CN101794446A - Line search type detection method of image corner point - Google Patents
Line search type detection method of image corner point Download PDFInfo
- Publication number
- CN101794446A CN101794446A CN 201010110067 CN201010110067A CN101794446A CN 101794446 A CN101794446 A CN 101794446A CN 201010110067 CN201010110067 CN 201010110067 CN 201010110067 A CN201010110067 A CN 201010110067A CN 101794446 A CN101794446 A CN 101794446A
- Authority
- CN
- China
- Prior art keywords
- nuclear
- class
- point
- pixel
- center
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 230000004044 response Effects 0.000 claims abstract description 18
- 238000001914 filtration Methods 0.000 claims abstract description 11
- 238000007689 inspection Methods 0.000 claims description 29
- 230000014509 gene expression Effects 0.000 claims description 19
- 238000012360 testing method Methods 0.000 claims description 9
- 238000013461 design Methods 0.000 claims description 4
- 238000000034 method Methods 0.000 abstract description 59
- 238000004422 calculation algorithm Methods 0.000 abstract description 8
- 230000001629 suppression Effects 0.000 abstract 2
- 230000000149 penetrating effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 12
- 238000002187 spin decoupling employing ultra-broadband-inversion sequences generated via simulated annealing Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 238000005259 measurement Methods 0.000 description 8
- 230000005764 inhibitory process Effects 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000012795 verification Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000001154 acute effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000002161 passivation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- GOLXNESZZPUPJE-UHFFFAOYSA-N spiromesifen Chemical compound CC1=CC(C)=CC(C)=C1C(C(O1)=O)=C(OC(=O)CC(C)(C)C)C11CCCC1 GOLXNESZZPUPJE-UHFFFAOYSA-N 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a line search type detection method of an image corner point, which overcomes the problem that the conventional detection method of the image corner point can not give consideration to both speed and accuracy. The invention is realized via the following technical scheme that: a circular mask taking the current pixel as a central core is affected, lines penetrating through the core are searched in the mask, and if a straight line which does not penetrates through other USAN areas outside the given neighborhood near the core exists, then the current pixel is the corner point. In the implementation, the number of search lines and the number of detection points on the search lines are designed in different steps, and a measure of filtering candidate corner points from coarse to fine in different steps is adopted so as to avoid the early intervention of large computational complexity and improve the detection speed. The invention also provides a novel non-maximum value suppression method based on the maximum equal distance, which is combined with a plurality of novel pseudo-response suppression measures to effectively improve the accuracy of the invention. The experimental results show that the method is superior to classical algorithms in accuracy, and has high speed and excellent comprehensive performance.
Description
Technical field
The present invention relates to the detection method of the extensive angle point that exists in the Flame Image Process, can be applicable to optical flow computation, Target Recognition, tracking, estimation and three-dimensionalreconstruction etc. and relate to the computer vision occasion that angle point is used.
Background technology
Angle point is a key character of image, has invariable rotary, does not change with illumination condition and the robustness condition for identification of significant change is widely used in occasions such as optical flow computation, target following, estimation and three-dimensionalreconstruction.About the definition of angle point, more unified saying is the marginal point of local curvature's maximum at present.General image all comprises a large amount of angle points, and the research of method has very strong versatility.A large amount of scholars carried out the research of this characteristic detection method, and were relatively more classical and simple relatively method has the MIC method, the SUSAN method, and Harris method etc., using maximum at present is Harris and SUSAN.Scholar Harris proposes famous plessey operator, and the outstanding advantage of this algorithm is to be simple and easy to implement, and good stability is widely used at present.Shortcoming is that bearing accuracy is undesirable, especially at " T ", and " X ", the angle point place of " Y " type; Also lose the angle point of the fuzzy type in the not remarkable or big obtuse angle of gradient easily, computing time is long in addition.1997, Smith proposed the notion of USAN first, and had designed the angular-point detection method of SUSAN based on this, and its great advantage is that method is very simple, has integral characteristic, and noiseproof feature is good, and bearing accuracy is not subjected to the influence of angle point type; Shortcoming is because there is a large amount of ambiguities edge in real image, is easy to generate pseudo-response or easily loses true angle point, causes the verification and measurement ratio of real image general not as good as the Harris algorithm, and integral process also causes consuming time more in addition.MiroslavTrajkovic should require to have proposed the MIC algorithm fast.This algorithm might be the fastest angular-point detection method during present gray level image is handled.Though this method simply easily produces pseudo-response regrettably, especially in oblique straight edge or ambiguity edge; Detection level is general, and is also quite responsive to noise.But its rapidity thought has but obtained excellent popularization, and including in can pick up speed in other algorithm.
The subject matter of these classical ways is that the detection execution time is longer, and levels of accuracy is not high; Short method puppet of execution time responds more, and it is poor to detect performance.So the short again method of a kind of detection level height and execution time that designs seems very necessary, especially has more value for real-time computer vision application scenario.
Summary of the invention
The invention provides the line search formula detection method of the image angle point that a kind of detection speed is fast, accuracy rate is high.
The present invention is achieved by following technical proposals:
A kind of line search formula detection method of image angle point, step is as follows:
Serve as nuclear with each pixel respectively and be the center of circle with this center of examining, with 8 pixel units is diameter, make imaginary first kind circle mask, search is called first kind scounting line by the imaginary line at nuclear center and with these imaginary lines in this first kind circle mask, being check point about the paired pixel examined on the centrosymmetric round mask border, if the same value retract USAN in described first kind circle mask can find paired pixel as check point with exterior domain, the first kind scounting line of then crossing described paired pixel is the non-ruler of the first kind, and the nuclear that the nuclear center is dropped on the non-ruler of the first kind is first kind candidate angular; Otherwise, abandon described nuclear.
