CN104680132B - A kind of sonar target recognition methods based on Shape context method - Google Patents
A kind of sonar target recognition methods based on Shape context method Download PDFInfo
- Publication number
- CN104680132B CN104680132B CN201510054144.XA CN201510054144A CN104680132B CN 104680132 B CN104680132 B CN 104680132B CN 201510054144 A CN201510054144 A CN 201510054144A CN 104680132 B CN104680132 B CN 104680132B
- Authority
- CN
- China
- Prior art keywords
- target
- context
- histogram
- value
- histograms
- 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
Links
Abstract
The invention discloses a kind of sonar target recognition methods based on Shape context method, including the following steps：Sample image statistics is carried out, is obtained using Shape context method in sample image per classification two context histograms of target；Images to be recognized is read, two context histograms of target to be identified are obtained using Shape context method；It will respectively be matched per the value of the corresponding color lump of classification two context histograms of target in the two context histograms and sample image of target to be identified, target identification carried out using maximum matching degree.Wherein Shape context method point on the basis of objective contour major axis end points.The present invention can improve the matching degree of target identification, and can reduce amount of calculation.
Description
Technical field
The invention belongs to sonar target to identify field, more particularly to the context by matching characteristic target and target to be measured
Histogram carries out target identification, a kind of sonar target recognition methods based on Shape context method.
Background technology
Shape context method is a kind of image outline characteristic extraction method, it lay particular emphasis on extraction profile in datum mark and its
The position relationship that he orders, can embodying on the basis of the profile point of selection point, profile, other are special relative to the structure of datum mark
Sign, it is a kind of feature extracting method for characterizing target macro contours, because it has the change to objective contour part insensitive
Property, therefore it is applied to the continually changing sonar image feature extraction of local configuration.
Shape context method is used under optical imagery, wherein Belnogie etc. proposes in shape by most researchers
The shape bidding documents method of text, abundant description containing " context " information is provided for the point in shape edges；Mori etc. is carried
A kind of generalized shape contextual feature is gone out, sampled point is replaced using the tangent vector of each sampled point；Roman-Rangel etc. is carried
Go out direction histogram Shape context feature, introduce the shape that histogram describes target；Domestic Han Min et al. proposes mould
The method for pasting Shape context, for improving matching precision.
The datum mark choosing method of traditional Shape context method can be classified as three classes：The first kind is according to local configuration
Semanteme determines, but the influence of stand under load displacement body and seanoise, in adjacent moment, even being caught under equal angular and distance
The similar target image obtained, its local configuration details is all more big changes, therefore first kind method can not reliably determine base
On schedule；Second class is to be traveled through all profile points all as datum mark, and this method amount of calculation for determining datum mark is very
Greatly, it is not suitable for needing the sonar requirement for considering power consumption considerations.3rd class is the improvement of traversal method, main thought be by
Profile point is divided into some regions, counts the profile point feature in each region, and final or needs circulate matching.Such as document 1：
Han Min, Zheng Danchen, based on the shape recognition algorithm of blurred shapes contextual feature, automate journal, Vol.38, No.1,201.
What is inquired into is optical imagery, is given a kind of " shape matching method based on cyclic shift ", test mesh fixed first during matching
B is marked, then target A is carried out to the rotation of limited number of time, finally find optimal circulation matching result.The essence of this method is exactly
Profile point sample and searching loop, the interval of sampling are the angles of rotation.Its shortcoming be rotation when do not have purpose and
Flexibility, amount of calculation is very big during matching.Document 2：Hu Zhengping, Gao Yanan, the target based on the description of angle spread Shape context
Detection algorithm is studied, signal transacting, Vol.26, No.6, and 2010.A kind of " description of angle spread Shape context " is proposed,
Timing determines the center of objective contour first, and profile point is divided into some regions, then counts the profile point of same area, most
Eventually circulation matching is carried out by region.This method is similar with the method for document [1], and shortcoming is apparent.Document 3：Tian Xiaodong, Liu
Loyalty, the Underwater Targets Recognition algorithm based on shape similarity, acoustic technique, Vol.26, No.3,2007.Matching problem is summed up
Divide for the contour area of image, segmented equivalent to by the profile point in document 2 in region, although improving matching precision,
Amount of calculation is bigger.
