the cultural relic disease investigation process, provides basic data for the establishment and research of the restoration, perfection and protection scheme of the cultural relics, and has practical significance and scientific research value for the digital protection of the cultural relics.

Description

Cultural relic disease high-precision automatic extraction method based on texture
Technical Field
The invention relates to a high-precision automatic cultural relic disease extraction method based on textures.
Background
The study on the extraction of the cultural relic diseases is the basis of cultural relic protection and cultural relic study, and the digitization of the cultural relics in the cultural relic protection project is the premise of the extraction of the cultural relic diseases. At present, for the extraction of cultural relic diseases, a manual extraction method based on cultural relic digital information is still generally adopted, and the problems are as follows: (1) when image data are obtained, if no field lighting environment adjusting parameter exists and a colorless card is used for performing subsequent color correction, the correctness of disease texture information cannot be ensured; (2) disease image data acquired many times in the disease investigation process are not orthoimages, and the image is used for disease investigation, so that the reliability of investigation results cannot be guaranteed; (3) disease extraction is mostly realized by manually extracting disease areas by using software, and the method has heavy workload and low efficiency. Therefore, the automatic extraction of cultural relic diseases by using undistorted images is the key point of the research of scholars at home and abroad at present. At present, methods such as edge extraction and image segmentation are mainly studied in the aspect of automatic region extraction. The problems that exist are mainly: (1) the extracted edge data caused by the complexity of the cultural relics is difficult to judge whether the extracted edge data is a disease edge; (2) it is difficult to count the area size of the disease itself and the length of the disease boundary. Conventional image segmentation methods such as multi-threshold segmentation, clustering-based segmentation, region-based segmentation, etc. The problems that exist are mainly: (1) multi-threshold segmentation methods do not necessarily enable obtaining a threshold on an image with flat histograms or peaks; (2) the clustering-based segmentation method needs to determine the number of clusters, and the clustering result is larger due to the image of an initial clustering center; (3) region-based segmentation methods are easy to generate over-segmentation combined with region merging to obtain better results, and are generally applied to the fields of medical image analysis and the like.
Disclosure of Invention
An object of the present invention is to solve at least the above problems and/or disadvantages and to provide at least the advantages described hereinafter.
The invention also aims to provide a texture-based cultural relic disease high-precision automatic extraction method.
The invention also aims to provide a texture-based cultural relic disease high-precision automatic extraction system.
Therefore, the technical scheme provided by the invention is as follows:
a high-precision automatic cultural relic disease extraction method based on textures is characterized by comprising the following steps:
step one, acquiring an orthoscopic image of a cultural relic;
taking the orthoimage of the cultural relic as a data processing basis, carrying out SLIC clustering on the orthoimage data of the cultural relic by using an improved simple linear iteration superpixel segmentation method SLIC, and segmenting the diseased region of the cultural relic into a plurality of pixel sets with similar sizes and same attributes;
wherein the improvement to the simple linear iterative superpixel segmentation method comprises the following steps: introducing an edge extraction algorithm, taking the edge recall rate of edge data of a segmentation result of a simple linear iteration superpixel segmentation method and edge data of an edge extraction operator as reference, adjusting the ratio of the edge recall rate by changing the number k of clustering centers and a compactness parameter m, and outputting the values of k and m and the segmentation result until the edge recall rate meets a set requirement;
the calculation of the edge recall rate is that the edge data of an edge extraction algorithm Canny is used as a true edge, and the super-pixel edge data obtained by the SLIC super-pixel segmentation result is calculated with the edge data, and is shown as the following formula:
Figure BDA0002232895620000021
where s represents a superpixel edge dataset;
g-represents a Canny edge dataset;
p-represents the number of edge pixels in the artificial standard segmentation result;
q-represents the number of edge pixels in the superpixel segmentation result;
δg-representing the set of all edge pixels in the manual standard segmentation result;
δs-representing the set of all edge pixels in the superpixel segmentation result;
and step three, fusing the disease areas of the cultural relics together by taking the pixel set as a clustering center through a clustering mode based on information transfer AP, and counting and calculating the disease condition of the cultural relics.
Preferably, in the texture-based high-precision automatic cultural relic disease extraction method, in the second step, a specific improved calculation process for the simple linear iteration superpixel segmentation method is as follows:
step 1): giving a clustering center number k and a compactness parameter m;
step 2): performing edge extraction on the image by using a Canny algorithm, and recording the row and column numbers of edge pixels;
step 3): performing superpixel segmentation according to the cluster center number and the compactness parameter given in the step 1) to obtain superpixel edge data;
step 4): calculating the edge recall rate;
step 5): if the edge recall rate reaches more than 0.9, outputting the number of the current clustering centers, if the edge recall rate is less than 0.9, increasing the number of the clustering centers by 10, and repeating the step 3);
step 6): when the number K of the clustering centers is increased to 300, if the edge recall rate still does not reach 0.9, updating the compactness parameter, and reducing the compactness parameter by 10 every time to perform superpixel segmentation;
step 7): after the edge recall rate reaches the set requirement, outputting the number of the clustering centers meeting the requirement and a compactness parameter to obtain an improved SLIC superpixel segmentation algorithm;
step 8): and segmenting the orthoimage image data of the cultural relic by using the SLIC superpixel segmentation algorithm to obtain k pixel sets.
Preferably, in the method for automatically extracting cultural relic diseases with high precision based on textures, in the third step, the specific steps of fusing the disease areas of the cultural relics together by using the pixel set as a clustering center through a clustering mode based on information transfer include:
step 9): calculating the color component mean value X of all points in the Lab color space in the super pixel block R (i)R(i)And using the color characteristic direction as the color characteristic direction of the super pixelAn amount;
step 10): calculating Euclidean distance between any super pixel blocks according to a formula (1) to obtain a similarity matrix S;
step 11): taking the median of the similarity matrix S as an initialization parameter, setting the damping coefficient lambda to 0.5, and enabling the attraction degree and the attribution degree parameters to be 0;
step 12): calculating the attribution degree and the attraction degree according to the formulas (2) to (4):
s(i，k)＝-||XR(i)-XR(k)||2，i≠k (1)
Figure BDA0002232895620000031
Figure BDA0002232895620000032
Figure BDA0002232895620000033
where r (i, j) -represents the attraction;
a (i, j) -represents degree of attribution;
i, j-representing two different sets of superpixel cluster pixels;
s (i, j) — represents the euclidean distance of the non-cluster centers;
k' -represents other cluster centers;
i' -represents data points in other classes;
a (i, k') — representing the degree of attribution to other cluster centers;
s (i, k') — represents the euclidean distance to other cluster center points;
r (k, k) -judging the attraction degree between the cluster centers;
r (i', k) — the attractiveness of other cluster data to the current cluster center;
s (i, k) — represents the euclidean distance from the center of the cluster;