The puppet response of the first kind candidate angular of step 2 pair lines and ridge-roof type image border suppresses
Be nuclear with each first kind candidate angular respectively and be the center of circle with this center of examining, with 16 pixel units is diameter, make imaginary second class circle mask, search is called the second class scounting line by all imaginary lines of nuclear center and pixel and with these imaginary lines in this second class circle mask, if the pixel that has adjacent with nuclear phase at least one the second class scounting lines and be positioned at the nuclear both sides and nuclear with value, then curb described nuclear; Otherwise, with described nuclear as the second class candidate angular,
The check of step 3 scounting line
Sampled point on the step 3.1 pair scounting line is tested
Be nuclear with each second class candidate angular respectively and be the center of circle with this center of examining, with 16 pixel units is diameter, make imaginary the 3rd class circle mask, search is called the 3rd class scounting line by the imaginary line of nuclear and with these imaginary lines in the 3rd class circle mask, get n article of the 3rd class scounting line as cut-off rule, the 3rd class circle mask is divided into the 2n equal portions, and n is for cutting apart number of lines, will be positioned on the cut-off rule again and be positioned at the pixel of round mask and the sub-pix point as Z sampled point P
i(x, y), Z is the number of sampled point, i is the sampled point numbering, i=1,2,, Z, x are the horizontal ordinate of sampled point, y is the ordinate of sampled point, is initial point O with the nuclear center, is X-axis with the cut-off rule of crossing the pixel of going together, with the cut-off rule of crossing the same column pixel is Y-axis, sets up the XOY orthogonal coordinate system, and is coordinate unit with 1 pixel unit, prohibit outside the inspection circle to being positioned at again, the 3rd class circle mask with sampled point test
If all sampled points that have in all cut-off rules of crossing each nuclear center at least one cut-off rule and the described cut-off rule all are positioned at beyond the same value retract USAN zone of the 3rd class circle mask, then described cut-off rule is the non-ruler of the 3rd class, and nuclear on the non-ruler of described the 3rd class is dropped on as the 3rd class candidate angular in the nuclear center; Otherwise, abandon described nuclear,
Described taboo inspection circle is to be that the center of circle, radius are that the circle of 4 pixel units is regional with the nuclear center,
Described sampled point is:
When | tan α | in the time of≤1, sampled point P
iHorizontal ordinate x be the coordinate unit integral multiple,
Sampled point P
iOrdinate y=x*tan α;
When | tan α | in the time of>1, sampled point P
iOrdinate y be the coordinate unit integral multiple,
Sampled point P
iHorizontal ordinate x=y*cot α,
Wherein, α is the angle of cut-off rule and X-axis,
Step 3.2 suppresses the pseudo-angle point on the ramp type image border
Be nuclear with each the 3rd class candidate angular respectively, and be the center of circle, in the 3rd class circle mask, satisfy if cross two sampled point brightness values of outermost end on any one article of the 3rd non-ruler of class at center of each nuclear with this center of examining:
Then the nuclear that the nuclear center is dropped on the non-ruler of described the 3rd class is suppressed; Otherwise, with described nuclear as the 4th class candidate angular,
Wherein: P
j, P
kTwo sampled points of expression outermost end,
The brightness value of representing described sampled point, I
NBe the brightness value of described nuclear,
Step 3.3 suppresses to determine the final image angle point based on the non-maximum value of Datong District value distance
Datong District value of calculating each the 4th class candidate angular respectively is apart from d
MH,
Design imaginary square window, this window is the center with each the 4th class candidate angular, and window width is 6 pixel units, respectively all candidate angular in each square window relatively
Value, therefrom select maximum
Value, and with maximum
Value corresponding candidate angular as final image angle point, with in each square window all the other
Be worth off-peak the 4th class candidate angular and curb,
Datong District value of described each the 4th class candidate angular is apart from d
MHRefer to: a minimum value in the Datong District of the first kind of the non-ruler of all the 3rd classes of each the 4th class candidate angular value distance,
The Datong District of described first kind value distance refers to: on arbitrary article the 3rd non-ruler of class of four class candidate angular, in prohibiting the inspection circle, and nuclear with value and apart from the nuclear center farthest point and the distance between the nuclear center,
Described distance nuclear point farthest is pixel or sub-pix point.
The present invention helps avoiding losing of true angle point feature under the situation that suitably increases scounting line quantity, add multinomial pseudo-response braking measure, make the verification and measurement ratio of the inventive method be guaranteed, the non-maximum value braking measure that proposes based on Datong District value distance, guaranteed the bearing accuracy of the inventive method, helped it and surpass classic methods.Take to be filtered out by thick and segmentation step the measure of candidate angular, avoided the too early intervention of intensive measure, this is the inventive method major reason fast.
The present invention compares with classic methods and has obtained following useful effect:
The evaluation criterion of introduction method contrast at first:
Evaluation map mainly contains accuracy (accuracy) and execution time as the standard of feature detection.Accuracy performance comprises the performance of two aspects again: 1) verification and measurement ratio; 2) bearing accuracy.
Verification and measurement ratio comprises just inspection rate and false drop rate two aspects.Those true angle points that are detected are designated as: and true positives (truepositive, TP); Those are not that true angle point is being designated as of angle point by flase drop but: and false positive (false positive, FP); Those were being designated as of but not being detected of true angle point originally: and false negative (false negative, FN).Just inspection rate (detectionrate) computing formula:
False drop rate (error rate) computing formula:
Bearing accuracy is weighed by positioning error, and positioning error is meant the offset distance between detected characteristic point position and the correct characteristic point position.The calculating formula of average positioning error:
Wherein n be true angle point place in the image (near) the angle point number of correct response, pseudo-response is not counted in (x
t, y
t) be the angular coordinate value that detects, (x
r, y
r) for the angle point correct coordinates value of reference.
Four kinds of angular-point detection methods of table 1 are than the average detected rate under the multisample
The inventive method and classical Several Methods at the subjective observation of artificial typical angular point diagram, building blocks figure, house figure to such as Figure 12, Figure 13, Figure 14.Wherein the figure of each figure (a) is former figure, and figure (b) is the exercising result of MIC method, and figure (c) is the exercising result of SUSAN method, and figure (d) is the exercising result of Harris method, and figure (e) is the exercising result of the inventive method.From the effect of these figure as can be known, the inventive method has demonstrated higher detection level than the several method of classics, and being embodied in not only most real features can respond, and pseudo-response is also fewer.
It is level and smooth that four angular-point detection methods of table 2 act on different Gausses
The average positioning error contrast of artificial figure
According to the verification and measurement ratio standard that provides previously, table 1 has provided several method at the verification and measurement ratio quantitative comparing results after being rich in the effect of angle point image more.The result shows: just inspection rate of the inventive method is all apparently higher than the several method of classics, and false drop rate is also low than them, has reached on the high detection level.
It is smoothly following in Gauss in various degree below to provide former figure, the contrast and experiment of several method aspect bearing accuracy:
Wherein table 2 is Figure 12 (artificial figure) ambiguity by the gaussian filtering actual figure of level and smooth simulation in various degree,
It is level and smooth that four angular-point detection methods of table 3 act on different Gausses
The average positioning error contrast of building blocks figure
It is level and smooth that four angular-point detection methods of table 4 act on different Gausses
The average positioning error contrast of house figure
Table 3 and table 4 item are that the Gauss in various degree of actual figure of Figure 13 and Figure 14 is level and smooth, because actual figure itself blurs, so only proceed to σ
2=1 level and smooth, high-intensity more smoothly meaningless.From then on the result as can be known, the bearing accuracy of the inventive method is better than classical certain methods generally, especially advantage is more outstanding on low level and smooth degree.
Four angular-point detection methods of table 5 are than the execution time scope under the multisample
Table 5 provides several method at the comparing result aspect the execution time.