The content of the invention
It is an object of the invention to provide a kind of matching precision is high, amount of calculation is small, the sonar based on Shape context method
Target identification method.
The present invention is achieved by the following technical solutions：
A kind of sonar target recognition methods based on Shape context method, including following steps：
Step 1：Sample image statistics is carried out, is obtained using Shape context method in sample image per classification target two
Individual context histogram；
Step 2：Images to be recognized is read, two contexts that target to be identified is obtained using Shape context method are straight
Fang Tu；
Step 3：By every classification two contexts of target in the two context histograms and sample image of target to be identified
The value of the corresponding color lump of histogram is matched respectively, and target identification is carried out using maximum matching degree；
Described Shape context method is：
A, sample image or images to be recognized are pre-processed, the bianry image of target area is partitioned into, by opening
Target area inside aperture is filled up in closed operation；
B, the profile point of target area is extracted；
C, the distance between any two profile point is calculated, takes the major axis that maximum is target area profile, the two of major axis
Individual end points is respectively first end point P1With the second end points P2；
D, with first end point P1On the basis of point, using the length of major axis as the radius of grating, build with P1On the basis of point pole sit
Mark grating；Similarly build with P2On the basis of put polar coordinates grating；
E, profile point is mapped to corresponding grating region by two polar coordinates gratings, obtains two context Nogatas
Figure, the value of context histogram color lump are the profile points included in corresponding grating region.
A kind of sonar target recognition methods based on Shape context method of the present invention, can also include：
1st, by every classification two context Nogatas of target in the two context histograms and sample image of target to be identified
The method that the value of the corresponding color lump of figure is matched respectively is：
Step 1：It is two benchmark context histograms to make every two context histograms of classification target in sample image,
Benchmark context histogram and the context histogram of target to be identified are subjected to binary conversion treatment, even the color lump value for having value is set
For 1, no value color lump value is set to 0, obtains benchmark binaryzation histogram and the binaryzation histogram of target to be identified；
Step 2：By every classification two benchmark two of target in the two binaryzation histograms and sample image of target to be identified
The corresponding color lump value of value histogram is matched, and the matching degree of target is：
Wherein, NMatchingBe matching color lump number, NBenchmarkOn the basis of have that value color lump is total in binaryzation histogram, NTargetTo treat
Surveying the binaryzation histogram of target has value color lump total.
2nd, the method for structure polar coordinates grating is：Using datum mark as origin, long axis length is radius, and long axis direction is pole
0 ° of direction of reference axis, takes ± 90 ° of angular ranges as grating, and the angular range of grating is divided into N number of section, by major axis etc.
Spacing is divided into M section.
Beneficial effect：
The present invention devises a kind of Shape context datum mark choosing method suitable for sonar image target identification, and gives
The matching process based on context histogram is gone out, the characteristic point of this method extraction has stability and uniqueness, during matching
Searching loop is not needed, not only increases matching precision, and reduce amount of calculation.The thought of realization is to utilize sonar image
The dimensional information of middle target determines datum mark, that is, chooses two end points of objective contour major axis respectively as Shape context
Datum mark, i.e. datum mark share two, and the position relationship by counting remaining profile point and datum mark obtains context Nogata
Figure.When entering line raster mapping under polar coordinates, chosen due to the parameter of grating and determined by major axis, therefore this method has rotation not
Denaturation and adaptivity.
Brief description of the drawings
Fig. 1 is sonar image to be identified；
Fig. 2 is target area bianry image；
Fig. 3 is objective contour and major axis end points；
Fig. 4 is the corresponding method of grating and objective contour；
Fig. 5 (a) is the context histogram obtained when being put on the basis of an end points of major axis, and Fig. 5 (b) is the another of major axis
The context histogram obtained when being put on the basis of individual end points；
Fig. 6 is the corresponding relation of context histogram and raster pattern；
Fig. 7 (a) is first kind propeller sample 1, and Fig. 7 (b) is a context Nogata of first kind propeller sample 1
Figure, Fig. 7 (c) are another context histogram of first kind propeller sample 1；
Fig. 8 (a) is first kind propeller sample 2, and Fig. 8 (b) is a context Nogata of first kind propeller sample 2
Figure, Fig. 8 (c) are another context histogram of first kind propeller sample 2；
Fig. 9 (a) is the second class propeller sample 1, and Fig. 9 (b) is a context Nogata of first kind propeller sample 1
Figure, Fig. 9 (c) are another context histogram of first kind propeller sample 1；
Figure 10 (a) is the second class propeller sample 2, and Figure 10 (b) is a context Nogata of first kind propeller sample 2
Figure, Figure 10 (c) are another context histogram of first kind propeller sample 2；
Figure 11 is context histogram and corresponding binaryzation template；
Figure 12 is the flow chart of the present invention.
Embodiment
The present invention is described in further details below in conjunction with accompanying drawing.
As shown in figure 12, a kind of sonar target recognition methods based on Shape context method of the present invention, including it is following several
Individual step：(1) images to be recognized is pre-processed, segmentation obtains the bianry image of target area, is filled up by opening and closing operation
Target area inside aperture.
(2) profile point of target area is extracted, profile point is carried out in various degree according to precision and processing speed demand
Sampling.
(3) major axis of profile is obtained by traversal, records two extreme coordinates P of major axis1And P2。
(4) respectively with P1And P2Point on the basis of point, polar coordinates grating is built according to objective contour major axis.
(5) sampled point on profile is mapped to by semicircle grating by corresponding section, obtains context histogram, on
Hereafter the value of histogram color lump represents the profile point number included in specific region.
Two benchmark context histograms in target sample storehouse and two context histograms of target to be identified are corresponding
Color lump is matched respectively, and target identification is carried out according to maximum matching degree.
Images to be recognized as shown in figure 1, be intercepted out from sonar image contain mesh target area.It is special carrying out target
Before sign extraction, first have to carry out view picture sonar image preliminary pretreatment.Determine the approximate region of target, and by it from whole
Interception comes out in width sonar image.Specific method may be referred to the pertinent literature of sonar image pretreatment.
1st, Target Segmentation：Images to be recognized is further processed, it is therefore an objective to remove ambient noise for target area
Influence, and utilize image partition method, we employ maximum between-cluster variance thresholding method, obtain target area binary map
Picture, as shown in Figure 2.
2nd, the profile of target area after extraction is split：If profile point quantity is very big, profile point is sampled first, subtracted
Few amount of calculation, it is done so that the statistical accuracy of context histogram can be reduced, as shown in Figure 3.
3rd, profile major axis is found：Distance on comparison object profile between each two sampled point, by comparing to obtain major axis, and
Mark two end points P of major axis1And P2。
4th, raster coordinate system is established：Establish respectively with P1And P2On the basis of the raster coordinate system put.As shown in figure 4, by P1Point
As datum mark, the radius of the length of major axis as grating, major axis is divided into 8 apart from section；If long axis direction is sat for pole
0 ° of direction is marked, ± 90 ° of angular ranges as grating is taken, 180 ° of scope is divided into 6 angular intervals, apart from section and angle
The number for spending section can be according to different cases classifications into the number arbitrarily needed.P2Point as datum mark way similarly.
5th, the grating mapping of configuration sampling point：The angle and distance of each sampled point and datum mark on objective contour is calculated, and
Grating section corresponding to mapping that to.The profile point number fallen into each grating section is counted, is obtained such as Fig. 5 (a) and Fig. 5
(b) two context histograms, corresponding P1And P2, the mapping schematic diagram of grating and profile point is as shown in Figure 6.The color of histogram
What block gray scale embodied is the profile point number included in grating section.
6th, the structure of benchmark context histogram：In target identification, multiple samples are counted first, obtain every classification target
Two benchmark context histograms.The acquiring method of two benchmark context histograms and two context Nogatas of target to be measured
The acquiring method of figure is the same, is all to use the Shape context method put on the basis of objective contour major axis end points.For sonar
The same target of image, although the context histogram color lump value of different samples is different, structure is substantially the same.
As shown in Fig. 7, Fig. 8, Fig. 9 and Figure 10, therefore benchmark context histogram can be obtained by the statistics of sample.
7th, match cognization：Pass through benchmark context histogram and the context histogram correspondence position of target to be identified
Color lump value carry out target match cognization.In order to improve processing speed, make have in context histogram value color lump value be 1,
The color lump value of void value is 0, obtains the binaryzation histogram of context histogram, such as Figure 11.During matching, by the two of target to be measured
The binaryzation histogram correspondence position of value histogram and benchmark context histogram carries out AND operation, if result is 1
It is otherwise mismatch for matching.The matching degree of target is defined as follows：
N in above formulaMatchingBe matching color lump number, NBenchmarkOn the basis of have that value color lump is total in binaryzation histogram, NTargetTo treat
Surveying the binaryzation histogram of target has value color lump total.Because each sample has two context histograms, thus it is to be identified
Target need to match 4 times altogether with benchmark binaryzation histogram, and obtained maximum matching degree represents the final result of matching.Test table
Bright, maximum matching degree is more than 85% corresponding to similar target difference sample, and the maximum matching degree of different target is less than
60%.
Claims (3)
1. a kind of sonar target recognition methods based on Shape context method, it is characterised in that including following steps：
Step 1：Sample image statistics is carried out, is obtained using Shape context method in sample image per in classification target two
Hereafter histogram；
Step 2：Images to be recognized is read, two context histograms of target to be identified are obtained using Shape context method；
Step 3：By every classification two context Nogatas of target in the two context histograms and sample image of target to be identified
The value of the corresponding color lump of figure is matched respectively, and target identification is carried out using maximum matching degree；
Described Shape context method is：
A, sample image or images to be recognized are pre-processed, is partitioned into the bianry image of target area, transported by being opened and closed
Target area inside aperture is filled up in calculation；
B, the profile point of target area is extracted；
C, the distance between any two profile point is calculated, takes the major axis that maximum is target area profile, two ends of major axis
Point is respectively first end point P1With the second end points P2；
D, with first end point P1On the basis of point, using the length of major axis as the radius of grating, build with P1On the basis of put polar coordinates light
Grid；Similarly build with P2On the basis of put polar coordinates grating；
E, profile point is mapped to corresponding grating region by two polar coordinates gratings, obtains two context histograms, on
Hereafter the value of histogram color lump is the profile points included in corresponding grating region.
A kind of 2. sonar target recognition methods based on Shape context method according to claim 1, it is characterised in that：
Per classification two context histograms of target in the described two context histograms and sample image by target to be identified
The method that the value of corresponding color lump is matched respectively is：
Step 1：It is two benchmark context histograms to make every two context histograms of classification target in sample image, by base
Quasi- context histogram and the context histogram of target to be identified carry out binary conversion treatment, even the color lump value for having value is set to 1,
0 is set to without value color lump value, obtains benchmark binaryzation histogram and the binaryzation histogram of target to be identified；
Step 2：By every classification two benchmark binaryzations of target in the two binaryzation histograms and sample image of target to be identified
The corresponding color lump value of histogram is matched, and the matching degree of target is：
Wherein, NMatchingBe matching color lump number, NBenchmarkOn the basis of have that value color lump is total in binaryzation histogram, NTargetFor mesh to be measured
Target binaryzation histogram has value color lump total.