updating the attribution degree and the attraction degree according to formulas (5) to (8):
Figure BDA0002232895620000041
Figure BDA0002232895620000042
Figure BDA0002232895620000043
in the formula rt-1(i, k) and at-1(i, k) -representing the updated attractiveness and attribution after the t-1 th iteration;
lambda represents a damping coefficient, and the convergence rate is higher if the numerical value is larger;
at(k, k) -representing the result of this iteration;
s (i, k') — represents euclidean distances from the center points of the other clusters;
rt-1(k, k) -representing the result of the last iteration of clustering;
step 13): determining the number of cluster centers by the formula (8):
Figure BDA0002232895620000044
if i is not equal to k, k is the number of the clustering centers of i, if the iteration times in the iteration process exceed 1000 times, the iteration is terminated, and the numerical value of the current clustering center is obtained as the number of the clustering centers;
step 14): when the number k of the clustering centers is 2, the foreground and the background of the cultural relic image are segmented, and the segmentation effect is optimal; if the number of cluster centers is not converged to 2, Sil is calculated according to equation (9)R(i)Continuing iteration by reducing the value of p by p ═ p +0.1min (S), and when the average value of the contour coefficients continuously drops for more than 3 times orStopping the clustering algorithm when the number of the clustering centers is equal to 2;
Figure BDA0002232895620000051
in the formula, SilR(i)The clustering quality of the AP algorithm is represented, and the larger the value of the clustering quality of the AP algorithm is, the higher the quality of the algorithm clustering result is, and the better the image segmentation effect is;
a (i) -means representing the average of the distance of the superpixel points in the color space;
b (i) -means representing the average of the distance in color space of a superpixel from a superpixel point to a superpixel in another image partition;
step 15): and taking the number of the clustering centers corresponding to the maximum value of the contour coefficient average value as a clustering result.
Preferably, in the method for high-precision automatic extraction of cultural relic diseases based on textures, the predetermined requirement is that the edge recall rate reaches 90% or more.
Preferably, in the method for extracting cultural relic diseases based on textures automatically with high precision, the predetermined requirement is that the edge recall rate reaches 90%.
Preferably, in the high-precision automatic extraction method of the texture-based cultural relic diseases, in the step 1), the compactness parameter takes a value of 30.
Preferably, in the method for automatically extracting cultural relic diseases with high precision based on textures, in the third step, the disease condition of the cultural relic is calculated statistically, including the area of the disease area and the length of the edge.
Preferably, in the method for automatically extracting cultural relic diseases with high precision based on textures, in the second step, the generation of the ortho-image of the cultural relic comprises a full-digital ortho-image generation method and a point cloud-based ortho-image generation method, and the full-digital ortho-image generation method comprises the following steps: acquiring original image data, establishing a Digital Elevation Model (DEM), registering the digital image and the DEM to form a color DEM, and projecting the color DEM onto a horizontal plane to obtain an orthoimage; the forward projection image generation method based on the point cloud comprises the following steps: acquiring point cloud data, constructing a triangulation network model and manufacturing an orthoimage.
A high-precision automatic cultural relic disease extraction method based on textures comprises the following steps:
the image segmentation module is used for segmenting the orthoimage data of the cultural relic by the improved simple linear iteration superpixel segmentation method and outputting the number of superpixel clustering centers, the compactness parameter and the image segmentation result data;
the clustering module is connected with the image segmentation module and used for clustering results obtained by the improved simple linear iteration superpixel segmentation method by using the AP clustering method;
and the disease counting module is connected with the clustering module and is used for counting the condition of the cultural relic diseases according to the clustering result, wherein the condition comprises the area of the disease area and the length of the edge.
The invention at least comprises the following beneficial effects:
the invention provides disease extraction and analysis based on an ortho-image, and the ortho-image is image data which has the characteristics of high resolution, easy interpretation, easy measurement, short production period, rich detail information, convenient storage and the like. The introduction of an ortho image into the field of cultural relic protection is a necessary trend in line with the development of the times, and is one of the important means and methods for the digitization of cultural relic information.
On the basis of utilizing an orthoimage, a simple linear iteration-based superpixel segmentation method (SLIC) is provided for segmenting the image into superpixel blocks with similar attributes, and the initial clustering center of the SLIC superpixel segmentation algorithm and the compactness parameters of the superpixel boundary need to be input manually, so that the SLIC superpixel segmentation algorithm is improved, the improved SLIC superpixel segmentation algorithm does not need manual intervention, and a program can automatically calculate a result according to a certain rule. On the basis of a super-pixel segmentation result, cultural relic diseases can be well extracted by giving a clustering rule through a clustering algorithm based on information transmission.
The cultural relic disease data automatically extracted through the orthographic images can greatly reduce the human involvement and interference in the cultural relic disease investigation process, improve the working efficiency, provide basic data for the establishment and research of the restoration, perfection and protection scheme of the cultural relic, and have practical significance and scientific research value for the digital protection of the cultural relic.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a schematic flow chart of superpixel clustering and superpixel segmentation in one embodiment of the present invention;
FIG. 2 is a vectorization disease graph of a cultural relic;
FIG. 3 is a schematic flow chart of one embodiment of the present invention;
FIG. 4 is a program interface diagram in accordance with one embodiment of the present invention;
FIG. 5 is a functional diagram of a program in one embodiment of the present invention;
FIG. 6 is a diagram of optimizing SLIC superpixel clusters in accordance with one embodiment of the present invention;
FIG. 7 is a graph of information-based clusters in one embodiment of the present invention;
FIG. 8 is a graph of disease edge extraction based on one embodiment of the present invention;
FIG. 9 is a statistical plot of lesion area and perimeter in one embodiment of the present invention;
FIG. 10 is a diagram illustrating the effect of histogram equalization in one embodiment of the present invention;
FIG. 11 is a graph of image contrast enhancement effect in one embodiment of the present invention;
FIG. 12 is a diagram of the brightness enhancement effect according to one embodiment of the present invention;
FIG. 13 is a graph of squared error template matching for image matching in one embodiment of the present invention;
FIG. 14 is a graph of correlation coefficient template matching in one embodiment of the present invention;
FIG. 15 is a diagram of the results of a thresholding method in accordance with one embodiment of the present invention;
FIG. 16 is a diagram of the results of using the GS partitioning method in one embodiment of the present invention;
FIG. 17 is a graph showing the results of the SEEDs segmentation method used in one embodiment of the present invention.
Detailed Description
The present invention is described in further detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description.
It will be understood that terms such as "having," "including," and "comprising," as used herein, do not preclude the presence or addition of one or more other elements or groups thereof.