Wherein the time of Flame Image Process can be different because of the difference of picture size, content, even if same width of cloth figure, the execution time neither be constant, so provided the time range of each algorithm under the different images size in table 5.For more objective, the result of table 5 is from the statistics of more processing result image.Picture size comprises 256 * 256 resolution, and the actual grey image of 640 * 480 resolution more commonly used is also arranged.What test was used is the PC of CPU as the AMD imperial 1600+ of speed (1.4GHz).From then on the result though the execution time of the inventive method is slightly more than MIC, has still lacked nearly order of magnitude than Harris and SUSAN as can be known, and the execution time is fast generally speaking.
Description of drawings
Fig. 1 is the ultimate principle figure of the inventive method, (a), and (b), (c) among the figure, 1,2,3 all represent with the current pixel to be nuclear, it is also as the center of circle (nuclear is idealized as a point in this schematic diagram) of the first kind or the 3rd class circle mask, and 4,5,6 all represent the first kind or the 3rd class circle mask, 7,8, the shadow region shown in 9 is all represented with value retract USAN zone; The inspection circle is prohibited in 10,11,12 expressions, and 13,14,15 all represent the first kind or the 3rd class scounting line.
Fig. 2 (a) is the inceptive filtering synoptic diagram, and each square is represented a pixel among the figure, and what there was stain at the center is the pixel that we are concerned about, wherein the pixel on the first kind circle mask border is as check point, and 21 represent the nuclear center O of circle masks
0, the arbitrary pixel r on the 22 expression mask border, 23 expression r are about O
0The point r ' of symmetry.In the inceptive filtering step, owing to be indifferent to the sub-pix point, so nuclear, pixel and pixel can be considered in the diagram with one deck notion.
Among Fig. 2 (b), the 24th, pass the nuclear center and with the round mask border of nuclear center same column on the first kind scounting line of pixel, the 25th, pass the nuclear center and the round mask border of going together with nuclear on the first kind scounting line of pixel, the 26th, the intersection point c of first kind scounting line 25 and circle mask border, the 27th, the intersection point d of first kind scounting line 25 and circle mask border.
Fig. 3 is the synoptic diagram that lines type image border easily produces pseudo-response, the pixel of shadow representation has been formed a lines type edge, j is one of them marginal point, the initially round mask of the first step of the general marginal point j of 30 expressions, bright area in 31 presentation videos, the dark areas in 32 presentation videos, 33,34 expressions are used to judge that j is the paired check point of the inceptive filtering of first kind candidate angular, and k is true angle point.Herein, marginal point and angle point all are the notions on the pixel aspect.
Fig. 4 is the puppet response inhibition schematic diagram to the first kind candidate angular of lines and ridge-roof type image border.
Fig. 5 is the synoptic diagram of Z sampled point on the scounting line in the basic checkout procedure.
Fig. 6 is that the ramp type marginal point that is mistaken for the 3rd class candidate angular shows with the brightness variation real image amplification of true angle point along the non-ruler of the 3rd class.Each pixel has been amplified to a tangible square size.The 61st, ramp type marginal point O
1The non-ruler of the 3rd class, the 62nd, true angle point O
2The non-ruler of the 3rd class.
Fig. 7 is the difference schematic diagram of ramp type marginal point and true angle point, (a) near the brightness value change curve of true angle point on the non-ruler of figure expression the 3rd class; (b) near the brightness value change curve the non-ruler upslope of figure expression the 3rd class type marginal point.
Fig. 8 is based on Datong District value apart from d
MHNon-maximum value suppress to determine the principle signal of qualified true angle point, bright area in 81 presentation videos, dark areas in 82 the shadow region presentation video, the taboo inspection circle of 83 expressions the 4th class candidate angular Oa, the taboo inspection circle of 84 expressions the 4th class candidate angular Ob, the non-ruler of the 3rd class of 85 expression Oa, the non-ruler of the 3rd class of 86 expression Ob.
Fig. 9 is a block pattern, the true angle point of 91,92,93 expressions.
Figure 10 is the process flow diagram of embodiment.
Figure 11 is the process flow diagram of the preliminary filtering of step 1.
Figure 12 is MIC, SUSAN, and Harris and the inventive method are compiled action effect on the figure at artificial angle point, (a) be original graph, (b) for MIC detects effect, (c) are SUAN detection effect, (d) for Harris detection effect, (e) detect effect for the present invention.
Figure 13 is MIC, SUSAN, and Harris and the inventive method action effect on block pattern (a) is original graph, (b) for MIC detects effect, (c) for SUAN detects effect, (d) for Harris detects effect, (e) detects effect for the present invention.
Figure 14 is MIC, SUSAN, and Harris and the inventive method action effect on house figure (a) is original graph, (b) for MIC detects effect, (c) for SUAN detects effect, (d) for Harris detects effect, (e) detects effect for the present invention.
Embodiment
The present invention is further detailed explanation below in conjunction with the drawings and specific embodiments.
The present invention is a kind of line search formula detection method of image angle point, ultimate principle is: acting on one is the round mask of centronucleus with the current pixel, search is by the straight line of nuclear in this mask, if exist straight line not pass near the given neighborhood of nuclear other USAN zone in addition, then current pixel point is an angle point.Ultimate principle is easy for narration, and " is the round mask of centronucleus with the current pixel " refers to " serve as nuclear and be the round mask of center of circle effect with this center of examining with each pixel "; " by nuclear " refers to " by the " center " of nuclear; Search is imaginary straight line by " straight line " of nuclear, is called " scounting line " in the present invention again; " do not pass near the straight line in the given neighborhood of nuclear other USAN zone in addition " and be called " non-ruler ".Given neighborhood is prohibited the inspection circle by one and delimited near the nuclear, and as 10,11,12 drawn circles among Fig. 1, relevant notion will illustrate in 3.1 steps of embodiment.There is a scounting line that satisfies above-mentioned requirements at least in angle point place as Fig. 1 (a), and 13 among the figure becomes non-ruler; Then do not have such scounting line for marginal point or flat site, as Fig. 1 (b) and 14,15 each scounting line (c) all can not become non-ruler.
The present invention has used the notion of USAN, and it is proposed in the SUSAN angular-point detection method by Smith scholar, is effect one circle mask in the center of circle with the current pixel point, " nuclear " that current pixel is called as mask (nucleus).In mask, the zone that has the pixel of identical or close brightness to form by those and nuclear is called as " USAN " (Univalue SegmentAssimilating Nucleus is with value retract).The present invention is not the improvement to classic method, but has proposed a kind of line search formula detection method of image angle point, has overcome the shortcoming that classical angular-point detection method can not be taken into account accuracy and rapidity simultaneously, thereby improves the level of aggregation of Corner Detection.