A kind of 3. sonar target recognition methods based on Shape context method according to claim 1, it is characterised in that：
The method of described structure polar coordinates grating is：Using datum mark as origin, long axis length is radius, and long axis direction is polar coordinates
0 ° of direction of axle, takes ± 90 ° of angular ranges as grating, and the angular range of grating is divided into N number of section, and major axis is equidistant
It is divided into M section.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510054144.XA CN104680132B (en) | 2015-01-30 | 2015-01-30 | A kind of sonar target recognition methods based on Shape context method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510054144.XA CN104680132B (en) | 2015-01-30 | 2015-01-30 | A kind of sonar target recognition methods based on Shape context method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104680132A CN104680132A (en) | 2015-06-03 |
| CN104680132B true CN104680132B (en) | 2017-11-21 |
Family
ID=53315150
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510054144.XA Active CN104680132B (en) | 2015-01-30 | 2015-01-30 | A kind of sonar target recognition methods based on Shape context method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104680132B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460773B (en) * | 2018-02-28 | 2022-03-18 | 哈尔滨工程大学 | Sonar image segmentation method based on offset field level set |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859382A (en) * | 2010-06-03 | 2010-10-13 | 复旦大学 | License plate detection and identification method based on maximum stable extremal region |
| CN102200999A (en) * | 2011-04-27 | 2011-09-28 | 华中科技大学 | Method for retrieving similarity shape |
| CN102467657A (en) * | 2010-11-16 | 2012-05-23 | 三星电子株式会社 | Gesture recognizing system and method |
| CN102609722A (en) * | 2012-02-07 | 2012-07-25 | 西安理工大学 | Method for fusing local shape feature structure and global shape feature structure of video image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010176380A (en) * | 2009-01-29 | 2010-08-12 | Sony Corp | Information processing device and method, program, and recording medium |
-
2015
- 2015-01-30 CN CN201510054144.XA patent/CN104680132B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859382A (en) * | 2010-06-03 | 2010-10-13 | 复旦大学 | License plate detection and identification method based on maximum stable extremal region |
| CN102467657A (en) * | 2010-11-16 | 2012-05-23 | 三星电子株式会社 | Gesture recognizing system and method |
| CN102200999A (en) * | 2011-04-27 | 2011-09-28 | 华中科技大学 | Method for retrieving similarity shape |
| CN102609722A (en) * | 2012-02-07 | 2012-07-25 | 西安理工大学 | Method for fusing local shape feature structure and global shape feature structure of video image |
Non-Patent Citations (2)
| Title |
|---|
| Shape Classification Using the Inner-Distance;Haibin Ling等;《IEEE》;20071231;第286-299页 * |
| 形状匹配方法研究与展望;周瑜等;《自动化学报》;20120630;第889-910页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104680132A (en) | 2015-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108121991B (en) | Deep learning ship target detection method based on edge candidate region extraction | |
| CN111611643B (en) | Household vectorization data acquisition method and device, electronic equipment and storage medium | |
| Li et al. | An edge embedded marker-based watershed algorithm for high spatial resolution remote sensing image segmentation | |
| CN107228860B (en) | Gear defect detection method based on image rotation period characteristics | |
| CN106530347B (en) | Stable high-performance circle feature detection method | |
| CN107038416B (en) | Pedestrian detection method based on binary image improved HOG characteristics | |
| CN106709500B (en) | Image feature matching method | |
| CN104778701A (en) | Local image describing method based on RGB-D sensor | |
| CN103400388A (en) | Method for eliminating Brisk (binary robust invariant scale keypoint) error matching point pair by utilizing RANSAC (random sampling consensus) | |
| CN104123554A (en) | SIFT image characteristic extraction method based on MMTD | |
| CN114862855B (en) | Textile defect detection method and system based on template matching | |
| CN108073940B (en) | Method for detecting 3D target example object in unstructured environment | |
| CN108961301B (en) | Chaetoceros image segmentation method based on unsupervised pixel-by-pixel classification | |
| CN110223310A (en) | A kind of line-structured light center line and cabinet edge detection method based on deep learning | |
| Li et al. | Application of multi-scale feature fusion and deep learning in detection of steel strip surface defect | |
| CN109101985A (en) | It is a kind of based on adaptive neighborhood test image mismatch point to elimination method | |
| CN111368865B (en) | Remote sensing image oil storage tank detection method and device, readable storage medium and equipment | |
| Jia | Fabric defect detection based on open source computer vision library OpenCV | |
| CN113989308A (en) | Polygonal target segmentation method based on Hough transform and template matching | |
| CN104680132B (en) | A kind of sonar target recognition methods based on Shape context method | |
| Kezia et al. | A color-texture based segmentation method to extract object from background | |
| Khamael et al. | Using adapted JSEG algorithm with fuzzy C mean for segmentation and counting of white blood cell and nucleus images | |
| CN104239411B (en) | A kind of detection method of the lattice-shaped radar based on color, position cluster and Corner Detection | |
| Akther et al. | Detection of Vehicle's Number Plate at Nighttime using Iterative Threshold Segmentation (ITS) Algorithm | |
| CN112364881A (en) | Advanced sampling consistency image matching algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |