CN104463896B - Image corner point detection method and system based on kernel similar region distribution characteristics - Google Patents
Image corner point detection method and system based on kernel similar region distribution characteristics Download PDFInfo
- Publication number
- CN104463896B CN104463896B CN201410830491.2A CN201410830491A CN104463896B CN 104463896 B CN104463896 B CN 104463896B CN 201410830491 A CN201410830491 A CN 201410830491A CN 104463896 B CN104463896 B CN 104463896B
- Authority
- CN
- China
- Prior art keywords
- area
- angle
- circle template
- nuclear phase
- image
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20164—Salient point detection; Corner detection
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image corner point detection method and system based on kernel similar region distribution characteristics. The image corner point detection method comprises the steps of establishing a round template and m corner templates, 2 obtaining the kernel similar region areas of the round template and the corner templates, 3 sorting the directions corresponding to the corner templates according to the kernel similar region areas to obtain a direction sequence, 4 determining main directions and judging the continuity of the main directions based on the direction sequence, 5 judging whether the centers of the templates are candidate corner points or nor according to the kernel similar region area of the round template, the difference value between the maximum kernel similar region area and the minimum kernel similar region area of the corner templates, the number of the main directions and the continuity of the main directions or not, and 6 performing non-maximum-value inhabitation based on a judgment result obtained in the step 5 and determining image corner points. In the corner point extraction process, the image corner point detection method and system simultaneously takes the areas of kernel similar regions and the distribution characteristics into account, mistaken corner point extraction can be avoided, and accordingly corner point detection accuracy is improved.
Description
Technical field
The invention belongs to technical field of image processing, be related to it is a kind of based on nuclear phase like area's distribution character image Corner Detection
Method and system.
Background technology
Angle point is one of important local feature of image, and it is to describe one of important information of target shape in image, because
This Corner Detection is one of key technology in the fields such as target recognition and tracking, images match, three-dimensional reconstruction.So far, it is existing
Many scholars are engaged in the research of Corner Detection, also generate a large amount of corner detection operators.These operators can be divided into two classes：It is based on
The operator of image border and the operator based on gradation of image, SUSAN (Small univalue segment assimilating
Nucleus) son is namely based on one of classical operator of gradation of image feature.SUSAN operators by Oxonian Smith and
Brady is proposed, using circle template, by the gray value detection angle for comparing template each pixel of inside and template center's pixel
Point.If the gray scale of certain pixel is less than certain value with the difference of template center's pixel grey scale in template, then it is assumed that the point and template
Center is similar, and the pixel quantity similar to central pixel point is nuclear phase like area (Univalue Segment in statistical mask
Assimilating Nucleus, USAN) area, when USAN is less than given threshold value, then judge that the central point is detected point.
Parameter needed for SUSAN operators is less to be easy to control, is affected little by template size, under the conditions of noisy according to
Required angle point can so be detected.Therefore, SUSAN operators have become one of corner detection operator more conventional at present, and in figure
As matching, camera are demarcated and the field such as target recognition and motion target tracking has obtained sufficient application.
So far, many scholars still are constantly improved to SUSAN operators.These improvement are mainly right in terms of three
SUSAN is improved：One is that the self adaptation choosing method for being directed to involved parameter in SUSAN operators is improved, typically right
The self adaptation of gray scale similarity threshold therein is chosen and is improved, such as the improvement that Bi Wuzhong etc. and Lv Haixia etc. is done.Two are
Similarity-rough set process in SUSAN operators is improved, so soldier etc. with pixel average in template center's certain area come
Replace the gray value of template center pixel to improve the effect of the Corner Detection in the image that signal to noise ratio is different, complexity is larger.
Three is to be combined SUSAN operators with the angular-point detection method for being based on image border with the advantage of comprehensive two class methods, such as Zhang Kunhua
On the basis of angle point is extracted in SUSAN, using edge unit to first angle point along edge direction tracking sequence and according to image border
The boundary direction situation of change of feature detection is rejecting the caused false angle point due to image digitization.