" with value " similar Smith scholar among the present invention in SUSAN angle point algorithm to definition with value, expression: with nuclear identical or close brightness is arranged, promptly be not strict be the brightness value that equates with nuclear, but delimit by the threshold value t in USAN zone.Concrete discrimination principle is: if certain pixel or sub-pix point P
iSatisfy:
Then this brightness with nuclear O is worth together or is close, for easy, and general designation " with value ", this pixel or sub-pix point P this moment
iThe same value retract USAN zone that belongs to this nuclear, wherein
I
OShow pixel or sub-pix point P
iWith the brightness value of nuclear, t is that the value of t is 2 to 40 with value retract USAN regional luminance difference threshold value, and for example, t gets 18; Otherwise, promptly
Pixel or sub-pix point P
iDo not belong to value retract USAN zone.
Brightness among the present invention and gray scale are identical concepts.
Among the present invention, " pixel " is the piece that size is arranged; " the promptly whole pixel of pixel " from accurately determining the angle of position, refers to the central point of pixel, is different from the whole pixel of sub-pix point that is:.
Among the present invention, nuclear is current pixel, so be the notion on the pixel aspect.The value of nuclear refers to the brightness value examined from accurate position, refer to examine central spot, i.e. the brightness value at whole pixel place, similarly, and " with nuclear with value " implication be: with the same brightness value of pixel as nuclear, brightness value is by the position acquisition at whole pixel place.Among the present invention, unless mention the positional precision of sub-pixel, the pixel brightness value under other situation all refers to the brightness value at pixel center point place.
Angle point described in the present invention is to locate by the precision of whole Pixel-level, so angle point is finally next definite with pixel (rather than sub-pix), so angle point and pixel are the notions on the same aspect.
Round mask described in the present invention comprises round mask border with interior zone.
Same value retract USAN described in the present invention does not comprise the border of value retract USAN together with exterior domain.
Zone outside the taboo inspection circle described in the present invention does not comprise prohibits inspection round edge circle.
Sampled point described in the present invention is the point of testing, so be also referred to as " check point ".
A pixel unit is meant the length of side of a pixel.Pixel is handled by desirable square block herein.
Concrete steps of the invention process are as follows:
A given width of cloth digital picture, image acquiescence are 8 gray level images, if coloured image, and then with R, G, three kinds of color component value of B obtain the brightness value of arbitrary pixel by weighting scheme, and relevant calculation all has known method.
Ultimate principle is so to embody in the first step: when specifically implementing, serve as nuclear with each pixel respectively and be the center of circle with this center of examining, with 8 pixel units is diameter, make imaginary first kind circle mask, search is called first kind scounting line by the imaginary line at nuclear center and with these imaginary lines in this first kind circle mask, with about examine on the centrosymmetric first kind circle mask border each to paired pixel one by one as check point.If in described first kind circle mask, exist certain paired pixel be positioned at value retract USAN with exterior domain, the first kind scounting line of then crossing described paired pixel is the non-ruler of the first kind, and the nuclear that the nuclear center is dropped on the non-ruler of the first kind is first kind candidate angular; Otherwise, abandon described nuclear.
Fig. 2 is the synoptic diagram of step 1, wherein the measurement of first kind circle mask diameters is as the criterion with the distance of two ends (on the border) pixel center of the colleague that crosses the center of circle or same column, so go together or 9 pixels of same column are that the actual circle of diameter forms for the round actual needs of the diameter of 8 pixel units.
The pixel that has comprised first kind circle mask border in the described first kind circle mask.Because the discrete feature of image, pixel on the first kind circle mask border differs to establish a capital and drops on just on the circular arc that desirable diameter is 8 pixel units, have only the boundary pixel point on the row or column just in time to drop on the desirable circular arc, for convenient and quick, the present invention this step all with whole pixel be pixel as check point, but not accurately the sub-pix point on the circular arc as check point.
Among Fig. 2 (a), with about examine 21 symmetries and arbitrary paired pixel r22 on first kind circle mask border and r ' 23 as check point.Fig. 2 (b) expression is for a kind of so desirable acute angle type angle point, on first kind scounting line 25, according to the aforementioned principle that should go on foot, with first kind circle mask border therewith the intersection point of first kind scounting line as check point, because pixel c26 and d27 can get not in the USAN zone: the nuclear of circle mask becomes first kind candidate angular.
Need observe following 3 rules during filtration: 1) on certain first kind scounting line, if a check point is arranged in same value retract USAN zone, then described first kind scounting line is not the non-ruler of the first kind, continues the check of other new first kind scounting line; 2) if two check points on certain first kind scounting line all are confirmed not in value retract USAN zone, then described first kind scounting line becomes the non-ruler of the first kind, and described nuclear is as first kind candidate angular.3) after all first kind scounting line checks through first kind circle mask, not existing to become the first kind of the non-ruler of first kind scounting line, then abandons described nuclear.Figure 11 is the process flow diagram of this step under this rule.I wherein
rWith I
R 'Arbitrary brightness value in the difference presentation graphs 2 (a) to pixel r and r ',
The brightness value of representing described nuclear.
Preliminary filtered pixel is rejected a large amount of ineligible pixels, makes that subsequent step only needs to detect in a spot of first kind candidate angular, helps improving whole detection speed.The actual image Fig. 2 (b) that is far from is so desirable, has a large amount of marginal points to become first kind candidate angular through this step.Below will progressively suppress pseudo-angle point.
The puppet response of the first kind candidate angular of step 2 pair lines and ridge-roof type image border suppresses
A large amount of puppet responses is arranged in first kind candidate angular, comprise lines and ridge-roof type image border point, as shown in Figure 3, such as pixel j, can there be the non-line more that passes of the first kind in k after back, for this this step has designed so pseudo-response rejector:
Be nuclear with each first kind candidate angular respectively and be the center of circle with this center of examining, with 16 pixel units is diameter, make imaginary second class circle mask, search is called the second class scounting line by all imaginary lines of nuclear center and pixel and with these imaginary lines in this second class circle mask.For arbitrary nuclear, if having at least one the second class scounting lines, on this second class scounting line, and the pixel that be positioned at nuclear both sides adjacent with nuclear phase and nuclear then curb described nuclear with value; Otherwise, with described nuclear as the second class candidate angular,
Pixel with shadow representation among Fig. 3 has constituted lines type edge.With wherein j marginal point is example, and 33,34 as the paired check point on the first step circle mask border, owing to all not in the USAN zone of nuclear (j), make j pixel (point) be mistaken as angle point, becomes first kind candidate angular.Herein, marginal point, angle point and pixel are with one deck notion.
Fig. 4 is the pseudo-response theory signal of the inhibition in this step, wherein (round here mask is the circle signal of 6 pixel units with diameter at the round mask of second class of nuclear 41, the situation of 16 pixel units is similar) in, search is by all imaginary lines of nuclear center and pixel.Here the possible imaginary line of institute of whole pixel and nuclear is passed through in a care, for the la among the figure, need check a
1, a
2Whether be worth together with nuclear, if eligible, then described nuclear 33, suppressed if being pseudo-angle point, if not then continuing check lb, check b on this line
1, b
2Whether be worth together,, then described nuclear suppressed, otherwise continue other possible imaginary line of check if eligible with nuclear.As long as condition is not inconsistent, then traversal is by the whole pixel in second class circle mask and all imaginary lines of described nuclear.After having traveled through, if do not deposit the second such class scounting line, on this second class scounting line, and the pixel that be positioned at nuclear both sides adjacent with nuclear phase and nuclear then keep described nuclear with value.
Process is as the method, and except pixel k (true angle point), other lines type marginal point is suppressed entirely among Fig. 3.For pixel k, though also there is one second class scounting line, but the described second class scounting line the most same value pixel of close nuclear is distributed in the one-sided of nuclear, and other direction do not exist have and the nuclear phase neighbour, with the second class scounting line of whole pixel that is worth and is positioned at the nuclear both sides, so do not think that pixel k is pseudo-response, kept, and entered next step screening as the second class candidate angular.
2 supplementary notes are arranged about this step:
Lines type edge is the situation under the artificial figure, and actual photographed image generally can be reflected as the ridge-roof type edge with lines type edge fogization.The ridge-roof type marginal point by as first kind candidate angular, and is suppressed the principle that device suppresses in second step through the first step, and all the situation with lines type marginal point is similarly, does not repeat to illustrate.
Even if the present invention finds these pseudo-response point and also is difficult to eliminate by step afterwards, because the non-line more that passes of the 3rd class of step after they also may have, and continue to be mistaken as is pseudo-angle point, so this step is absolutely necessary to the inhibition that puppets such as lines type, ridge-roof type edge respond.
This step is to operate on whole pixel, calculated amount is little, before the step, follow-up complex steps is still less being carried out on the pixel of ratio (the second class candidate angular) than first kind candidate angular after being placed on, help reducing the overall calculation amount of the inventive method, further improve execution speed.This mode also is the deep reflection that " goes on foot the measure that filters out candidate angular by thick and segmentation, avoided the too early intervention of intensive measure " in the inventive method basic thought.
This step is to operate on whole pixel, so angle point, marginal point, pixel and pixel are with one deck notion.
The check of step 3 scounting line
Sampled point on the step 3.1 pair scounting line is tested
The ultimate principle of the inventive method was demonstrated fully in this step.
Mention in the ultimate principle: acting on one is the round mask of centronucleus with the current pixel.Concrete enforcement of this step be: being nuclear with each second class candidate angular respectively and being the center of circle with this center of examining, is diameter with 16 pixel units, does imaginary the 3rd class and justifies mask.The similar step 1 of measurement of round here mask diameters is that the distance with two ends (on the border) pixel center of the colleague that crosses the center of circle or same column is as the criterion.
Mention in the ultimate principle: search is by the straight line of nuclear in this mask, and such imaginary line is called scounting line.Consideration is unlimited by the imaginary line of nuclear, and reality obviously can not also unnecessaryly be tested to unlimited bar scounting line, in concrete the enforcement, gets n bar scounting line, will justify mask and be divided into the 2n five equilibrium.Owing in other step scounting line is arranged also, be obvious difference, be referred to as " the 3rd class scounting line ", the circle mask is referred to as " the 3rd class circle mask ".The 3rd class search this moment has the n bar as cut-off rule, and n gets 8 or 16 usually.
Mention in the ultimate principle: if exist straight line not pass near the given neighborhood of nuclear other USAN zone in addition, then current pixel point is an angle point.In concrete the enforcement, whether not passing near other USAN zone beyond the given neighborhood nuclear for these imaginary straight lines (cut-off rule), is to justify the pixel that is positioned in the mask on the cut-off rule and sub-pix as Z sampled point P
i(on behalf of a cut-off rule, x y), test, wherein Z is the number of sampled point, and i is the sampled point numbering, i=1,2 ..., Z, x is the horizontal ordinate of sampled point, and y is the ordinate of sampled point, is initial point O with the nuclear center, cut-off rule with the pixel of going together excessively is an X-axis, with the cut-off rule of crossing the same column pixel is Y-axis, sets up the XOY orthogonal coordinate system, and is coordinate unit with 1 pixel unit.
Described sampled point is:
When | tan α | in the time of≤1, sampled point P
iHorizontal ordinate x be the coordinate unit integral multiple,
Sampled point P
iOrdinate y=x*tan α;
When | tan α | in the time of>1, sampled point P
iOrdinate y be the coordinate unit integral multiple,
Sampled point P
iHorizontal ordinate x=y*cot α,
Wherein, α is the angle of cut-off rule and X-axis,
These sampled points and α angle are like Fig. 5.
Mention in the ultimate principle: " near the given neighborhood nuclear ", front claims and technical scheme were once mentioned " prohibiting the inspection circle " simultaneously, did description below here:
In the ideal case, angle point should be a point that does not have size, and the non-ruler of ultimate principle should not pass other USAN zone except this point, can be referring to shown in the ideal case of Fig. 2 (b).But in real image, determine whether that the pixel for angle point is the piece that size is arranged.Moreover, no matter actual object, the still image of taking, angle point is by a large amount of passivation, obfuscation, near the zone that this makes existence and the same value of described angle point true angle point, whether this zone must be considered as the part of angle point, cannot be verified scounting line as the USAN zone and pass; The difference otherwise the point of angle point and other type is had no way of.For the ease of realizing, define near the given neighborhood of nuclear with one around less circle, in this zone as candidate angular (nuclear), owing to comprised the same value district of a part that is regarded as examining, the operation so forbid testing, we are referred to as " prohibiting the inspection circle ", and radius is expressed as: r
n(radius of no testing circle) is as Fig. 5.Prohibiting inspection circle and be with the nuclear center is that the center of circle, radius are that the circle of 4 pixel units is regional.
So far, those are positioned at the USAN zones of prohibiting outside the inspection circle and have just constituted " near other USAN zone the nuclear beyond the given neighborhood ", only prohibit outside the inspection circle in this step to being positioned at, the 3rd class circle mask with cut-off rule on sampled point test.
Described by ultimate principle, the judgement measure concrete in this step is: if all sampled points that have in all cut-off rules of crossing each nuclear center at least one cut-off rule and the described cut-off rule all are positioned at beyond the same value retract USAN zone of the 3rd class circle mask, then described cut-off rule is the non-ruler of the 3rd class, and nuclear on the non-ruler of described the 3rd class is dropped on as the 3rd class candidate angular in the nuclear center; Otherwise, abandon described nuclear.Wherein when each nuclear was judged, described nuclear was as current pixel; The number Z of sampled point is by cut-off rule pitch angle, taboo inspection circle size, the big or small decision of the 3rd class circle mask, as Fig. 5.