The invention provides a high-precision automatic cultural relic disease extraction method based on textures, which comprises the following steps:
step one, acquiring an orthoscopic image of a cultural relic;
taking the orthoimage of the cultural relic as a data processing basis, carrying out SLIC clustering on the orthoimage data of the cultural relic by using an improved simple linear iteration superpixel segmentation method SLIC, and segmenting the diseased region of the cultural relic into a plurality of pixel sets with similar sizes and same attributes;
wherein the improvement to the simple linear iterative superpixel segmentation method comprises the following steps: introducing an edge extraction algorithm, taking the edge recall rate of edge data of a segmentation result of a computational simple linear iteration superpixel segmentation method and edge data of an edge extraction operator as reference, adjusting the ratio of the edge recall rate by changing the number k of clustering centers and a compactness parameter m, and outputting the values of k and m and a segmentation result until the edge recall rate meets a set requirement;
the calculation of the edge recall rate is that the edge data of an edge extraction algorithm Canny is used as a true edge, and the super-pixel edge data obtained by the SLIC super-pixel segmentation result is calculated with the edge data, and is shown as the following formula:
Figure BDA0002232895620000071
where s represents a superpixel edge dataset;
g-represents a Canny edge dataset;
p-represents the number of edge pixels in the artificial standard segmentation result;
q-represents the number of edge pixels in the superpixel segmentation result;
δg-representing the set of all edge pixels in the manual standard segmentation result;
δs-representing the set of all edge pixels in the superpixel segmentation result;
and step three, fusing the disease areas of the cultural relics together by taking the pixel set as a clustering center through a clustering mode based on information transfer AP, and counting and calculating the disease conditions of the cultural relics.
In one embodiment of the present invention, preferably, in the step two, the specific calculation process for the improvement of the simple linear iterative superpixel segmentation method is as follows:
step 1): giving a clustering center number k and a compactness parameter m;
step 2): performing edge extraction on the image by using a Canny algorithm, and recording the row and column numbers of edge pixels;
step 3): performing superpixel segmentation according to the cluster center number and the compactness parameter given in the step 1) to obtain superpixel edge data;
step 4): calculating the edge recall rate;
step 5): if the edge recall rate reaches more than 0.9, outputting the number of the current clustering centers, if the edge recall rate is less than 0.9, increasing the number of the clustering centers by 10, and repeating the step 3);
step 6): when the number K of the clustering centers is increased to 300, if the edge recall rate still does not reach 0.9, updating the compactness parameter, and reducing the compactness parameter by 10 every time to perform superpixel segmentation;
step 7): after the edge recall rate reaches the set requirement, outputting the number of the clustering centers meeting the requirement and a compactness parameter to obtain an improved SLIC superpixel segmentation algorithm;
step 8): and segmenting the orthoimage image data of the cultural relic by using the SLIC superpixel segmentation algorithm to obtain k pixel sets.
In one embodiment of the present invention, preferably, in step three, the specific step of fusing the damaged areas of the cultural relics together by using the pixel set as a clustering center in a clustering manner based on information transfer includes:
step 9): calculating the color component mean value X of all points in the Lab color space in the super pixel block R (i)R(i)And using the color feature vector as the color feature vector of the super pixel;
step 10): calculating Euclidean distance between any super pixel blocks according to a formula (1) to obtain a similarity matrix S;
step 11): taking the median of the similarity matrix S as an initialization parameter, setting the damping coefficient lambda to 0.5, and enabling the attraction degree and the attribution degree parameters to be 0;
step 12): calculating the attribution degree and the attraction degree according to the formulas (2) to (4):
s(i，k)＝-||XR(i)-XR(k)||2，i≠k (1)
Figure BDA0002232895620000091
Figure BDA0002232895620000092
Figure BDA0002232895620000093
where r (i, j) -represents the attraction;
a (i, j) -represents degree of attribution;
i, j-representing two different sets of superpixel cluster pixels;
s (i, j) — represents the euclidean distance of the non-cluster centers;
k' -represents other cluster centers;
i' -represents data points in other classes;
a (i, k') — representing the degree of attribution to other cluster centers;
s (i, k') — represents euclidean distances from the center points of the other clusters;
r (k, k) -judging the attraction degree between the cluster centers;
r (i', k) — the attractiveness of other cluster data to the current cluster center;
s (i, k) — represents the euclidean distance from the center of the cluster;
updating the attribution degree and the attraction degree according to formulas (5) to (8):
Figure BDA0002232895620000094
Figure BDA0002232895620000095
Figure BDA0002232895620000096
in the formula rt-1(i, k) and at-1(i, k) -representing the updated attractiveness and attribution after the t-1 th iteration;
lambda represents a damping coefficient, and the convergence speed is higher when the numerical value is larger;
at(k, k) -representing the result of this iteration;
s (i, k') — represents euclidean distances from the center points of the other clusters;
rt-1(k, k) -representing the result of the last iteration of clustering;
step 13): determining the number of cluster centers by the formula (8):
Figure BDA0002232895620000101
if i is not equal to k, k is the number of the clustering centers of i, if the iteration times in the iteration process exceed 1000 times, the iteration is terminated, and the numerical value of the current clustering center is obtained as the number of the clustering centers;
step 14): when the number k of the clustering centers is 2, the foreground and the background of the cultural relic image are segmented, and the segmentation effect is optimal; if the number of cluster centers is not converged to 2, Sil is calculated according to equation (9)R(i)Continuing iteration by reducing the value p in p +0.1min (S), and stopping the clustering algorithm when the average value of the contour coefficients continuously drops for more than 3 times or the number of clustering centers is equal to 2;
Figure BDA0002232895620000102
in the formula, SilR(i)The clustering quality of the AP algorithm is represented, and the larger the value of the clustering quality of the AP algorithm is, the higher the quality of the algorithm clustering result is, and the better the image segmentation effect is;
a (i) -means representing the average of the distance of the superpixel points in the color space;
b (i) -means representing the average of the distance in color space of a superpixel from a superpixel point to a superpixel in another image partition;
step 15): and taking the number of the clustering centers corresponding to the maximum value of the contour coefficient average value as a clustering result.
In one of the embodiments of the present invention, the intended requirement is preferably an edge recall of 90% and above.
In one of the embodiments of the present invention, the intended requirement is preferably that the edge recall rate be up to 90%.
In one embodiment of the present invention, preferably, in step 1), the compactness parameter takes a value of 30.
In one embodiment of the present invention, preferably, in step three, the disease condition of the cultural relic is statistically calculated, including the area of the disease area and the length of the edge.
In one embodiment of the present invention, preferably, in the second step, the generating of the ortho-image of the cultural relic includes a full-digital ortho-image generating method and a point cloud-based ortho-image generating method, and the full-digital ortho-image generating method includes: acquiring original image data, establishing a Digital Elevation Model (DEM), registering a digital image and the DEM to form a color DEM, and projecting the color DEM onto a horizontal plane to obtain an orthoimage; the orthoimage generation method based on the point cloud comprises the following steps: acquiring point cloud data, constructing a triangulation network model and manufacturing an orthoimage.
The invention also provides a texture-based cultural relic disease high-precision automatic extraction method, which comprises the following steps:
an image segmentation module which segments the orthoimage data of the cultural relic by the improved simple linear iterative superpixel segmentation method as claimed in claim 1, and outputs the number of the superpixel clustering centers, the compactness parameter and the image segmentation result data;
a clustering module, connected to the image segmentation module, for clustering results obtained by the improved simple linear iterative superpixel segmentation method using the AP clustering method according to claim 1;
the disease counting module is connected with the clustering module and used for counting the cultural relic disease conditions including the area of the disease area and the length of the edge according to the clustering result
In order to make the technical solution of the present invention better understood by those skilled in the art, the following examples are now provided for illustration:
based on the problems, the invention provides an automatic extraction method of a super-pixel segmentation disease based on an orthoimage. The main research contents include: (1) the method for generating an orthoimage mainly comprises a point cloud-based orthoimage generating method and a full-digital orthoimage generating method; (2) the super-pixel segmentation algorithm research mainly comprises the comparison of several super-pixel segmentation algorithms and the optimization of a simple linear iteration super-pixel segmentation method (SLIC); (3) and (3) extracting the disease area segmented by the superpixels by using an AP clustering method based on an information transfer clustering method (AP).
As shown in fig. 1, a disease extraction method based on superpixel segmentation. Because each super-pixel segmentation method has advantages and disadvantages, the defect that the super-pixel initial clustering center is not easy to select is improved on the basis of the compactness of the ASLIC self-adaptive super-pixel, so that the SLIC super-pixel segmentation method can be suitable for the segmentation requirements of cultural relic diseases: the edge attaching degree is high, the super-pixel area is compact and coherent, and the label distribution is convenient for disease statistics. And on the basis of finishing the SLIC super-pixel algorithm optimization, combining a clustering method based on information transmission, and clustering the over-segmented super-pixel regions together again. The label and the edge data are fused to facilitate information extraction of the cultural relic diseases.
Aiming at the particularity of cultural heritage, the method for extracting cultural relic diseases by combining superpixel segmentation with information transfer clustering is provided by taking an orthoimage as a data source and taking high-precision cultural relic disease automatic extraction as a research target. Firstly, in order to avoid the influence of different projection modes on the subsequent disease extraction and statistics, the orthoimage of the cultural relic is used as the data processing basis. And then, SLIC clustering is carried out on the cultural relic disease area, and the cultural relic disease area is divided into a plurality of areas with similar sizes and the same attributes. And finally, fusing the cultural relic disease areas together by using the superpixel blocks as the clustering centers in an information transfer-based clustering (AP) mode, and counting and calculating data such as the areas of the disease areas, the lengths of the edges and the like through a given label. The main contents are divided into two parts of an image edge extraction and segmentation based on an orthoimage and a disease extraction method based on super-pixel segmentation.
(1) Image edge extraction and segmentation based on orthoimages. The generation of the ortho-image mainly includes a full-digital ortho-image generation method and a point cloud-based ortho-image generation method. The all-digital orthographic image generation method comprises the following steps: the method comprises the steps of obtaining original image data, establishing a Digital Elevation Model (DEM), registering the digital image and the DEM to form a color DEM, and projecting the color DEM onto a horizontal plane to obtain an orthoimage. The orthoimage generation method based on the point cloud comprises the following steps: acquiring point cloud data, constructing a triangulation network model and manufacturing an orthoimage. Image segmentation includes several superpixel segmentation methods: NC, SEEDS, GS, Mean Shift, turbopexes, SLIC, etc.
(2) A disease extraction method based on super-pixel segmentation. The Simple Linear Iteration (SLIC) clustering method has the advantages of high edge fitting degree, controllable super-pixel compactness and high calculation efficiency. However, for complex images, the parameters of superpixel cluster center and superpixel compactness which need to be initialized need to be judged by experience to ensure that the edge fitting degree is highest. Therefore, the algorithm is optimized in a mode of judging the edge recall rate on the basis that the ASLIC is self-adaptive to the super-pixel compactness, so that SLIC super-pixel segmentation can be completely self-adaptive to segmentation work for dealing with cultural relic diseases. And selecting an information transfer (AP) clustering-based method on the basis of segmentation, searching other clusters meeting the conditions by taking each super pixel as a cluster, and fusing the clusters of the contract types to obtain the final disease clustering result. And calculating data such as the area of the region, the boundary length and the like by counting the labels of the disease regions.
The traditional cultural relic disease extraction mode mainly adopts manual drawing, and has large error and low efficiency. At present, the disease extraction firstly carries out site survey on the basis of collecting data including historical photos, written records and the like and drawing the appearance of the cultural relic and the condition of suffering from the disease by contrasting the cultural relic by hands. However, such disease mapping can only satisfy the investigation work for roughly distributing the disease range in the early stage of the engineering, and does not have the basic conditions for studying the disease itself and its formation. The method is characterized in that images of cultural relics are shot on the basis of earlier research and photo shooting, the images are vectorized by importing the images into AutoCAD, and the diseases are scaled according to the size information measured on site, and the images correspond to the real size.
Based on the defects of the method, the disease extraction and analysis based on the ortho-image is provided, and the ortho-image is image data which has the characteristics of high resolution, easiness in interpretation, easiness in measurement, short production period, rich detail information, convenience in storage and the like. The introduction of orthoimages into the field of cultural relic protection is a necessary trend in accordance with the development of the times, and is one of the important means and methods for the digitization of cultural relic information.
On the basis of utilizing an orthoimage, a simple linear iteration-based superpixel segmentation method (SLIC) is provided for segmenting the image into superpixel blocks with similar attributes, and the initial clustering center of the SLIC superpixel segmentation algorithm and the compactness parameters of the superpixel boundary need to be input manually, so that the SLIC superpixel segmentation algorithm is improved, the improved SLIC superpixel segmentation algorithm does not need manual intervention, and a program can automatically calculate a result according to a certain rule. On the basis of a super-pixel segmentation result, cultural relic diseases can be well extracted by giving a clustering rule through a clustering algorithm based on information transmission.
The cultural relic disease data automatically extracted through the orthographic images can greatly reduce the human involvement and interference in the cultural relic disease investigation process, improve the working efficiency, provide basic data for the establishment and research of the restoration, perfection and protection scheme of the cultural relic, and have practical significance and scientific research value for the digital protection of the cultural relic. As shown in fig. 2.
At present, in practical engineering, two methods are mainly used for manual disease extraction.
(1) And (4) placing the scale around the disease during on-site exploration, and taking pictures of the disease and the scale. Firstly, when processing disease image data, according to a scale, a photo is stretched in Photoshop or the image data is imported into CAD to scale the image, and the image is adjusted to the actual size. Then, vector quantization operation is performed on the disease region in the CAD, and a vector diagram is drawn. And counting the area and the perimeter information of the disease area in the CAD for the drawn vector diagram.
(2) When image data are acquired on site, light is arranged according to the light condition on site, and the color temperature and the exposure degree are adjusted; and adjusting the focal length of the camera to enable the picture to be clear, and performing array type image shooting on the disease area after the adjustment is completed to acquire image data. And correcting the image distortion of the shot image data. And then splicing the disease images to form complete disease image data. And finally, importing the complete disease image data into the CAD for vectorization drawing and counting disease information.
The SLIC super-pixel clustering method is adopted for subsequent research, SLIC super-pixel segmentation has remarkable advantages, and is better than other super-pixel segmentation methods in both segmentation quality and segmentation efficiency, but the SLIC algorithm also has certain limitation. Firstly, from the aspect of controllability of the number of super pixels, the method is one of the advantages and disadvantages of the SLIC algorithm, and because the types of diseases are various and the formation forms are complex and diverse, the number of a cluster center needs to be estimated according to the image size and the disease condition when the SLIC algorithm is used for segmentation. The estimated initial clustering centers cannot meet the condition that most superpixel boundaries fall on the real edge of the disease, so that the number of the superpixel initial clustering centers needs to be redistributed. Secondly, the compactness degree of the super pixel region also needs to be calculated by presetting empirical values, and if the result is not ideal, the given order of magnitude of the compactness parameter needs to be judged again. The initial cluster center N and the compactness parameter M add a lot of labor effort while bringing sufficient openness and are inefficient.
The adjustable parameters in the SLIC superpixel segmentation algorithm comprise the number K of clustering centers and a compactness parameter m. The significance of the clustering center K is how many clustering results will be generated given how many K, and if the initial clustering center is set to 50, the segmentation result will present 50 superpixels, each superpixel representing a category. The compactness parameter M is in a value range of 1-40, the closer the value is to 40, the more regular the super-pixel block finally generated, and the smaller the value is, the more complex the boundary of the super-pixel block is.
The number of the super-pixel clustering centers and the super-pixel compactness parameter have obvious influence on the segmentation result, the super-pixel clustering centers K have positive correlation with the quality of the segmentation result, and the larger the K value is, the higher the edge fitting degree of the segmentation result is. The compactness m has negative correlation with the result of the super-pixel segmentation, the smaller the value of m is, the higher the edge fitting degree of the segmentation result is, and the larger the value of m is, the lower the edge fitting degree is, but the more regular the super-pixel is. Hence, the optimization improvements herein for SLIC superpixels will proceed from both of these aspects.
The method introduces an edge extraction operator, takes the edge extracted by the edge extraction operator as a true value edge, and takes the edge obtained by SLIC super pixel segmentation as an extraction target. And adjusting the ratio of the edge recall rate by calculating the edge data of the SLIC superpixel segmentation result and the edge recall rate of the boundary of the edge extraction operator as reference and changing the number of the clustering centers and the compactness parameter until the edge recall rate meets the set requirement, and outputting the superpixel number, the compactness parameter and the segmentation result. The specific flow is shown in fig. 3:
the method comprises the following specific steps:
step 1): giving a clustering center number k and a compactness parameter m;
step 2): performing edge extraction on the image by using a Canny algorithm, and recording the row and column numbers of edge pixels;
step 3): performing superpixel segmentation according to the cluster center number and the compactness parameter given in the step 1) to obtain superpixel edge data;
step 4): calculating the edge recall rate of the superpixel edge data obtained by the SLIC superpixel segmentation result by taking the edge data of an edge extraction algorithm Canny as a true edge; as shown in the following formula:
Figure BDA0002232895620000141
where s represents a superpixel edge dataset;
g-represents a Canny edge dataset;
p-represents the number of edge pixels in the artificial standard segmentation result;
q-represents the number of edge pixels in the super-pixel segmentation result;
δg-representing the set of all edge pixels in the manual standard segmentation result;
δsrepresenting the set of all edge pixels in the superpixel segmentation result
Step 5): if the edge recall rate reaches more than 0.9, outputting the number of the current clustering centers, if the edge recall rate is less than 0.9, increasing the number of the clustering centers by 10, and repeating the step 3);
step 6): when the number K of the clustering centers is increased to 300, if the edge recall rate still does not reach 0.9, updating the compactness parameter, and reducing the compactness parameter by 10 every time to perform superpixel segmentation;
step 7): after the edge recall rate reaches the set requirement, outputting the number of the clustering centers meeting the requirement and a compactness parameter to obtain an improved SLIC superpixel segmentation algorithm;
step 8): and segmenting the orthoimage image data of the cultural relic by using the SLIC superpixel segmentation algorithm to obtain k pixel sets.
In one embodiment of the present invention, preferably, in step three, the specific step of fusing the damaged areas of the cultural relics together by using the pixel set as a clustering center in a clustering manner based on information transfer includes:
step 9): calculating the color component mean value X of all points in the Lab color space in the super pixel block R (i)R(i)And using the color feature vector as the color feature vector of the super pixel;
step 10): calculating Euclidean distance between any super pixel blocks according to a formula (1) to obtain a similarity matrix S;
step 11): taking the median of the similarity matrix S as an initialization parameter, setting the damping coefficient lambda to 0.5, and enabling the attraction degree and the attribution degree parameters to be 0;
step 12): calculating the attribution degree and the attraction degree according to the formulas (2) to (4):
s(i，k)＝-||XR(i)-XR(k)||2，i≠k (1)
Figure BDA0002232895620000151
Figure BDA0002232895620000152
Figure BDA0002232895620000153
where r (i, j) -represents the attraction;
a (i, j) -represents degree of attribution;
i, j-representing two different sets of superpixel cluster pixels;
s (i, j) — represents the euclidean distance of the non-cluster centers;
k' -represents other cluster centers;
i' -represents data points in other classes;
a (i, k') — representing degree of attribution to other cluster centers;
s (i, k') — represents euclidean distances from the center points of the other clusters;
r (k, k) -judging the attraction degree between the cluster centers;
r (i', k) — the attractiveness of other cluster data to the current cluster center;
s (i, k) — represents the euclidean distance from the center of the cluster;
updating the attribution degree and the attraction degree according to formulas (5) - (8):
Figure BDA0002232895620000161
Figure BDA0002232895620000162
Figure BDA0002232895620000163
in the formula rt-1(i, k) and at-1(i, k) -indicating an update after the t-1 st iterationAttraction degree and attribution degree;
lambda represents a damping coefficient, and the convergence speed is higher when the numerical value is larger;
at(k, k) -representing the result of this iteration;