Although SUSAN operators have the advantages that in terms of Corner Detection it is prominent, and under the updating of numerous scholars by
Step is perfect, but can still there is the phenomenon for detecting error corner point in some cases, such as noisy image, can be by noise
It is mistaken for angle point.
The content of the invention
Extraction problem is missed for the angle point that SUSAN operators detection noise image in prior art is present, the invention provides
It is a kind of it is avoiding that angle point in noise image extracts by mistake, based on nuclear phase is like the image angular-point detection method of area's distribution character and is
System.
To solve above-mentioned technical problem, the present invention is adopted the following technical scheme that：
First, Corner Detection is carried out for gray level image like the image angular-point detection method of area's distribution character based on nuclear phase, is wrapped
Include step：
Step 1, one circle template of construction and m Angle formwork, by circle template the m equal sector of central angle i.e. angle is divided into
Template, each Angle formwork represents different directions, and m takes 6～12；
Step 2, successively with each pixel in gray level image as circle template center, in pixel and circle template in circle template
The gray value of the heart carries out similarity judgement, obtains the nuclear phase of circle template and Angle formwork under each circle template center like area's area；
Step 3, direction sequence { O is obtained by nuclear phase like area's area from big to small to the corresponding direction sequencing of Angle formwork1,
O2,...Om}；
Step 4, determines principal direction and judges principal direction seriality, this step further includes sub-step based on direction sequence
Suddenly：
4.1 sequentially in calculated direction sequence the corresponding Angle formwork nuclear phase in adjacent two direction like area's area attenuation degree, i.e.,
The corresponding Angle formwork nuclear phase in two directions like area's area difference and sequence in the corresponding Angle formwork nuclear phase of front direction like area's area
Ratio, obtains and direction sequence { O1,O2,...OmCorresponding attenuation degree sequence { D1,2,D2,3,...Dj,j+1,...Dm-1,m}；
4.2 sequentially look for first element more than 0.5 from attenuation degree sequence, are designated as Dk,k+1, then principal direction collection be
{O1,O2,…,Ok}；If attenuation degree sequence all elements are no more than 0.5, principal direction collection is { O1,O2,...Om}；
4.3 when principal direction concentrates the corresponding Angle formwork in all directions adjacent in circle template, then principal direction is continuous；
Step 5, the similar area of maximum kernel in threshold value aS, Angle formwork is less than by the nuclear phase for meeting circle template simultaneously like area's area
Area and minimum nuclear phase are more than threshold value aS like the difference of area's areai, principal direction quantity be not m and the continuous circle template of principal direction
Center is designated as candidate angular, and a is according to the acuity value of angle point, S and SiThe pixel that respectively circle template and Angle formwork are included
Points；
Step 6, the judged result based on step 5 carries out non-maxima suppression, to determine angle point.
If original image is single band image, above-mentioned gray level image is original image；If original image be coloured image or
Multi-band image, above-mentioned gray level image is the meansigma methodss of each wave band grey scale pixel value.
Above-mentioned steps 2 further include sub-step：
2.1 successively using each pixel in gray level image as circle template center；
2.2 compare one by one all pixels point and the gray value at circle template center in circle template and Angle formwork, obtain and circle mould
The similar pixel in plate center；
2.3 count respectively pixel similar to circle template center in circle template and Angle formwork, obtain circle template and angle mould
Nuclear phase of the plate under the mould plectane center is like area's area.
The further sub-step of above-mentioned steps 6：
6.1 judged results based on step 5 calculate angle point response value, i.e. the angle point response value of candidate angular is threshold value a
, like the difference of area's area, the angle point response value of non-candidate angle point is 0 for S and circle template nuclear phase；
6.2 carry out non-maxima suppression according to angle point response value, to determine angle point from candidate angular.