Among Fig. 5, r
nWhat represent is to prohibit the inspection radius of a circle, is 4 pixel units.Cut-off rule S meets at the K point with prohibiting the inspection circle, meets at the L point with the circle mask border, and KL has determined the sample range of check point.E, the exact position of hollow dots remarked pixel points such as g.Sub-pix point P between the KL
1(x, y), P
2(x, y), P
3(x, y) and pixel P
4(x y) tests as KL section sampled point.KL pitch angle among the figure | tan α |≤1, sampled point horizontal ordinate x is the coordinate unit integral multiple, and sampled point is in respectively up and down between two whole pixels, and their brightness value can use known single linear interpolation to obtain by two whole pixels up and down.The single linear interpolation need be learned sampled point in the two whole residing positions of pixel, final linear interpolation and sampled point with the distance dependent of pixel up and down, with i sampled point P
i(x y) is example:
Horizontal ordinate x is the coordinate unit integral multiple, ordinate y=x*tan α
The coordinate of whole pixel g be (x,
), the coordinate of whole pixel e be (x,
), wherein
Expression rounds downwards.Sub-pix point P
i(x, brightness value y) is:
Wherein
Obvious 0≤k≤1, k is and sub-pix point P
i(x, y) (at e, between the g) position relevant apart from weights.The coordinate of more than whole pixel and sub-pix point is the value of local coordinate system XOY, and it is the world coordinates of image with coordinate transformation in advance that the acquisition of brightness value also needs, and promptly needs local coordinate is superimposed with the world coordinates of coordinate system XOY initial point O in image.With whole pixel g is example, and its world coordinates in image is
(x wherein
GO, y
GO) be the world coordinates of initial point O (working as pronucleus) in image.So
The acquisition of other whole pixel brightness value also is similar.
For the cut-off rule pitch angle | tan α |>1 situation, sampled point ordinate x is the coordinate unit integral multiple, about sampled point is in respectively between two whole pixels, their brightness value can by about two whole pixels use known single linear interpolation to obtain, method is ditto described.
P among Fig. 5
5(x, y), P
6(x, y), P
7(x, y), P
8(x, y) with KL on sampled point be symmetrically distributed in the opposite side of cut-off rule S about the center of circle.The compute classes of brightness value seemingly.
Whether in the USAN zone of nuclear, the front is existing is introduced, and omits herein for any pixel or sub-pix point.
This step need be observed following 3 rules when filtering: 1) on certain cut-off rule, if a sampled point is arranged in same value retract USAN zone, then described cut-off rule is not the non-ruler of the 3rd class, continues the check of other new cut-off rule; 2) if all sampled points on certain first kind scounting line all are confirmed not in value retract USAN zone, described cut-off rule becomes the non-ruler of the 3rd class, and described nuclear (current pixel) is as the 3rd class candidate angular.3) after the check through all n bar cut-off rules of the 3rd class circle mask, there is not the cut-off rule that can become the non-ruler of the 3rd class, then abandons described nuclear.
Step 3.2 suppresses the pseudo-angle point on the ramp type image border
Real image is owing to the low frequency characteristic of sensor, and the edge blurs basically, and major part is the ramp type edge.After all the sampled point checkout procedures on the previous step cut-off rule, though dispose a large amount of marginal points, still having a small amount of point to be mistaken on the ramp type edge is the 3rd class candidate angular.The present invention finds that on the non-ruler of the 3rd class, the brightness of true angle point is an extreme value along the non-ruler of the 3rd class, its luminance directions derivative
Usually understand the positive negativity of reindexing, amplify the true angle point O of the real image of demonstration as Fig. 6
2Along the situation of the non-ruler of its 3rd class, Fig. 7 (a) is with the situation of change of curve representation I with l.And the ramp type marginal point in the 3rd class candidate angular is the median luminance value point on the non-ruler of the 3rd class, and brightness is not extreme value, common monotone increasing or successively decrease, i.e. and brightness is along the directional derivative of the non-ruler of the 3rd class
Usually can reindexing, amplify the O of the real image of demonstration as Fig. 6
1Point is along the situation of the non-ruler of its 3rd class, and Fig. 7 (b) is with the situation of change of curve representation I with l.Therefore, by judging the luminance directions derivative
Whether reindexing can suppress so pseudo-response of ramp type marginal point along the non-ruler of the 3rd class.For easy, with
The positive negativity of symbol differentiate, in Fig. 6, h=1 or 2.Concrete enforcement is as follows:
Be nuclear with each the 3rd class candidate angular respectively, and be the center of circle, in the 3rd class circle mask, satisfy if cross two sampled point brightness values of outermost end on any one article of the 3rd non-ruler of class at center of each nuclear with this center of examining:
Then the nuclear that the nuclear center is dropped on the non-ruler of described the 3rd class is suppressed; Otherwise, with described nuclear as the 4th class candidate angular,
Wherein: P
j, P
kTwo sampled points of expression outermost end,
The brightness value of representing described sampled point,
Be the brightness value of described nuclear, it gets center (whole pixel) brightness value.
Step 3.3 suppresses to determine the final image angle point based on the non-maximum value of Datong District value distance
In nearly all angular-point detection method, all can face such problem: at the angle point and the neighborhood thereof of real image, often have a group candidate angular rather than only pixel can satisfy the core concept and the condition of correlation method.These angular-point detection methods all use a kind of being called the measure of " non-maximum value inhibition " to determine the most qualified unique angle point at present.So-called non-maximum value suppresses to refer to that the most qualified final all real features certain metric under the correlation method definition should all be the maximal value in the subrange separately, general designation " maximum value ", and near their other separately candidate feature metrics are non-greatly, need to suppress." non-maximum value inhibition " hence obtains one's name.
Yet by ultimate principle of the present invention, the metric that non-maximum value suppresses does not appear can be used as.For this reason, a kind of new non-maximum value inhibition method has been proposed.
Datong District value has at first been proposed apart from d
MH(maximal homo-value distance) this notion: on the non-ruler of certain the 3rd class that has found, from prohibiting the inspection circle outside inward, near a certain section of nuclear must with nuclear with value.On the non-ruler of relevant the 3rd class, with nuclear value together, and point has farthest been determined the Datong District of first kind value distance, because the same value distance that other point is determined is than its little (with regard to a side), (Oa is a pixel as the 4th class candidate angular Oa of Fig. 8, the distance that below relates to Oa and certain point refers to the distance of central point and the described point of pixel Oa) the non-ruler 85 of the 3rd class on, Hw be with Oa with value and apart from Oa point farthest.Distance nuclear point farthest is pixel or sub-pix point.Other same value point between Hw and Oa is all near than the former apart from Oa.In addition, Oa also needs to calculate and Oa distance same value point farthest along the other direction of the non-ruler 85 of the 3rd class, and relatively which is bigger in both sides then, as the Datong District of the final first kind of Oa value apart from d
MHThe Hw of Fig. 8 is the point of the Datong District of the first kind of Oa value distance, and the Datong District of first kind value distance is the distance between Hw and the Oa.
Owing to may have more than one article of the 3rd non-ruler of class to the 4th class candidate angular, Datong District value of stipulating the 4th class candidate angular is apart from d
MHRefer to: a minimum value in the Datong District of the first kind on the non-ruler of all the 3rd classes of the 4th class candidate angular value distance.
The most qualified angle point (real angle point) should have local curvature's maximum according to known notion and definition or the angle point drift angle the sharpest, this has minimum Datong District value apart from d corresponding to the most qualified angle point on the non-ruler of the 3rd class to the present invention is based on this fact-finding
MHDatong District value of having determined the 4th class candidate angular Oa and Ob respectively as the Hw and the Hv point of (each grid among the figure is represented a pixel) among Fig. 8 is apart from d
MH(suppose the Datong District of first kind value distance be exactly final Datong District value apart from) this moment.As the Datong District value of true angle point Ob among Fig. 8 apart from d
MHLittler than near the 4th class candidate angular Oa.In this step, non-maximum value inhibition method with
As metric, promptly real angle point should have maximum in its neighborhood in many candidate angular
Concrete implementation step is as follows:
Datong District value of calculating each the 4th class candidate angular respectively is apart from d
MH,
Design imaginary square window, this window is the center with each the 4th class candidate angular, and window width is 6 pixel units, respectively all candidate angular in each square window relatively
Value, therefrom select maximum
Value, and with maximum
Value corresponding candidate angular as final image angle point, with in each square window all the other
Be worth off-peak candidate angular and curb,
About Datong District value apart from d
MHComputing method, be example with the 4th class candidate angular Oa point among Fig. 8, be described below:
For convenient, suppose the pitch angle of the non-ruler of the 3rd class | tan α |≤1, on the non-ruler of the 3rd class, the point Hw point left and right sides of the Datong District of the first kind of Oa value distance must exist two sampled point Px and Py, their horizontal ordinate is got continuous rounded coordinate value, and the close Px that examines is certain to be worth together with nuclear, and farther Py is then certain to be worth together with nuclear is non-.Sampled point P in similar 3.1 steps of sampled point Px and Py
i(x, y), but Px herein and Py be not defined in and prohibit outside the inspection circle, generally prohibit the inspection circle with interior (have a few cases Py prohibiting outside the inspection circle, but Px one fix on prohibit within the inspection circle).Concrete operations can be along the non-ruler of the 3rd class from prohibiting the inspection circle outside inward, if certain sub-pix point (horizontal ordinate is an integer) and nuclear with value, then with this point as Px, just process with nuclear non-with the sub-pix point that is worth as Py.The compute classes of sampled point Px and Py brightness value is like the sampled point P in 3.1 steps
i(x, calculating y) adopt known single linear interpolation to get final product.
Can be with the distance of value point Hw and Px by the brightness value of Px and Py, and the distance between the PxPy, obtain by known single linear interpolation.Single linear interpolation herein (or claim linear interpolation) though on the principle linear interpolation with the front identical, but the linear interpolation of front is the brightness value that point to be calculated is calculated in the brightness according to 2, the to be calculated some position between 2 o'clock, promptly obtains brightness value by position (in other words with distance) at 2; Then be herein according to determine with the brightness value of value point Hw Hw and Px apart from d
HP
So need to determine with the definite brightness value that is worth some Hw earlier.Same value among the present invention be not with nuclear (the 4th class candidate angular) be strict same brightness value, but by a close brightness value of the threshold value t of USAN decision.The brightness value of Px is near the scope of nuclear brightness value that threshold value t determines, Hw then must be on the border of the definite nuclear brightness value environs of threshold value t.The calculating formula of determining is:
If
Then
Otherwise,
Hw and Px apart from d
HPCan by
And the distance between the PxPy (| tan α | under≤1 situation, this value is
Parasang: pixel), negating to separate according to known linear interpolation way obtains, and omits herein.So the same value of a Hw and nuclear is apart from d
OH=d
HP+ d
OP, d wherein
OPThe distance of expression Px to Oa, it can be by the horizontal ordinate of Px simply divided by | cos α | obtains.Both sides at nuclear all need to calculate so same value distance, with the maximum as the Datong District of the first kind of nuclear value distance; If there are many non-rulers, then calculate the Datong District of first kind value distance of all non-rulers, with the reckling in them as final Datong District value apart from d
MH
Claims (1)
1. the line search formula detection method of an image angle point is characterized in that step is as follows:
Step 1 inceptive filtering
Serve as nuclear with each pixel respectively and be the center of circle with this center of examining, with 8 pixel units is diameter, make imaginary first kind circle mask, search is called first kind scounting line by the imaginary line at nuclear center and with these imaginary lines in this first kind circle mask, being check point about the paired pixel examined on the centrosymmetric round mask border, if the same value retract USAN in described first kind circle mask can find paired pixel as check point with exterior domain, the first kind scounting line of then crossing described paired pixel is the non-ruler of the first kind, and the nuclear that the nuclear center is dropped on the non-ruler of the first kind is first kind candidate angular; Otherwise, abandon described nuclear.