s (i, k') — represents euclidean distances from the center points of the other clusters;
rt-1(k, k) -representing the result of the last iteration of clustering;
step 13): determining the number of cluster centers by the formula (8):
Figure BDA0002232895620000164
if i is not equal to k, k is the number of the clustering centers of i, if the iteration times in the iteration process exceed 1000 times, the iteration is terminated, and the numerical value of the current clustering center is obtained as the number of the clustering centers;
step 14): when the number k of the clustering centers is 2, the foreground and the background of the cultural relic image are segmented, and the segmentation effect is optimal; if the number of clustering centers is not converged to 2, Sil is calculated according to equation (9)R(i)Continuing iteration by reducing the value p in p +0.1min (S), and stopping the clustering algorithm when the average value of the contour coefficients continuously drops for more than 3 times or the number of clustering centers is equal to 2;
Figure BDA0002232895620000171
in the formula, SilR(i)The clustering quality of the AP algorithm is represented, and the larger the value of the clustering quality of the AP algorithm is, the higher the quality of the algorithm clustering result is, and the better the image segmentation effect is;
a (i) -means representing the average of the distance of the superpixel points in the color space;
b (i) -means representing the average of the distance in color space of a superpixel from a superpixel point to a superpixel in another image partition;
step 15): and taking the number of the clustering centers corresponding to the maximum value of the contour coefficient average value as a clustering result.
Overall design of program
Aiming at realizing high-precision automatic extraction of the cultural relic diseases, the method, the process and the specific steps are combined to carry out SLIC superpixel segmentation on the normal incidence image of the cultural relic, then the superpixel blocks with the same attribute are clustered together by using an information transfer clustering method, and finally the area and the perimeter of the disease area are calculated.
And developing a library template by combining Opencv3.0 and QT5 open source user interfaces on the basis of C + + language. The main program interface is designed as shown in fig. 4 according to the overall design requirements, each main function module is shown in fig. 5, and the main function modules have the functions of image enhancement by segmentation, image matching, image segmentation, image clustering, disease statistics, disease representation and the like.
The image enhancement module mainly comprises several commonly used image enhancement methods, including histogram equalization, contrast enhancement, brightness enhancement and brightness contrast enhancement; the template matching function mainly comprises a square error matching method, a standardized square error matching method, a correlation coefficient matching method and a standardized correlation coefficient matching method; the image segmentation module comprises SLIC superpixel segmentation, segmentation by other segmentation methods and other methods for superpixel clustering; the clustering module is mainly a clustering method based on information transmission; the disease statistical module obtains data of the area and the perimeter of a disease area which a user wants to be interested in on the basis of clustering; the disease identification function can fill corresponding disease international identification in a disease area obtained by manual drawing or program automatic calculation, so that the disease identification is facilitated.
Implementation of specific design
For the main functional modules in the program, the functions of the main functional modules will be described correspondingly by taking the normal image of the cultural relic diseases as an example, and the detailed steps and flows of each part will be described.
1. Improved SLIC super pixel segmentation method
And introducing an edge extraction operator, taking the edge extracted by the edge extraction operator as a true value edge, and taking the edge obtained by SLIC superpixel segmentation as an extraction target. And adjusting the ratio of the edge recall rate by calculating the edge data of the SLIC superpixel segmentation result and the edge recall rate of the boundary of the edge extraction operator as reference and changing the number of the clustering centers and the compactness parameter until the edge recall rate meets the set requirement, and outputting the superpixel number, the compactness parameter and the segmentation result. As shown in fig. 6.
2. Image clustering module
The improved SLIC superpixels are clustered using AP clustering (Affinity prediction clustering algorithm), and all points have the same ability to become cluster centers before the algorithm starts to run, so the reference is usually the minimum or median of all values in the similarity matrix. The algorithm will continuously update the attraction degree and the attribution degree in the iterative process until K optimal clustering centers are generated, and the result is shown in fig. 7:
3. disease statistics module
The disease statistical function is mainly used for counting data such as surface area, perimeter and the like of cultural relic diseases, and the size of a cultural relic disease area has a certain corresponding relation with an actual disease because the text source data is an orthographic image. And if the scale of the image data is 1:1, the counted disease result is the real result. The disease statistical function is mainly realized through functions in an Opencv library. The statistical lesion area uses a function contourArea (), which requires contour points of the input image as starting points. And selecting the disease contour points through the mark after clustering, selecting the complete edge which is most overlapped with the Canny operator as the cultural relic disease edge, and counting the cultural relic disease edge. As shown in fig. 8 and 9.
4. Other functions
Image enhancement module
4.1 histogram equalization
If the gray values of the image include all the gray values and are uniformly distributed, the image must have strong contrast and rich color variation. The purpose of histogram equalization is to map the pixels in the image to make the gray distribution uniform, so as to enhance the contrast of the image. The histogram equalization calculation formula (8) is as follows:
Figure BDA0002232895620000181
the results are shown in FIG. 10.
4.2 Brightness and contrast
Calling the coverto () function in opengcv3.0 can change the brightness and contrast parameters of the image, so the brightness and contrast adjustments can be made to the image using the coverto () function, and the results are shown in fig. 11 and 12.
Image matching module
4.3 squared error matching method and Standard squared error matching
The method for matching two images by using the square difference is characterized in that the worse the matching is, the larger the matching value obtained by calculation is, and the matching value is 0 when the matching is successful in an ideal state. The square error matching model is as formula (9), and the standard square error matching model is as formula (10):
Figure BDA0002232895620000191
Figure BDA0002232895620000192
where T represents a band matching image;
i-represents the matching image.
The results are shown in FIG. 13.
4.4 correlation coefficient matching method and normalized correlation coefficient matching method
The method matches the correlation value of the template to the correlation value of the image to the mean value, and uses 1, -1 and 0 to respectively represent matching, mismatching and no correlation. Correlation coefficient matching mathematic model formula (11)
Figure RE-GDA0002345290260000013
Where R (m, n) -represents the correlation between images;
m, n-represent the band matching image and the matching image.
As shown in fig. 14.
4.4.3 other segmentation methods
Other segmentation methods include image segmentation methods such as threshold segmentation, SEEDS segmentation, NC segmentation, and GS segmentation, as shown in FIGS. 15, 16, and 17.
The number of modules and the processing scale described herein are intended to simplify the description of the invention. Applications, modifications and variations of XX's of the present invention will be apparent to those skilled in the art.
While embodiments of the invention have been disclosed above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in a variety of fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.

Claims (8)

1. A high-precision automatic cultural relic disease extraction method based on textures is characterized by comprising the following steps:
step one, acquiring an orthoscopic image of a cultural relic;
taking the orthoimage of the cultural relic as a data processing basis, carrying out SLIC clustering on the orthoimage data of the cultural relic by using an improved simple linear iteration superpixel segmentation method SLIC, and segmenting the diseased region of the cultural relic into a plurality of pixel sets with similar sizes and same attributes;
the improvement of the simple linear iteration super-pixel segmentation method comprises the following steps: introducing an edge extraction algorithm, taking the edge recall rate of edge data of a segmentation result of a simple linear iteration superpixel segmentation method and edge data of an edge extraction operator as reference, adjusting the ratio of the edge recall rate by changing the number k of clustering centers and a compactness parameter m, and outputting values of k and m and a segmentation result until the edge recall rate meets a set requirement;
the calculation of the edge recall rate is that the edge data of an edge extraction algorithm Canny is used as a true edge, and the super-pixel edge data obtained by the SLIC super-pixel segmentation result is calculated with the edge data, and is shown as the following formula:
Figure FDA0003509285200000011
where s represents a superpixel edge dataset;
g-represents a Canny edge dataset;
p-represents the number of edge pixels in the artificial standard segmentation result;
q-represents the number of edge pixels in the superpixel segmentation result;
δg-representing the set of all edge pixels in the manual standard segmentation result;
δs-representing a set of all edge pixels in the superpixel segmentation result;
the specific calculation process for the improvement of the simple linear iterative superpixel segmentation method is as follows:
step 1): giving a clustering center number k and a compactness parameter m;
step 2): performing edge extraction on the image by using a Canny algorithm, and recording the row and column numbers of edge pixels;
step 3): performing superpixel segmentation according to the cluster center number and the compactness parameter given in the step 1) to obtain superpixel edge data;
step 4): calculating the edge recall rate;
and step 5): if the edge recall rate reaches more than 0.9, outputting the number of the current clustering centers, if the edge recall rate is less than 0.9, increasing the number of the clustering centers by 10, and repeating the step 3);
step 6): when the number K of the clustering centers is increased to 300, if the edge recall rate still does not reach 0.9, updating the compactness parameter, and reducing the compactness parameter by 10 every time to perform superpixel segmentation;
step 7): after the edge recall rate reaches the set requirement, outputting the cluster center number and the compactness parameter meeting the requirement to obtain an improved SLIC super-pixel segmentation algorithm;
step 8): segmenting the orthoimage image data of the cultural relic by using the SLIC superpixel segmentation algorithm to obtain k pixel sets;
and step three, fusing the disease areas of the cultural relics together by taking the pixel set as a clustering center through a clustering mode based on information transfer AP, and counting and calculating the disease conditions of the cultural relics.
2. The method for automatically extracting the cultural relic disease high precision based on the texture as claimed in claim 1, wherein in the third step, the concrete step of fusing the disease areas of the cultural relic together by taking the pixel set as a clustering center through a clustering mode based on information transfer comprises the following steps:
step 9): calculating the color component mean value X of all points in the Lab color space in the super pixel block R (i)R(i)And using the color feature vector as the color feature vector of the super pixel;
step 10): calculating Euclidean distance between any super pixel blocks according to a formula (1) to obtain a similarity matrix S;
step 11): taking the median of the similarity matrix S as an initialization parameter, setting the damping coefficient lambda to 0.5, and enabling the attraction degree and the attribution degree parameters to be 0;
step 12): calculating the attribution degree and the attraction degree according to the formulas (2) to (4):
Figure FDA0003509285200000021
Figure FDA0003509285200000022
Figure FDA0003509285200000023
Figure FDA0003509285200000024
where r (i, j) -represents the attraction;
a (i, j) -represents degree of attribution;
i, j-representing two different sets of superpixel cluster pixels;
s (i, j) — represents the euclidean distance of the non-cluster centers;
k' -represents other cluster centers;
i' -represents data points in other classes;
a (i, k') — representing the degree of attribution to other cluster centers;
s (i, k') — represents euclidean distances from the center points of the other clusters;
r (k, k) -judging the attraction degree between the cluster centers;
r (i', k) — the attractiveness of other cluster data to the current cluster center;
s (i, k) -represents the Euclidean distance from the center of the cluster;
updating the attribution degree and the attraction degree according to formulas (5) - (8):
Figure FDA0003509285200000031
Figure FDA0003509285200000032
Figure FDA0003509285200000033
in the formula rt-1(i, k) and at-1(i, k) -representing the updated attractiveness and attribution after the t-1 th iteration;
lambda represents a damping coefficient, and the convergence speed is higher when the numerical value is larger;
at(k, k) -representing the result of this iteration;
s (i, k') — represents euclidean distances from the center points of the other clusters;
rt-1(k, k) -representing the result of the last iteration of clustering;
step 13): determining the number of cluster centers by the formula (8):
Figure FDA0003509285200000034
if i is not equal to k, k is the number of the clustering centers of i, if the iteration times in the iteration process exceed 1000 times, the iteration is terminated, and the numerical value of the current clustering center is obtained as the number of the clustering centers;
step 14): when the number k of the clustering centers is 2, the foreground and the background of the cultural relic image are segmented, and the segmentation effect is optimal; if the number of cluster centers is not converged to 2, Sil is calculated according to equation (9)R(i)Continuing iteration by reducing the value p by p +0.1min (S), and stopping the clustering algorithm when the average value of the contour coefficients continuously drops for more than 3 times or the number of clustering centers is equal to 2;
Figure FDA0003509285200000041
in the formula, SilR(i)The clustering quality of the AP algorithm is represented, and the larger the value of the clustering quality of the AP algorithm is, the higher the quality of the algorithm clustering result is, and the better the image segmentation effect is;
a (i) -means representing the average of the distance of the superpixel points in the color space;
b (i) -means representing the average of the distance in color space of a superpixel from a superpixel point to a superpixel in another image partition;
step 15): and taking the number of clustering centers corresponding to the maximum value of the contour coefficient average value as a clustering result.
3. The method for high-precision automatic extraction of texture-based cultural relic diseases according to claim 1, wherein the established requirement is that the edge recall rate reaches 90% and above.
4. The method for high-precision automatic extraction of texture-based cultural relic diseases according to claim 3, wherein the established requirement is that the edge recall rate reaches 90%.
5. The texture-based cultural relic disease high-precision automatic extraction method as claimed in claim 1, wherein in the step 1), the compactness parameter takes a value of 30.
6. The method for automatically extracting cultural relic diseases based on textures as claimed in claim 1, wherein in the third step, the disease condition of the cultural relic is calculated statistically, which comprises the area of the disease area and the length of the edge.
7. The texture-based cultural relic disease high-precision automatic extraction method according to claim 1, wherein in the second step, the generation of the orthoscopic image of the cultural relic comprises an all-digital orthoscopic image generation method and a point cloud-based orthoscopic image generation method, and the all-digital orthoscopic image generation method comprises the following steps: acquiring original image data, establishing a Digital Elevation Model (DEM), registering the digital image and the DEM to form a color DEM, and projecting the color DEM onto a horizontal plane to obtain an orthoimage; the orthoimage generation method based on the point cloud comprises the following steps: acquiring point cloud data, constructing a triangulation network model and manufacturing an orthoimage.
8. A high-precision automatic cultural relic disease extraction method based on textures is characterized by comprising the following steps:
an image segmentation module which segments the orthoimage data of the cultural relic by the improved simple linear iterative superpixel segmentation method as claimed in claim 1, and outputs the number of the superpixel clustering centers, the compactness parameter and the image segmentation result data;
a clustering module, connected to the image segmentation module, for clustering results obtained by the improved simple linear iterative superpixel segmentation method using the information-passing AP-based clustering approach of claim 1;
and the disease counting module is connected with the clustering module and is used for counting the cultural relic disease conditions including the area of the disease area and the length of the edge according to the clustering result.
CN201910973573.5A 2019-06-28 2019-10-14 Cultural relic disease high-precision automatic extraction method based on texture Active CN110796181B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2019105778924 2019-06-28
CN201910577892 2019-06-28

Publications (2)

Publication Number Publication Date
CN110796181A CN110796181A (en) 2020-02-14
CN110796181B true CN110796181B (en) 2022-05-03

Family

ID=69439286

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910973573.5A Active CN110796181B (en) 2019-06-28 2019-10-14 Cultural relic disease high-precision automatic extraction method based on texture

Country Status (1)

Country Link
CN (1) CN110796181B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113838061A (en) * 2021-07-28 2021-12-24 中科云谷科技有限公司 Method and device for image annotation and storage medium
CN116051976B (en) * 2022-11-23 2023-09-19 河南理工大学 Processing method of remote sensing image fused with elevation information

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106056155B (en) * 2016-05-30 2019-04-23 西安电子科技大学 Superpixel segmentation method based on boundary information fusion
CN109345582A (en) * 2018-08-14 2019-02-15 西安电子科技大学 The depth image acquisition method of belief propagation algorithm and rarefaction representation based on optimization
CN109919159A (en) * 2019-01-22 2019-06-21 西安电子科技大学 A kind of semantic segmentation optimization method and device for edge image

Also Published As

Publication number Publication date
CN110796181A (en) 2020-02-14

Similar Documents

Publication Publication Date Title
CN110363858B (en) Three-dimensional face reconstruction method and system
CN107516319B (en) High-precision simple interactive matting method, storage device and terminal
CN108038906B (en) Three-dimensional quadrilateral mesh model reconstruction method based on image
CN111652892A (en) Remote sensing image building vector extraction and optimization method based on deep learning
WO2018053952A1 (en) Video image depth extraction method based on scene sample library
CN110796181B (en) Cultural relic disease high-precision automatic extraction method based on texture
CN109389569B (en) Monitoring video real-time defogging method based on improved DehazeNet
CN102982524B (en) Splicing method for corn ear order images
CN112929626B (en) Three-dimensional information extraction method based on smartphone image
CN112270688A (en) Foreground extraction method, device, equipment and storage medium
CN109859249B (en) Scene flow estimation method based on automatic layering in RGBD sequence
CN110544294A (en) dense three-dimensional reconstruction method based on panoramic video
CN111008642A (en) High-resolution remote sensing image classification method and system based on convolutional neural network
CN110390724B (en) SLAM method with instance segmentation
CN115731146A (en) Multi-exposure image fusion method based on color gradient histogram feature light stream estimation
CN111127353B (en) High-dynamic image ghost-removing method based on block registration and matching
CN110517348B (en) Target object three-dimensional point cloud reconstruction method based on image foreground segmentation
CN115222884A (en) Space object analysis and modeling optimization method based on artificial intelligence
CN117611542B (en) Fetal intrauterine craniocerebral image-based detection method and system
WO2021000302A1 (en) Image dehazing method and system based on superpixel segmentation, and storage medium and electronic device
CN113409332B (en) Building plane segmentation method based on three-dimensional point cloud
CN113034681B (en) Three-dimensional reconstruction method and device for spatial plane relation constraint
CN116993947B (en) Visual display method and system for three-dimensional scene
Li et al. Global color consistency correction for large-scale images in 3-D reconstruction
CN115115847B (en) Three-dimensional sparse reconstruction method and device and electronic device

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