2nd, Corner Detection is carried out for gray level image like the image angle point detecting system of area's distribution character based on nuclear phase, is wrapped
Include：
(1) target formation module, for constructing a circle template and m Angle formwork, by circle template m central angle is divided into
Equal sector is Angle formwork, and each Angle formwork represents different directions, and m takes 6～12；
(2) nuclear phase is like area's area statistics module, for successively with each pixel in gray level image as circle template center, to circle
The gray value at pixel and circle template center carries out similarity judgement in template, and acquisition circle template and Angle formwork are in each circle template
Nuclear phase under the heart is like area's area；
(3) direction sequence obtains module, for pressing nuclear phase like area's area from big to small to the corresponding direction sequencing of Angle formwork,
Obtain direction sequence { O1,O2,...Om}；
(4) principal direction seriality judge module, for determining principal direction based on direction sequence and judging principal direction seriality,
This module further includes submodule：
Attenuation degree sequence obtains submodule, for the corresponding Angle formwork core in adjacent two direction in sequentially calculated direction sequence
The corresponding Angle formwork nuclear phase of the attenuation degree of similar area's area, i.e. two directions is corresponding in front direction with sequence like the difference of area's area
Angle formwork nuclear phase like area's area ratio, obtain with direction sequence { O1,O2,...OmCorresponding attenuation degree sequence { D1,2,
D2,3,...Dj,j+1,...Dm-1,m}；
Principal direction collection obtains submodule, for sequentially looking for first element more than 0.5, note from attenuation degree sequence
For Dk,k+1, then principal direction collection is { O1,O2,…,Ok}；If attenuation degree sequence all elements are no more than 0.5, principal direction collection
For { O1,O2,...Om}；
Seriality judging submodule, when principal direction concentrates the corresponding Angle formwork in all directions adjacent in circle template,
Then principal direction is continuous；
(5) candidate angular judge module, for the nuclear phase for meeting circle template simultaneously is less than into threshold value aS, angle like area's area
The similar area's area of maximum kernel and minimum nuclear phase are more than threshold value aS like the difference of area's area in templatei, principal direction quantity for m and
The continuous circle template center of principal direction is designated as candidate angular, and a is according to the acuity value of angle point, S and SiRespectively circle template
The pixel number included with Angle formwork；
(6) non-maxima suppression module, for the judged result based on candidate angular judge module non-maximum suppression is carried out
System, to determine angle point.
Above-mentioned nuclear phase further includes submodule like area's area statistics module：
Circle template center obtains submodule, for successively using each pixel in gray level image as circle template center；
Pixel similarity judging submodule, for comparing all pixels point and circle template in circle template and Angle formwork one by one
The gray value at center, obtains the pixel similar to circle template center；
Nuclear phase is similar to circle template center in circle template and Angle formwork for counting respectively like area's area statistics submodule
Pixel, obtains the nuclear phase of circle template and Angle formwork under the mould plectane center like area's area.
Above-mentioned non-maxima suppression module further includes submodule：
Angle point response value obtains submodule, and for the judged result based on candidate angular judge module angle point response is calculated
Value, i.e. the angle point response value of candidate angular is the difference of threshold value aS and circle template nuclear phase like area's area, the angle of non-candidate angle point
Point response value is 0；
Non-maxima suppression submodule, for carrying out non-maxima suppression according to angle point response value, with from candidate angular
Determine angle point.
When using SUSAN operator extraction angle points, circle template need to be typically defined, make circle template center and image pixel weight
Close, with the gray value of circle template center respective pixel (hereinafter template center) in all pixels included in circle template
Similar pixel constitutes nuclear phase like area, and the number of pixels i.e. nuclear phase that nuclear phase is included like area is designated as USAN like area's area.Work as circle template
When moving on image, nuclear phase can produce change like area's area with characteristics of image.In circle template from target external through target
Angle point is carried out in the moving process of target internal, and nuclear phase can form the regular change changed from small to big again from large to small like area's area
Change, and nuclear phase is minimum like area's area and nuclear phase is concentrated on like the pixel that area includes when circle template center overlaps with Corner
In a direction of circle template.
Area of the present invention using image angle point nuclear phase like area carries out Corner Detection with distribution character, to former SUSAN operators
It is improved, m Angle formwork is defined based on circle template, and nuclear phase is counted respectively like area in different directions using Angle formwork
Area, is designated as O-USAN；Judge that nuclear phase, like area's area, according to the O-USAN of Angle formwork distribution of the nuclear phase like area is determined according to USAN
Situation, and then whether judge templet center is angle point.
Compared to the prior art, the invention has the advantages that and beneficial effect：
During angle point grid, on the basis of Variation Features of the present invention in consideration angle point nuclear phase like area's area, fully
The distribution character of angle point is considered, Angle formwork is constructed based on circle template, and adopted based on Angle formwork nuclear phase like area's size
The direction sequence of structure describes distribution character, and then carries out angle point grid.Therefore, the present invention not only remains traditional SUSAN and calculates
The advantage of son detection angle point, the mistake that angle point can be avoided again is extracted, so as to improve Corner Detection precision.
Description of the drawings
Fig. 1 is the flow chart of the embodiment of the present invention；
Fig. 2 is by the Corner Detection template schematic diagram that constructs in the embodiment of the present invention.
Specific embodiment
For a better understanding of the present invention technical scheme, does into one with reference to the accompanying drawings and detailed description to the present invention
Step explanation.It is as follows with reference to the step of Fig. 1, the specific embodiment of the invention：
Step a, image gray processing.
The present invention is to carry out angle point grid to gray level image, it is therefore desirable to which original image is converted to into gray level image.For
Multi-band image, its gray level image is the meansigma methodss of multiband grey scale pixel value, i.e.,：
In formula (1), (x, y) represents location of pixels, and I (x, y) represents the gray value of pixel (x, y) in gray level image, Mk(x,
Y) gray value of k-th wave band pixel (x, y) in original image is represented.
Step b, constructs Corner Detection template.
The Corner Detection template of construction includes a circle template and m Angle formwork, is originally embodied as middle m and takes 8, will justify mould
Plate is divided into the fan-shaped acquisition Angle formwork that 8 central angles are 45 degree.Angle formwork has to comply with each Angle formwork, and to include pixel total
Number identical principle.Angle formwork not is one-to-one with its pixel for being included, and partial pixel can simultaneously belong to two phases
Adjacent angle template.
Fig. 2 is the Corner Detection template schematic diagram constructed in specific embodiment, and this group of Corner Detection template is that radius is
The circle template (figure (a)) of 5.5 pixels and 8 Angle formworks (figure (b)～figure (i)) built based on the circle template, figure (b)～figure
In (i) Angle formwork represent respectively direction 1,2 ... 8.For convenience of stating, the set of the pixel that circle template is included is designated as T0, angle mould
Set of the plate comprising pixel is designated as Ti, and wherein i represents the direction numbering of Angle formwork, i ∈ { 1,2 ..., 8 }.
Step c, pixel similarity judges and nuclear phase is like area's area statistics.
The template of the circle template for being constructed as step b using each pixel in the calculated gray level image of step a successively
Center, and compare the gray value of circle template inside all pixels point and template center one by one, if certain pixel and mould in circle template
The gray scale difference value at plate center is less than pixel similarity threshold, then it is assumed that the pixel is similar to template center.
The judge process of pixel similarity can be represented by the formula：
In formula (2), (x0,y0) expression template center position, pixel in the range of (x, y) expression circle template, c ((x, y),
(x0,y0)) represent pixel (x, y) and the similarity of template center；I(x0,y0), I (x, y) represent template center (x respectively0,
y0) and circle template in pixel (x, y) gray value；T be pixel similarity threshold, pixel similarity threshold t with it is to be checked
The minimum contrast of angle measurement point is relevant, and it determines that method is the routine techniquess in SUSAN operator extraction angle point algorithms, and here is not made
Repeat.
When | I (x, y)-I (x0,y0) | during no more than t, represent pixel (x, y) and template center (x0,y0) similar；It is no
Then, it is dissimilar.
All pixels in circle template are carried out after similarity judgement, respectively to meeting phase in circle template and 8 Angle formworks
Counted like the pixel quantity of condition, i.e., c ((x, y), (x of all pixels point in each template that adds up respectively0,y0)), formula
(3)～(4)：
In formula (3)～(4), N (x0,y0) represent the heart (x in a template0,y0) under the nuclear phase that includes of circle template like area's area, Ni
(x0,y0) represent the heart (x in a template0,y0) under direction for i the nuclear phase that includes of Angle formwork like area's area；T0Represent template center
For (x0,y0) the set of all pixels point that includes of circle template, TiRepresent all pixels point that direction includes for the Angle formwork of i
Set.
Step d, constructs the direction sequence of Angle formwork.
The nuclear phase of each Angle formwork obtained according to step c statistics like area's area, according to nuclear phase like area's area from big to small
Order is ranked up to the corresponding direction of 8 Angle formworks, constitutes direction sequence { O1,O2,O3,O4,O5,O6,O7,O8, wherein, Oj
Angle formwork direction i, i.e. O are corresponded to respectivelyj=i, wherein, j represents that direction sequence is numbered, and i represents that Angle formwork direction is numbered.
Step e, determines principal direction and judges principal direction seriality.
Sequentially in calculated direction sequence the nuclear phase of the corresponding Angle formwork in adjacent direction like area's area attenuation degree Dj,j+1：
In formula (5),O in direction sequence is represented respectivelyj、Oj+1The core of corresponding Angle formwork
Similar area's area；Dj,j+1Represent Oj+1And OjAttenuation degree of the corresponding Angle formwork nuclear phase like area's area.
The O from direction sequence1Start, using all adjacent directions in formula (5) sequentially calculated direction sequence nuclear phase seemingly
Area's area attenuation degree, obtains attenuation degree sequence of the following nuclear phase like area's area：
{D1,2,D2,3,D3,4,D4,5,D5,6,D6,7,D7,8}
From nuclear phase like sequentially find in area's area attenuation degree of sequence first more than 0.5 Dj,j+1, then principal direction set
For { O1,O2,…,Oj}；If nuclear phase is like all D in area's area attenuation degree of sequencej,j+1No more than 0.5, then principal direction collection is combined into
{O1,O2,…,O8}。
It is presently believed that principal direction is continuous and principal direction quantity is differed with general direction number, then show that nuclear phase is distributed like area
There is obvious directivity.Whether principal direction continues through principal direction seriality CO (x0,y0) judge, judge principal direction set { O1,
O2,…,OjIn the corresponding Angle formwork of all directions it is whether adjacent in circle template, when the corresponding angle in all directions in principal direction set
When template is adjacent in circle template, then CO (x0,y0)=1, shows that principal direction is continuous；Otherwise, CO (x0,y0)=0, shows master
Direction is discontinuous.
Step f, judges angle point and calculates angle point response value.
According to the nuclear phase of circle template like area area N (x0,y0), the nuclear phase of Angle formwork is like area area Ni(x0,y0), principal direction number
Amount NO (x0,y0), principal direction seriality CO (x0,y0) carry out angle point judgement.As template center pixel (x0,y0) meet formula (6)
When, by pixel (x0,y0) candidate angular is labeled as, with corner (x0,y0)=1 represents；Otherwise, corner (x0,y0)=0.
In formula (6), g be given circle template nuclear phase like area's area threshold, its determine method be SUSAN operator extraction angle points
Conventional meanses in algorithm, therefore not to repeat here；G is typically represented by aS, and S counts for all pixels that circle template is included
Amount, a is proportionality coefficient, according to the acuity value for extracting angle point；Og is that given Angle formwork nuclear phase is maximum poor like area's area
Different threshold value, og=aSi, SiRepresent all pixels point quantity that circle template is included.
Angle point response value need to be calculated in order to carry out non-maxima suppression, the angle point response value of candidate angular is circle template nuclear phase
Like area area threshold g and circle template nuclear phase like area's area difference, the angle point response value of non-candidate angle point is 0, sees formula (7)：
Step g, non-maxima suppression.
According to the angle point response value R (x that step f is obtained0,y0) non-maxima suppression is carried out, finally determine from candidate angular
Image angle point.
Claims (7)
1. Corner Detection is carried out, its feature for gray level image like the image angular-point detection method of area's distribution character based on nuclear phase
It is, including step：
Step 1, one circle template of construction and m Angle formwork, by circle template the m equal sector of central angle i.e. angle mould is divided into
Plate, each Angle formwork represents different directions, and m takes 6～12；
Step 2, successively with each pixel in gray level image as circle template center, to pixel in circle template and circle template center
Gray value carries out similarity judgement, obtains the nuclear phase of circle template and Angle formwork under each circle template center like area's area；
Step 3, direction sequence { O is obtained by nuclear phase like area's area from big to small to the corresponding direction sequencing of Angle formwork1,O2,
...Om}；
Step 4, determines principal direction and judges principal direction seriality, this step further includes sub-step based on direction sequence：
4.1 sequentially in calculated direction sequence the corresponding Angle formwork nuclear phase in adjacent two direction like area's area attenuation degree, i.e. two sides
To corresponding Angle formwork nuclear phase like area's area difference and sequence the corresponding Angle formwork nuclear phase of front direction like area's area ratio,
Obtain and direction sequence { O1,O2,...OmCorresponding attenuation degree sequence { D1,2,D2,3,...Dj,j+1,...Dm-1,m}；
4.2 sequentially look for first element more than 0.5 from attenuation degree sequence, are designated as Dk,k+1, then principal direction collection is { O1,
O2,…,Ok}；If attenuation degree sequence all elements are no more than 0.5, principal direction collection is { O1,O2,...Om}；
4.3 when principal direction concentrates the corresponding Angle formwork in all directions adjacent in circle template, then principal direction is continuous；
Step 5, the similar area's area of maximum kernel in threshold value aS, Angle formwork is less than by the nuclear phase for meeting circle template simultaneously like area's area
It is more than threshold value aS with difference of the minimum nuclear phase like area's areai, principal direction quantity be not m and the continuous circle template center of principal direction
It is designated as candidate angular, a is according to the acuity value of angle point, S and SiThe pixel number that respectively circle template and Angle formwork are included；
Step 6, the judged result based on step 5 carries out non-maxima suppression, to determine angle point.
2. image angular-point detection method of the nuclear phase like area's distribution character is based on as claimed in claim 1, it is characterised in that：
If original image is single band image, described gray level image is original image；If original image is coloured image or many
Band image, described gray level image is the meansigma methodss of each wave band grey scale pixel value.
3. image angular-point detection method of the nuclear phase like area's distribution character is based on as claimed in claim 1, it is characterised in that：
Step 2 further includes sub-step：
2.1 successively using each pixel in gray level image as circle template center；
2.2 gray values for comparing all pixels point and circle template center in circle template and Angle formwork one by one, in acquisition and circle template
The similar pixel of the heart；
2.3 count respectively pixel similar to circle template center in circle template and Angle formwork, obtain circle template and Angle formwork exists
Nuclear phase under the circle template center is like area's area.
4. image angular-point detection method of the nuclear phase like area's distribution character is based on as claimed in claim 1, it is characterised in that：
The further sub-step of step 6：
6.1 based on step 5 judged results calculate angle point response values, i.e. the angle point response value of candidate angular be threshold value aS with
Like the difference of area's area, the angle point response value of non-candidate angle point is 0 to circle template nuclear phase；
6.2 carry out non-maxima suppression according to angle point response value, to determine angle point from candidate angular.
5. Corner Detection is carried out, its feature for gray level image like the image angle point detecting system of area's distribution character based on nuclear phase
It is, including：
(1) target formation module, for constructing a circle template and m Angle formwork, is divided into m central angle equal by circle template
Sector be Angle formwork, each Angle formwork represents different directions, and m takes 6～12；
(2) nuclear phase is like area's area statistics module, for successively with each pixel in gray level image as circle template center, to circle template
The gray value at interior pixel and circle template center carries out similarity judgement, and acquisition circle template and Angle formwork are under each circle template center
Nuclear phase like area's area；
(3) direction sequence obtains module, for pressing nuclear phase like area's area from big to small to the corresponding direction sequencing of Angle formwork, obtains
Direction sequence { O1,O2,...Om}；
(4) principal direction seriality judge module, for determining principal direction based on direction sequence and judging principal direction seriality, this mould
Block further includes submodule：
Attenuation degree sequence obtains submodule, for the corresponding Angle formwork nuclear phase in adjacent two direction in sequentially calculated direction sequence seemingly
Difference and sort in front direction corresponding angle of the corresponding Angle formwork nuclear phase of the attenuation degree of area's area, i.e. two directions like area's area
The ratio of the similar area's area of template core, obtains and direction sequence { O1,O2,...OmCorresponding attenuation degree sequence { D1,2,
D2,3,...Dj,j+1,...Dm-1,m}；
Principal direction collection obtains submodule, for sequentially looking for first element more than 0.5 from attenuation degree sequence, is designated as
Dk,k+1, then principal direction collection is { O1,O2,…,Ok}；If attenuation degree sequence all elements are no more than 0.5, principal direction collection is
{O1,O2,...Om}；
Seriality judging submodule, when principal direction concentrates the corresponding Angle formwork in all directions adjacent in circle template, then leads
Direction is continuous；
(5) candidate angular judge module, for the nuclear phase for meeting circle template simultaneously is less than into threshold value aS, Angle formwork like area's area
The middle similar area's area of maximum kernel and minimum nuclear phase are more than threshold value aS like the difference of area's areai, principal direction quantity be not m and main formula
Candidate angular is designated as to continuous circle template center, a is according to the acuity value of angle point, S and SiRespectively circle template and angle
The pixel number that template is included；
(6) non-maxima suppression module, for the judged result based on candidate angular judge module non-maxima suppression is carried out, with
Determine angle point.
6. image angle point detecting system of the nuclear phase like area's distribution character is based on as claimed in claim 5, it is characterised in that：
Described nuclear phase further includes submodule like area's area statistics module：
Circle template center obtains submodule, for successively using each pixel in gray level image as circle template center；
Pixel similarity judging submodule, for comparing all pixels point and circle template center in circle template and Angle formwork one by one
Gray value, obtain the pixel similar to circle template center；
Nuclear phase like area's area statistics submodule, for counting pixel similar to circle template center in circle template and Angle formwork respectively
Point, obtains the nuclear phase of circle template and Angle formwork under the circle template center like area's area.
7. image angle point detecting system of the nuclear phase like area's distribution character is based on as claimed in claim 5, it is characterised in that：
Described non-maxima suppression module further includes submodule：
Angle point response value obtains submodule, and for the judged result based on candidate angular judge module angle point response value is calculated, i.e.
The angle point response value of candidate angular is the difference of threshold value aS and circle template nuclear phase like area's area, and the angle point of non-candidate angle point is responded
It is worth for 0；
Non-maxima suppression submodule, for carrying out non-maxima suppression according to angle point response value, to determine from candidate angular
Angle point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410830491.2A CN104463896B (en) | 2014-12-26 | 2014-12-26 | Image corner point detection method and system based on kernel similar region distribution characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410830491.2A CN104463896B (en) | 2014-12-26 | 2014-12-26 | Image corner point detection method and system based on kernel similar region distribution characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104463896A CN104463896A (en) | 2015-03-25 |
| CN104463896B true CN104463896B (en) | 2017-04-12 |
Family
ID=52909878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410830491.2A Active CN104463896B (en) | 2014-12-26 | 2014-12-26 | Image corner point detection method and system based on kernel similar region distribution characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104463896B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106023171B (en) * | 2016-05-12 | 2019-05-14 | 惠州学院 | A kind of image angular-point detection method based on turning radius |
| CN106846412B (en) * | 2017-01-23 | 2020-03-13 | 上海兴芯微电子科技有限公司 | Checkerboard angular point detection method and device |
| CN109919957B (en) * | 2019-01-08 | 2020-11-27 | 同济大学 | Corner detection method based on dynamic vision sensor |
| CN116128804B (en) * | 2022-11-28 | 2024-02-27 | 深圳技术大学 | Multi-scale multi-direction unmanned aerial vehicle image corner detection method, system and terminal |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789637A (en) * | 2012-07-12 | 2012-11-21 | 北方工业大学 | Salient region extraction based on improved SUSAN (small univalue segment assimilating nucleus) operator |
| CN103279956A (en) * | 2013-05-30 | 2013-09-04 | 华南理工大学 | Method for detecting positional accuracy of components of chip mounter |
-
2014
- 2014-12-26 CN CN201410830491.2A patent/CN104463896B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789637A (en) * | 2012-07-12 | 2012-11-21 | 北方工业大学 | Salient region extraction based on improved SUSAN (small univalue segment assimilating nucleus) operator |
| CN103279956A (en) * | 2013-05-30 | 2013-09-04 | 华南理工大学 | Method for detecting positional accuracy of components of chip mounter |
Non-Patent Citations (3)
| Title |
|---|
| SUSAN—A New Approach to Low Level Image Processing;Stephen M.Smith等;《International Journal of Computer Vision》;19970531;第23卷(第1期);第45-78页 * |
| 一种基于USAN的特征点检测算法;杨幸芳等;《机械科学与技术》;20110731;第30卷(第7期);第1120-1123页 * |
| 一种稳健的高效角点特征提取变换;陈敏 等;《武汉大学学报(信息科学版)》;20130831;第38卷(第10期);第1142-1147页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104463896A (en) | 2015-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107248159A (en) | A kind of metal works defect inspection method based on binocular vision | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| CN106683091A (en) | Target classification and attitude detection method based on depth convolution neural network | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| CN109447065A (en) | A kind of method and device of breast image identification | |
| CN107392885A (en) | A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism | |
| CN110232379A (en) | A kind of vehicle attitude detection method and system | |
| CN104463896B (en) | Image corner point detection method and system based on kernel similar region distribution characteristics | |
| CN106204540B (en) | Visible detection method | |
| CN109363698A (en) | A kind of method and device of breast image sign identification | |
| CN109977968B (en) | SAR change detection method based on deep learning classification comparison | |
| CN109363699A (en) | A kind of method and device of breast image lesion identification | |
| CN105678792A (en) | Method and system for extracting body profile | |
| CN103218605A (en) | Quick eye locating method based on integral projection and edge detection | |
| CN108550166B (en) | Spatial target image matching method | |
| CN107909593B (en) | Non-directional selective contour detection method based on receptive field region | |
| CN104657980A (en) | Improved multi-channel image partitioning algorithm based on Meanshift | |
| CN104966305A (en) | Foreground detection method based on motion vector division | |
| CN108664994A (en) | A kind of remote sensing image processing model construction system and method | |
| CN109363697A (en) | A kind of method and device of breast image lesion identification | |
| CN112802054A (en) | Mixed Gaussian model foreground detection method fusing image segmentation | |
| CN104282027A (en) | Circle detecting method based on Hough transformation | |
| CN108257153A (en) | A kind of method for tracking target based on direction gradient statistical nature | |
| CN105069459B (en) | One kind is directed to High Resolution SAR Images type of ground objects extracting method | |
| CN106683125A (en) | RGB-D image registration method based on 2D/3D mode switching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |