CN110555444A - Feature matching screening algorithm based on local clustering - Google Patents
Feature matching screening algorithm based on local clustering Download PDFInfo
- Publication number
- CN110555444A CN110555444A CN201810557874.5A CN201810557874A CN110555444A CN 110555444 A CN110555444 A CN 110555444A CN 201810557874 A CN201810557874 A CN 201810557874A CN 110555444 A CN110555444 A CN 110555444A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- omega
- characteristic points
- points
- clustering
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
- G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
-
- 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
- G06V10/443—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 by matching or filtering
Abstract
The invention relates to a feature matching screening algorithm based on local clustering. The algorithm divides the region of the feature extraction image, counts the number of effective feature points in the region and carries out local approximate clustering statistical processing; and then, carrying out constraint processing on the number of the effective characteristic points, and carrying out characteristic matching screening. The method of the invention deletes the characteristic point pairs with wrong matching connection, reduces the number of characteristic matches, reduces the computational complexity and verifies the effectiveness of the method through related experiments.
Description
Technical Field
The invention relates to the field of image processing, in particular to a feature matching screening algorithm based on local clustering.
background
the feature matching is one of key steps in image splicing, a matching algorithm based on the ratio of the nearest Euclidean distance to the next nearest Euclidean distance often has a large amount of mismatching, and a good screening algorithm can reduce the mismatching rate and improve the processing efficiency, so that the feature matching method has important significance for the research of the algorithm. The RANSAC algorithm is a screening algorithm widely used, but has the defects of uncertain iteration times, unfriendly beam adjustment method BA (bundle adjustment) process and the like.
disclosure of Invention
aiming at the defects in the prior art, the invention provides a brand-new matching screening algorithm (LCMF) based on local clustering based on the local similarity principle of images. The invention aims to solve the problem of screening the feature point pairs after image feature matching, and screening is carried out by utilizing an LCMF algorithm so as to obtain higher screening efficiency and screening results.
The technical scheme adopted by the invention for realizing the purpose is as follows: a feature matching screening algorithm based on local clustering comprises the following steps:
Step 1: loading the matching relation of the characteristic point pairs of the images M1 and M2 after the characteristic extraction and the images M1 and M2;
Step 2, dividing the image M1 into n × n rectangular areas in equal proportion, and counting the number C i of effective characteristic points in each area, wherein the serial number i of the rectangular area partition is 0, and … … n × n-1;
Step 3, performing descending sorting according to the number C i of the effective characteristic points to obtain a partition array Cx of the descending sorting of the effective characteristic points;
and 4, step 4: if the first element Cx [0] < alpha of the array Cx, calling RANSAC algorithm and ending the whole process;
And 5: if Cx 0 is not less than alpha, building cluster point set omega; wherein alpha is a preset threshold value;
step 6: if Cx [0] < beta, go to step 8; wherein beta is a preset threshold value;
and 7: if Cx 0 is not less than beta, thinning each rectangular area of the image and repeating the iteration once again;
And 8: summing up elements in the omega set, setting a deletion proportion y if sigma omega > gamma, and carrying out block random deletion on effective characteristic points; wherein γ is a preset threshold value;
and step 9: rescanning the images M1 and M2, and if the effective feature points in M1 are deleted, deleting the corresponding feature points in M2; returning to the step 2 to perform an iterative algorithm flow on the image M2;
step 10: if the screening is more important than the efficiency, the algorithm is ended; if the algorithm is heavier than the quality, the RANSAC algorithm is called once and then the algorithm process is ended.
The effective characteristic points are characteristic points with matching relations.
the threshold alpha is an empirical value and has a value range of 15-30.
The construction of the cluster point set omega comprises the following steps:
Step 5.1: defining a set omega, and adding Cx [0] in the set omega;
step 5.2: adding Cx [ i +1] within the set Ω if Cx [ i ]/Cx [ i +1] < P and Cx [ i ] ∈ set Ω;
The number of effective characteristic points of the block areas with the clustering characteristics is in the set omega; cx [ i ] is an element in the array Cx; i is the element number of the array Cx, and the value of i is more than or equal to 0 and less than or equal to (n multiplied by n-2); p is a preset threshold value.
the threshold value beta is an empirical value and has a value range of 40-60.
The γ is an empirical value with a value range of 100-150.
the setting of the deletion ratio y includes:
setting a deletion proportion y to enable y to be (beta-1)/beta, and carrying out block random deletion on the effective characteristic points;
And beta is equal to sigma omega/W, sigma omega is the sum of the number of effective characteristic points of each partitioned area in the set omega, and W is the total number of preset characteristic matching point pairs.
if Cx [0] is not less than beta, refining each rectangular region of the image and repeating the iteration once again comprises the following steps:
step 7.1, dividing the equal proportion of the regions represented by the elements in the set omega into m multiplied by m rectangular regions, and counting the number of effective characteristic points C j in each region, wherein the partition serial number j of the rectangular regions is 0, … … m multiplied by k-1, and k is the number of the elements in the omega;
7.2, sorting the data in a descending order according to the number C j of the effective characteristic points to obtain an array Cv;
step 7.3: constructing a clustering point set lambda; if Cv [ i ]/Cv [ i +1] < P and Cx [ i ] ∈ lambda, adding Cv [ i +1] in a set lambda, wherein the set lambda is the number of effective characteristic points of the partitioned areas with the clustering characteristics;
Step 7.4: and updating the original point set omega to be a point set lambda.
The invention has the following beneficial effects and advantages:
1. the invention replaces fitting operation by statistical clustering, and has faster calculation speed.
2. The invention replaces one-time operation with multiple small-scale iterative branch operations, reduces the calculation scale and simplifies the calculation process.
3. The invention utilizes RANSAC to process the situation of too few matching features, and has stronger universality.
4. the invention restrains the quantity of the matching results, selects the threshold beta, gives consideration to the effect and the efficiency and has self-adaption.
5. The method and the device can be used for rapidly screening the feature matching result and have important significance for image optimization processing.
Drawings
FIG. 1 is a flow chart of a feature matching screening algorithm for local clustering according to the method of the present invention;
FIG. 2(a) is a graph of feature point matching relationship before processing by the algorithm of the present invention;
Fig. 2(b) is a feature point matching relationship diagram after the algorithm of the present invention is screened.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
As shown in fig. 1, a feature matching screening algorithm based on local clustering includes the following steps:
step 1: loading the matching relation of the characteristic point pairs of the images M1 and M2 after the characteristic extraction and the images M1 and M2;
Step 2, dividing the image M1 into 3x3 rectangular areas in equal proportion, and counting the number C i of effective characteristic points in each area, wherein the serial number i of the rectangular area partition is 0, … … 8;
Step 3, performing descending sorting according to the number C i of the effective characteristic points to obtain a partition array Cx of the descending sorting of the effective characteristic points;
And 4, step 4: if the first element Cx [0] < alpha of the array Cx, calling RANSAC algorithm and ending the whole process;
And 5: if Cx 0 is not less than alpha, building cluster point set omega;
step 6: if Cx [0] < beta, jump to step 8;
and 7: if Cx 0 is not less than beta, thinning each rectangular area of the image and repeating the iteration once again;
step 7.1, dividing the equal proportion of the regions represented by the elements in the set omega into 3 × 3 rectangular regions, and counting the number of effective characteristic points C j in each region, wherein the partition sequence number j of the rectangular regions is 0, … … 3 × 3 × k-1, and k is the number of the elements in the omega;
7.2, sorting the data in a descending order according to the number C j of the effective characteristic points to obtain an array Cv;
step 7.3: constructing a clustering point set lambda; if Cv [ i ]/Cv [ i +1] < P and Cx [ i ] ∈ lambda, adding Cv [ i +1] in a set lambda, wherein the set lambda is the number of effective characteristic points of the partitioned areas with the clustering characteristics;
Step 7.4: and updating the original point set omega to be a point set lambda. Resulting in an omega set having at most 64 elements.
And 8: summing up elements in the omega set, setting a deletion proportion y if sigma omega > gamma, and carrying out block random deletion on effective characteristic points; wherein γ is a preset threshold value;
and step 9: rescanning M1 and M2, and if the effective characteristic points in M1 are deleted, deleting the corresponding characteristic points in M2; returning to the step 2 to perform an iterative algorithm flow on the image M2;
step 10: if the screening is more important than the efficiency, the algorithm is ended; if the algorithm is heavier than the quality, the RANSAC algorithm is called once and then the algorithm process is ended. The screening bias is pre-set with respect to efficiency or quality.
in step 2, the effective feature points: feature extraction is carried out to obtain a feature point set S of the images, feature matching is carried out to match similar feature points in the two images to form a point pair set M, and effective feature points are defined as feature points contained in M. As shown in fig. 2(a), the end points at both ends of the connecting line are valid feature points, and the other points are invalid points.
In step 4, the threshold α is an empirical value, generally 15 to 30, and represents a threshold of the number of effective points in an area of an image subjected to 3 × 3 cutting, and if the value is too small, the effective points often do not have obvious clustering characteristics, and there is no need for cluster screening. This value was an empirical value, and 20 was chosen in the experiment.
in step 6, beta is an empirical value and represents a threshold value of the number of effective points in a region, if the value is larger, the graph has more effective characteristic points, the effective points have obvious clustering characteristics, and further optimization can be achieved. The value was selected to be 50 in the experiment.
in step 8, feature matching can be solved theoretically only by 4 pairs of feature points, γ represents the number of effective points after commander selection, and as the value increases, the influence on the experimental result is smaller and smaller, but the calculation amount is geometrically increased, so that λ ═ Σ Ω/150, a random factor y is set, the deletion ratio (β -1)/β is set, the effective feature points are deleted randomly in blocks, wherein γ is an empirical value, and 100 is selected as a standard in the experiment.
example (c): FIG. 2(a) is a graph of feature point matching relationship before processing by the algorithm of the present invention; FIG. 2(b) is a graph of the feature point matching relationship after the algorithm of the present invention is screened. It can be seen that: by using the method, the characteristic point pairs with wrong matching connection lines are deleted, the number of characteristic matches is reduced, and the calculation complexity is reduced.
Claims (8)
1. a feature matching screening algorithm based on local clustering is characterized by comprising the following steps:
Step 1: loading the matching relation of the characteristic point pairs of the images M1 and M2 after the characteristic extraction and the images M1 and M2;
Step 2, dividing the image M1 into n × n rectangular areas in equal proportion, and counting the number C i of effective characteristic points in each area, wherein the serial number i of the rectangular area partition is 0, and … … n × n-1;
step 3, performing descending sorting according to the number C i of the effective characteristic points to obtain a partition array Cx of the descending sorting of the effective characteristic points;
And 4, step 4: if the first element Cx [0] < alpha of the array Cx, calling RANSAC algorithm and ending the whole process;
and 5: if Cx 0 is not less than alpha, building cluster point set omega; wherein alpha is a preset threshold value;
step 6: if Cx [0] < beta, go to step 8; wherein beta is a preset threshold value;
and 7: if Cx 0 is not less than beta, thinning each rectangular area of the image and repeating the iteration once again;
And 8: summing up elements in the omega set, setting a deletion proportion y if sigma omega > gamma, and carrying out block random deletion on effective characteristic points; wherein γ is a preset threshold value;
And step 9: rescanning the images M1 and M2, and if the effective feature points in M1 are deleted, deleting the corresponding feature points in M2; returning to the step 2 to perform an iterative algorithm flow on the image M2;
Step 10: if the screening is more important than the efficiency, the algorithm is ended; if the algorithm is heavier than the quality, the RANSAC algorithm is called once and then the algorithm process is ended.
2. The local clustering-based feature matching and screening algorithm according to claim 1, wherein the valid feature points are feature points having a matching relationship.
3. the feature matching and screening algorithm based on local clustering of claim 1, wherein the threshold α is an empirical value with a value range of 15-30.
4. The local clustering-based feature matching screening algorithm of claim 1, wherein the construction of the cluster point set Ω comprises:
Step 5.1: defining a set omega, and adding Cx [0] in the set omega;
step 5.2: adding Cx [ i +1] within the set Ω if Cx [ i ]/Cx [ i +1] < P and Cx [ i ] ∈ set Ω;
The number of effective characteristic points of the block areas with the clustering characteristics is in the set omega; cx [ i ] is an element in the array Cx; i is the element number of the array Cx, and the value of i is more than or equal to 0 and less than or equal to (n multiplied by n-2); p is a preset threshold value.
5. the feature matching and screening algorithm based on local clustering of claim 1, wherein the threshold β is an empirical value with a value range of 40-60.
6. The feature matching screening algorithm based on local clustering as claimed in claim 1, wherein said γ is an empirical value with a value range of 100-150.
7. the local clustering based feature matching screening algorithm of claim 1, wherein the setting of the deletion ratio y comprises:
setting a deletion proportion y to enable y to be (beta-1)/beta, and carrying out block random deletion on the effective characteristic points;
and beta is equal to sigma omega/W, sigma omega is the sum of the number of effective characteristic points of each partitioned area in the set omega, and W is the total number of preset characteristic matching point pairs.
8. the local clustering based feature matching screening algorithm of claim 1, wherein if Cx [0] β ≧ β, refining each rectangular region of the image for a new iteration comprises:
Step 7.1, dividing the equal proportion of the regions represented by the elements in the set omega into m multiplied by m rectangular regions, and counting the number of effective characteristic points C j in each region, wherein the partition serial number j of the rectangular regions is 0, … … m multiplied by k-1, and k is the number of the elements in the omega;
7.2, sorting the data in a descending order according to the number C j of the effective characteristic points to obtain an array Cv;
Step 7.3: constructing a clustering point set lambda; if Cv [ i ]/Cv [ i +1] < P and Cx [ i ] ∈ lambda, adding Cv [ i +1] in a set lambda, wherein the set lambda is the number of effective characteristic points of the partitioned areas with the clustering characteristics;
step 7.4: and updating the original point set omega to be a point set lambda.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810557874.5A CN110555444B (en) | 2018-06-01 | 2018-06-01 | Feature matching screening algorithm based on local clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810557874.5A CN110555444B (en) | 2018-06-01 | 2018-06-01 | Feature matching screening algorithm based on local clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110555444A true CN110555444A (en) | 2019-12-10 |
| CN110555444B CN110555444B (en) | 2022-09-20 |
Family
ID=68735363
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810557874.5A Active CN110555444B (en) | 2018-06-01 | 2018-06-01 | Feature matching screening algorithm based on local clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110555444B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111507991A (en) * | 2020-04-20 | 2020-08-07 | 西安邮电大学 | Method and device for segmenting remote sensing image of characteristic region |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819835A (en) * | 2012-07-26 | 2012-12-12 | 中国航天科工集团第三研究院第八三五七研究所 | Method for screening matching pairs of feature points to splice images |
| CN106529573A (en) * | 2016-10-14 | 2017-03-22 | 北京联合大学 | Real-time object detection method based on combination of three-dimensional point cloud segmentation and local feature matching |
| WO2017181892A1 (en) * | 2016-04-19 | 2017-10-26 | 中兴通讯股份有限公司 | Foreground segmentation method and device |
| CN107578376A (en) * | 2017-08-29 | 2018-01-12 | 北京邮电大学 | The fork division of distinguished point based cluster four and the image split-joint method of local transformation matrix |
-
2018
- 2018-06-01 CN CN201810557874.5A patent/CN110555444B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819835A (en) * | 2012-07-26 | 2012-12-12 | 中国航天科工集团第三研究院第八三五七研究所 | Method for screening matching pairs of feature points to splice images |
| WO2017181892A1 (en) * | 2016-04-19 | 2017-10-26 | 中兴通讯股份有限公司 | Foreground segmentation method and device |
| CN106529573A (en) * | 2016-10-14 | 2017-03-22 | 北京联合大学 | Real-time object detection method based on combination of three-dimensional point cloud segmentation and local feature matching |
| CN107578376A (en) * | 2017-08-29 | 2018-01-12 | 北京邮电大学 | The fork division of distinguished point based cluster four and the image split-joint method of local transformation matrix |
Non-Patent Citations (2)
| Title |
|---|
| 薛佳乐等: "针对大视差图像拼接的显性子平面配准", 《中国图象图形学报》 * |
| 赵春阳等: "多模态鲁棒的局部特征描述符", 《光学精密工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111507991A (en) * | 2020-04-20 | 2020-08-07 | 西安邮电大学 | Method and device for segmenting remote sensing image of characteristic region |
| CN111507991B (en) * | 2020-04-20 | 2023-03-21 | 西安邮电大学 | Method and device for segmenting remote sensing image of characteristic region |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110555444B (en) | 2022-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8117609B2 (en) | System and method for optimizing changes of data sets | |
| CN109165215A (en) | Construction method, device and the electronic equipment of temporal index under a kind of cloud environment | |
| JPH02227725A (en) | Zoning of list sorted | |
| US20220005546A1 (en) | Non-redundant gene set clustering method and system, and electronic device | |
| CN110555444B (en) | Feature matching screening algorithm based on local clustering | |
| Arya et al. | A simple entropy-based algorithm for planar point location | |
| CN110097581B (en) | Method for constructing K-D tree based on point cloud registration ICP algorithm | |
| CN108509532B (en) | Point gathering method and device applied to map | |
| CN105988995A (en) | HFile based data batch loading method | |
| CN111384972A (en) | Optimization method and device of multi-system LDPC decoding algorithm and decoder | |
| CN107562762A (en) | Data directory construction method and device | |
| CN109710633B (en) | Method and device for determining man-in-middle information and intelligent terminal | |
| CN111915091A (en) | Parking space arrangement method, device, terminal and medium | |
| CN104794129A (en) | Data processing method and system based on query logs | |
| CN116502595A (en) | Random walking-based capacitance extraction method and device, equipment and medium | |
| CN108521376B (en) | Flow table design method based on attribute similarity in software defined network | |
| CN111177190A (en) | Data processing method and device, electronic equipment and readable storage medium | |
| CN111859040B (en) | Data matching method, device and related equipment | |
| CN110599568B (en) | Line generation method, device, equipment and storage medium | |
| CN109828894B (en) | Equipment state data acquisition method and device, storage medium and electronic equipment | |
| CN111782645A (en) | Data processing method and device | |
| WO2018196246A1 (en) | Fast sinc interpolation method and system | |
| CN111224674A (en) | Decoding method, device and decoder of multi-system LDPC code | |
| CN116882849B (en) | Method and device for measuring and calculating yield of clean coal | |
| CN115409744A (en) | Method for realizing assembly line of image recursive median filtering |
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 |