CN107679131A - A kind of quick spectrogram matching process - Google Patents
A kind of quick spectrogram matching process Download PDFInfo
- Publication number
- CN107679131A CN107679131A CN201710863532.1A CN201710863532A CN107679131A CN 107679131 A CN107679131 A CN 107679131A CN 201710863532 A CN201710863532 A CN 201710863532A CN 107679131 A CN107679131 A CN 107679131A
- Authority
- CN
- China
- Prior art keywords
- matrix
- matching
- value
- initializing
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
Abstract
A kind of quick spectrogram matching process of the disclosure of the invention, the figure matching field being related in computer vision, more particularly to a kind of spectrogram matching process of quick high accuracy.Method space complexity proposed by the present invention is only O (n1n2), time complexity is O (kn1n2), wherein k is the number of basic matrix;Normalized by increasing expansion factor Y and ranks, accelerate the speed of algorithmic statement, effectively inhibit the interference of noise and exterior point for figure matching, improve the precision of figure matching.The present invention also has the advantages of matching speed is fast, reliability is high, cost is low, the saving energy.
Description
Technical Field
The invention relates to the field of graph matching in computer vision, in particular to a fast high-precision spectrogram matching method.
Background
In the field of computer vision, graph matching can be used for solving the problem of feature corresponding points, and has wide application in image retrieval based on geometric shapes, target recognition, shape matching, target tracking and the like. Simply speaking, a graph is a topological structure of points and lines, and graph matching is to find the correspondence between points and points in two graphs. Spectrogram Matching is an "art-level" method proposed by Leordeanu and Hebert in 2005, known as Spectrogram Matching (SM). They model the graph matching problem for the corresponding points as a quadratic distribution problem. As the quadratic distribution problem is an NP difficult problem, in the solving process, duchenne and the like relax spectrogram distribution into a problem of eigenvector corresponding to the maximum eigenvalue, namely solving the eigenvector corresponding to the maximum eigenvalue of two graph similarity matrixes, and then performing HungaryThe odonglis algorithm performs 0-1 transformation to obtain the matching relation of points. However, the biggest disadvantage of the above-described spectrogram-based matching method is the need to construct O (n) 4 ) And thus are not suitable for solving medium or ultra-large scale graph matching problems. To reduce the storage space of the similarity matrix, kang et al propose to approximate the similarity matrix as a series of O (n) 2 ) The sum form of the kronecker product of the sparse basis matrix and the index matrix is called the FaSM method. Although this approximate expression method does not show the construction of a similarity matrix, in the calculation process thereof, O (n) needs to be constructed 4 ) Similar matrix is approximated, and because of the approximation of the matrix, the similar matrix has information loss, which causes the precision of the similar matrix to be reduced compared with the precision of spectrogram matching.
Disclosure of Invention
The invention aims to solve the problem that a spectrogram matching method is not suitable for solving the medium or large-scale spectrogram matching problem, and provides a fast high-precision spectrogram matching method. The proposed method requires only O (n) 2 ) Spatial complexity of O (kn) 2 ) The time complexity of the method is reduced, the interference of noise and external points on spectrogram matching can be effectively inhibited, and the spectrogram matching precision is improved.
The technical scheme of the invention is a rapid spectrogram matching method, which comprises the following steps:
step 1: known diagram G 1 And graph G 2 The set of the vertex and the edge of (b) is (V) 1 ,E 1 ) And (V) 2 ,E 2 ) The number of vertexes is n 1 ,n 2 (ii) a Set up the drawing G 1 The distance between the ith vertex and the jth vertex isDrawing G 2 The distance between the ith vertex and the jth vertex isFor picture G 1 Is D as a distance matrix between any two points in the array 1 To D, to 1 The middle elements are sorted, assuming the maximum value is d max Minimum value of d min D is mixing d max And d min M segments are divided into m segments, and the width w of each segment is = (d) max -d min ) (ii)/m; to d is paired ij The approximate expression is carried out, and the expression method is as follows:
d 'is judged' ij Which segment of the m segments to locate, and then taking the median d 'of that segment' k Replaces the original d' ij ;
And 2, step: constructing a base matrix B k And index matrix H k (ii) a Initialization H k Is n 1 xn 1 K is [0, m-1 ]]M is preset according to the approximate precision required;
step 2.1 setting B k =4.5-(d′ k -D 2 ) 2 /2σ 2 Wherein: d 2 Represents G 2 Distance matrix between vertices of (1), σ 2 Represents an adjustable factor;
step 2.2 search for G 1 Distance matrix D of 1 Element value between [ d min +w×(k-1),d min +w×k]Of elements between, let H k The corresponding index element in (1);
and step 3: initializing a matrixInitializing M to n 1 ×n 2 All-zero matrix of (2); let X = X 0 (ii) a Initializing an Error threshold Error =1, initializing a maximum iteration number ItersMax, and initializing alpha, beta and p;
and 4, step 4: calculating an initial matching matrix X;
step 4.1, take k as [0, m-1 ]]Is sequentially repeatedly calculatedCalculate M m
Step 4.2, assignment calculation:Y=e βX/max(X) ;
and 4.3, normalizing the rows and columns of the Y, namely repeatedly calculating the assignment:
up to Y k -Y k+1 || 2 <1e-25;Y aj Represents the element of the a-th row and the j-th column in Y, and the Y value is Y k+1 ;
Step 4.4, assigning alpha X + (1-alpha) Y to X, and then normalizing the X after assignment, namely, repeating assignment calculation
Up to | X | k -X k+1 || 2 <1e-25,X aj Represents the element in row a and column j in X, where X has the value X k+1 ;
Step 4.5, calculate Error = | | X p -X p-1 || 2 ,X p Representing the value of X, calculated for the current cycle p-1 Representing the X value calculated in the previous cycle, reassigning the calculation p = p +1;
step 4.6, if Error >1e-25 and p < ItersMax, returning to step 4.1, otherwise, saving the value of X at the moment, and entering the next step;
step 5, performing 0-1 discretization on the solved X by using Hungarian algorithm to obtain n 1 ×n 2 The matching matrix represents the corresponding relationship of the vertexes in the two images.
Compared with the prior art, the space complexity of the method provided by the invention is only O (n) 1 n 2 ) Time complexity of O (k.n) 1 n 2 ) Which isThe middle k is the number of the base matrixes; by increasing the expansion factor Y and row-column normalization, the convergence speed of the algorithm is accelerated, the interference of noise and outliers on the image matching is effectively inhibited, and the accuracy of the image matching is improved. In the specific embodiment, different noises or outliers are added to the graph, as shown in fig. 5 and fig. 6, which illustrates the advantages of the invention, such as fast matching speed, high reliability, low cost and energy saving.
Drawings
FIG. 1 randomly generated graph G 1 And G 2 A schematic diagram;
FIG. 2 is a comparison test with outliers added to the generated data, in SM, faSM;
FIG. 3 is a comparison test with noise added to the generated data, in SM, faSM;
FIG. 4 is an example of a real graph;
figure 5 comparative test results on real graphical examples. (a) Is the matching precision comparison result of the proposed method and FaSM; (b) Is the run-time comparison of the proposed method with the FaSM;
figure 6 outlier comparison test on real data. (a) The method is a comparison result of matching accuracy under the condition of adding different proportions of outer points and fixed noise (sigma = 50); (b) Is a comparison of the run time of the proposed method with the addition of different proportions of outliers and fixed noise (σ = 50).
Detailed Description
A fast spectrogram matching method, the method comprising:
step 1: known graph G 1 And graph G 2 The set of the vertex and the edge of (b) is (V) 1 ,E 1 ) And (V) 2 ,E 2 ) The number of vertexes is n 1 ,n 2 (ii) a Set up the drawing G 1 The distance between the ith vertex and the jth vertex isDrawing G 2 The distance between the ith vertex and the jth vertex isFor graph G 1 A distance matrix between any two points in (1) is D 1 To D, pair 1 The middle elements are sorted, assuming the maximum value is d max Minimum value of d min D is mixing d max And d min M segments between, the width w of each segment = (d) max -d min ) (ii)/m; to d ij The approximate expression is carried out, and the expression method is as follows:
d 'is judged' ij Which segment of the m segments to locate, and then taking the median d 'of that segment' k Instead of the original d' ij ;
Step 2: constructing a base matrix B k And index matrix H k (ii) a Initialization H k Is n 1 xn 1 Is k is [0, m-1 ]]M =11;
step 2.1 setting B k =4.5-(d′ k -D 2 ) 2 /2σ 2 Wherein: d 2 Represents G 2 Distance matrix between vertices of (1), σ 2 Represents an adjustable factor;
step 2.2 search for G 1 Distance matrix D of 1 Element value between [ d min +w×(k-1),d min +w×k]Element (b) in between, let H k The corresponding index element in (1);
and step 3: initializing a matrixInitializing an all-zero matrix with M being n1 multiplied by n 2; let X = X 0 (ii) a Initializing an Error threshold Error =1, and initializing a maximum iteration number ItersMax =300; initializing α =0.3, β =30, p =0;
and 4, step 4: calculating an initial matching matrix X;
step 4.1, take k as [0, m-1 ]]Is sequentially repeatedly calculatedCalculate M m
Step 4.2, assignment calculation:Y=e βX/max(X) ;
and 4.3, normalizing the rows and columns of the Y, namely repeatedly calculating the assignment:
up to Y k -Y k+1 || 2 <1e-25;Y aj Represents the element of the a-th row and the j-th column in Y, and the Y value is Y k+1 ;
Step 4.4, assigning alpha X + (1-alpha) Y to X, and then normalizing the X with the column after assignment, namely, repeatedly assigning value calculation
Up to | X | k -X k+1 || 2 <1e-25,X aj Represents the element in row a and column j in X, where X has the value X k+1 ;
Step 4.5, calculate Error = | | X p -X p-1 || 2 ,X p Representing the value of X, calculated for the current cycle p-1 Representing the X value calculated in the previous cycle, reassigning the calculation p = p +1;
step 4.6, if Error >1e-25 and p < ItersMax, returning to step 4.1, otherwise, saving the value of X at the moment, and entering the next step;
step 5, performing the X of the solution by using Hungarian algorithm0-1 discretization, thereby obtaining n 1 ×n 2 The matching matrix represents the corresponding relationship of the vertexes in the two images.
In the embodiment, the randomly generated image and the real image are selected for image matching. In the generation of graph or real graph, the experiment runs 20 times to randomly add noise and outlier to generate G 2 And calculating the average matching precision and the operation time. FIG. 1 is a randomly generated example to be matched implementing the efficient graph matching method of the present invention. The red dot in FIG. 1 is G 1 A point of (1); by adding different proportions of noise and outliers, G is formed 2 A dot of (1), i.e., a dot of green. G 1 Obeys a normal distribution of N (0, 30). Wherein outliers and noise are randomly generated; noise addition to G 2 The distance between a point and a point. Fig. 3 is an actual graphic example. FIG. 2 shows G generated by adding 0-20 outliers to the generated data 2 And comparing the matching result with the matching result of the SM and the FaSM. FIG. 3 shows the results of matching by adding noise that varies from 10-100 in generation to the generated data, and comparing with the matching results for SM, faSM. The edges of the real object clock in FIG. 4 are extracted first and down-sampled to 200 to 1000 points. Adding noise and outliers is similar to generating a graph. FIG. 5 shows the comparison result of matching precision and runtime of graph matching by randomly adding 10-100 unequal outliers when the number of graph vertices is 200. Fig. 6 shows the comparison results under the condition that 10% to 50% of the outliers are added to the fixed noise condition (σ = 50) when the number of vertices of the real figure is 300 to 1000, respectively.
In summary, the method of the present invention does not require the construction of n 1 n 2 ×n 1 n 2 The spatial complexity of the similarity matrix of (2) is only O (n) 1 n 2 ) Time complexity of O (k.n) 1 n 2 ) (ii) a And meanwhile, an expansion factor and a row-column normalization processing method are integrated, so that the running speed of the algorithm is accelerated, and the accuracy of the graph matching is improved.
Claims (1)
1. A fast spectrogram matching method, the method comprising:
step 1: known graph G 1 And graph G 2 The set of the vertex and the edge of (b) is (V) 1 ,E 1 ) And (V) 2 ,E 2 ) The number of vertexes is n 1 ,n 2 (ii) a Set up the drawing G 1 The distance between the ith vertex and the jth vertex isDrawing G 2 The distance between the ith vertex and the jth vertex isFor graph G 1 A distance matrix between any two points in (1) is D 1 To D, pair 1 The middle elements are sorted, assuming the maximum value is d max Minimum value of d min D is mixing max And d min M segments between, the width w of each segment = (d) max -d min ) (ii)/m; to d ij The approximate expression is carried out, and the expression method is as follows:
d 'is judged' ij Which segment of the m segments to locate, and then taking the median d 'of that segment' k Instead of the original d' ij ;
And 2, step: constructing a base matrix B k And index matrix H k (ii) a Initialization H k Is n 1 xn 1 K is [0, m-1 ]]M is preset according to the approximate precision required;
step 2.1 setting B k =4.5-(d′ k -D 2 ) 2 /2σ 2 Wherein: d 2 Represents G 2 Distance matrix between vertices of (1), σ 2 Represents an adjustable factor;
step 2.2 search for G 1 Distance matrix D of 1 Element value between [ d min +w×(k-1),d min +w×k]BetweenOf (a) is H k The corresponding index element in (1);
and step 3: initializing a matrixInitializing M to n 1 ×n 2 All-zero matrix of (2); let X = X 0 (ii) a Initializing an Error threshold Error =1, initializing a maximum iteration number ItersMax, and initializing alpha, beta and p;
and 4, step 4: calculating an initial matching matrix X;
step 4.1, take k as [0, m-1 ]]Is sequentially repeatedly calculatedCalculate M m
Step 4.2, assignment calculation:Y=e βX/max(X) ;
and 4.3, normalizing the rows and columns of the Y, namely repeatedly calculating the assignment:
up to | | Y k -Y k+1 || 2 <1e-25;Y aj The element in row a and column j in Y is represented, and the Y value is Y k+1 ;
Step 4.4, assigning alpha X + (1-alpha) Y to X, and then normalizing the X after assignment, namely, repeating assignment calculation
Up to | | | X k -X k+1 || 2 <1e-25,X aj Represents the element in row a and column j in X, where X has the value X k+1 ;
Step 4.5, meterCalculation of Error = | | X p -X p-1 || 2 ,X p Representing the value of X, calculated for the current cycle p-1 Representing the X value calculated in the previous cycle, reassigning the calculation p = p +1;
step 4.6, if Error >1e-25 and p < ItersMax, returning to step 4.1, otherwise, saving the value of X at the moment, and entering the next step;
step 5, performing 0-1 discretization on the solved X by using Hungarian algorithm to obtain n 1 ×n 2 The matching matrix represents the corresponding relationship of the vertexes in the two images.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710863532.1A CN107679131B (en) | 2017-09-22 | 2017-09-22 | A kind of quick spectrogram matching process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710863532.1A CN107679131B (en) | 2017-09-22 | 2017-09-22 | A kind of quick spectrogram matching process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107679131A true CN107679131A (en) | 2018-02-09 |
CN107679131B CN107679131B (en) | 2019-10-01 |
Family
ID=61137802
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710863532.1A Expired - Fee Related CN107679131B (en) | 2017-09-22 | 2017-09-22 | A kind of quick spectrogram matching process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107679131B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110334433A (en) * | 2019-07-03 | 2019-10-15 | 电子科技大学 | A kind of PCB package file automatic generation method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103514606A (en) * | 2013-10-14 | 2014-01-15 | 武汉大学 | Heterology remote sensing image registration method |
CN104268896A (en) * | 2014-10-27 | 2015-01-07 | 武汉大学 | Hyper spectrum dimensionality reduction matching method and system based on spectrum sampling histogram |
US9213030B2 (en) * | 2008-01-28 | 2015-12-15 | National University Of Singapore | Lipid tumour profile |
CN105740890A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | Method for solving secondary diagram matching model |
-
2017
- 2017-09-22 CN CN201710863532.1A patent/CN107679131B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9213030B2 (en) * | 2008-01-28 | 2015-12-15 | National University Of Singapore | Lipid tumour profile |
CN103514606A (en) * | 2013-10-14 | 2014-01-15 | 武汉大学 | Heterology remote sensing image registration method |
CN104268896A (en) * | 2014-10-27 | 2015-01-07 | 武汉大学 | Hyper spectrum dimensionality reduction matching method and system based on spectrum sampling histogram |
CN105740890A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | Method for solving secondary diagram matching model |
Non-Patent Citations (2)
Title |
---|
WANG, PF 等: "MULTI-SUBGRAPH MATCHING FOR LOGO LOCALIZATION USING GENETIC ALGORITHM", 《PROCEEDINGS OF 2016 INTERNATIONAL CONFERENCE ON WAVELET ANALYSIS AND PATTERN RECOGNITION (ICWAPR)》 * |
刘志忠: "基于谱图理论的结构描述子及其在点模式匹配中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110334433A (en) * | 2019-07-03 | 2019-10-15 | 电子科技大学 | A kind of PCB package file automatic generation method |
CN110334433B (en) * | 2019-07-03 | 2022-03-15 | 电子科技大学 | Automatic generation method for PCB (printed circuit board) packaging file |
Also Published As
Publication number | Publication date |
---|---|
CN107679131B (en) | 2019-10-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10229092B2 (en) | Systems and methods for robust low-rank matrix approximation | |
US10782396B2 (en) | Variational track management | |
Zheng et al. | Robust point matching for nonrigid shapes by preserving local neighborhood structures | |
Zaslavskiy et al. | A path following algorithm for the graph matching problem | |
US10713566B2 (en) | Efficient calculations of negative curvature in a hessian free deep learning framework | |
KR101581112B1 (en) | Method for generating hierarchical structured pattern-based descriptor and method for recognizing object using the descriptor and device therefor | |
US10762425B2 (en) | Learning affinity via a spatial propagation neural network | |
JP2015197702A (en) | Information processor and information processing method | |
CN108171258B (en) | Independent component analysis multi-shape prior level set method and image segmentation system | |
US20210173895A1 (en) | Apparatus and method of performing matrix multiplication operation of neural network | |
WO2019199307A1 (en) | Second-order optimization methods for avoiding saddle points during the training of deep neural networks | |
Yu et al. | Local part chamfer matching for shape-based object detection | |
CN111983927A (en) | Novel maximum entropy ellipsoid collective filtering method | |
CN105389343A (en) | Vectorized dimension reduction method | |
US20150277435A1 (en) | Computing apparatus and computing method | |
Sahloul et al. | An accurate and efficient voting scheme for a maximally all-inlier 3D correspondence set | |
CN107679131A (en) | A kind of quick spectrogram matching process | |
US20050273447A1 (en) | Support vector classification with bounded uncertainties in input data | |
Dan et al. | Multifeature energy optimization framework and parameter adjustment-based nonrigid point set registration | |
Panaganti et al. | Robust nonrigid point set registration using graph-laplacian regularization | |
CN113670253B (en) | Space target attitude inversion method and device, computing equipment and storage medium | |
Dapogny et al. | Face alignment with cascaded semi-parametric deep greedy neural forests | |
Wang et al. | Point cloud registration algorithm based on Cauchy mixture model | |
Fan et al. | Star skeleton for human behavior recognition | |
Maier et al. | Fast interpolation of dense motion fields from synthetic phantoms |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191001 |