The puppet response of the first kind candidate angular of step 2 pair lines and ridge-roof type image border suppresses
Be nuclear with each first kind candidate angular respectively and be the center of circle with this center of examining, with 16 pixel units is diameter, make imaginary second class circle mask, search is called the second class scounting line by all imaginary lines of nuclear center and pixel and with these imaginary lines in this second class circle mask, if the pixel that has adjacent with nuclear phase at least one the second class scounting lines and be positioned at the nuclear both sides and nuclear with value, then curb described nuclear; Otherwise, with described nuclear as the second class candidate angular,
The check of step 3 scounting line
Sampled point on the step 3.1 pair scounting line is tested
Be nuclear with each second class candidate angular respectively and be the center of circle with this center of examining, with 16 pixel units is diameter, make imaginary the 3rd class circle mask, search is called the 3rd class scounting line by the imaginary line of nuclear and with these imaginary lines in the 3rd class circle mask, get n article of the 3rd class scounting line as cut-off rule, the 3rd class circle mask is divided into the 2n equal portions, and n is for cutting apart number of lines, will be positioned on the cut-off rule again and be positioned at the pixel of round mask and the sub-pix point as Z sampled point P
i(x, y), Z is the number of sampled point, i is the sampled point numbering, i=1,2,, Z, x are the horizontal ordinate of sampled point, y is the ordinate of sampled point, is initial point O with the nuclear center, is X-axis with the cut-off rule of crossing the pixel of going together, with the cut-off rule of crossing the same column pixel is Y-axis, sets up the XOY orthogonal coordinate system, and is coordinate unit with 1 pixel unit, prohibit outside the inspection circle to being positioned at again, the 3rd class circle mask with sampled point test
If all sampled points that have in all cut-off rules of crossing each nuclear center at least one cut-off rule and the described cut-off rule all are positioned at beyond the same value retract USAN zone of the 3rd class circle mask, then described cut-off rule is the non-ruler of the 3rd class, and nuclear on the non-ruler of described the 3rd class is dropped on as the 3rd class candidate angular in the nuclear center; Otherwise, abandon described nuclear,
Described taboo inspection circle is to be that the center of circle, radius are that the circle of 4 pixel units is regional with the nuclear center,
Described sampled point is:
When | tan α | in the time of≤1, sampled point P
iHorizontal ordinate x be the coordinate unit integral multiple,
Sampled point P
iOrdinate y=x*tan α;
When | tan α | in the time of>1, sampled point P
iOrdinate y be the coordinate unit integral multiple,
Sampled point P
iHorizontal ordinate x=y*cot α,
Wherein, α is the angle of cut-off rule and X-axis,
Step 3.2 suppresses the pseudo-angle point on the ramp type image border
Be nuclear with each the 3rd class candidate angular respectively, and be the center of circle, in the 3rd class circle mask, satisfy if cross two sampled point brightness values of outermost end on any one article of the 3rd non-ruler of class at center of each nuclear with this center of examining:
Then the nuclear that the nuclear center is dropped on the non-ruler of described the 3rd class is suppressed; Otherwise, with described nuclear as the 4th class candidate angular,
Wherein: P
j, P
kTwo sampled points of expression outermost end,
The brightness value of representing described sampled point, I
NBe the brightness value of described nuclear,
Step 3.3 suppresses to determine the final image angle point based on the non-maximum value of Datong District value distance
Datong District value of calculating each the 4th class candidate angular respectively is apart from d
MH,
Design imaginary square window, this window is the center with each the 4th class candidate angular, and window width is 6 pixel units, respectively all candidate angular in each square window relatively
Value, therefrom select maximum
Value, and with maximum
Value corresponding candidate angular as final image angle point, with in each square window all the other
Be worth off-peak the 4th class candidate angular and curb,
Datong District value of described each the 4th class candidate angular is apart from d
MHRefer to: a minimum value in the Datong District of the first kind of the non-ruler of all the 3rd classes of each the 4th class candidate angular value distance,
The Datong District of described first kind value distance refers to: on arbitrary article the 3rd non-ruler of class of four class candidate angular, in prohibiting the inspection circle, and nuclear with value and apart from the nuclear center farthest point and the distance between the nuclear center,
Described distance nuclear point farthest is pixel or sub-pix point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101100672A CN101794446B (en) | 2010-02-11 | 2010-02-11 | Line search type detection method of image corner point |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101100672A CN101794446B (en) | 2010-02-11 | 2010-02-11 | Line search type detection method of image corner point |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101794446A true CN101794446A (en) | 2010-08-04 |
CN101794446B CN101794446B (en) | 2011-12-14 |
Family
ID=42587118
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010101100672A Expired - Fee Related CN101794446B (en) | 2010-02-11 | 2010-02-11 | Line search type detection method of image corner point |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101794446B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102339464A (en) * | 2011-09-01 | 2012-02-01 | 上海大学 | Line search corner detection method |
GB2514397A (en) * | 2013-05-23 | 2014-11-26 | Linear Algebra Technologies Ltd | Corner detection |
CN104992144A (en) * | 2015-06-11 | 2015-10-21 | 电子科技大学 | Method for distinguishing transmission line from road in remote sensing image |
CN105787912A (en) * | 2014-12-18 | 2016-07-20 | 南京大目信息科技有限公司 | Classification-based step type edge sub pixel localization method |
CN106682678A (en) * | 2016-06-24 | 2017-05-17 | 西安电子科技大学 | Image angle point detection and classification method based on support domain |
CN107909555A (en) * | 2017-11-27 | 2018-04-13 | 北京大恒图像视觉有限公司 | A kind of gridding noise elimination method for keeping acutance |
CN112308797A (en) * | 2020-10-30 | 2021-02-02 | 维沃移动通信有限公司 | Corner detection method and device, electronic equipment and readable storage medium |
CN113643176A (en) * | 2021-07-28 | 2021-11-12 | 沈阳先进医疗设备技术孵化中心有限公司 | Rib display method and device |
CN113888456A (en) * | 2020-07-01 | 2022-01-04 | 长春工业大学 | Corner detection algorithm based on contour |
CN116228634A (en) * | 2022-12-07 | 2023-06-06 | 辉羲智能科技(上海)有限公司 | Distance transformation calculation method, application, terminal and medium for image detection |
CN117114971A (en) * | 2023-08-01 | 2023-11-24 | 北京城建设计发展集团股份有限公司 | Pixel map-to-vector map conversion method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20050008247A (en) * | 2003-07-14 | 2005-01-21 | (주)워치비젼 | An Apparatus and Method for mixing 3D graphic images in sports video |
CN101216941A (en) * | 2008-01-17 | 2008-07-09 | 上海交通大学 | Motion estimation method under violent illumination variation based on corner matching and optic flow method |
CN101251926A (en) * | 2008-03-20 | 2008-08-27 | 北京航空航天大学 | Remote sensing image registration method based on local configuration covariance matrix |
CN101458814A (en) * | 2007-12-13 | 2009-06-17 | 东软集团股份有限公司 | Method and apparatus for separating objects interesting region from image |
CN101477687A (en) * | 2009-01-22 | 2009-07-08 | 上海交通大学 | Checkerboard angle point detection process under complex background |
-
2010
- 2010-02-11 CN CN2010101100672A patent/CN101794446B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20050008247A (en) * | 2003-07-14 | 2005-01-21 | (주)워치비젼 | An Apparatus and Method for mixing 3D graphic images in sports video |
CN101458814A (en) * | 2007-12-13 | 2009-06-17 | 东软集团股份有限公司 | Method and apparatus for separating objects interesting region from image |
CN101216941A (en) * | 2008-01-17 | 2008-07-09 | 上海交通大学 | Motion estimation method under violent illumination variation based on corner matching and optic flow method |
CN101251926A (en) * | 2008-03-20 | 2008-08-27 | 北京航空航天大学 | Remote sensing image registration method based on local configuration covariance matrix |
CN101477687A (en) * | 2009-01-22 | 2009-07-08 | 上海交通大学 | Checkerboard angle point detection process under complex background |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102339464A (en) * | 2011-09-01 | 2012-02-01 | 上海大学 | Line search corner detection method |
GB2514397A (en) * | 2013-05-23 | 2014-11-26 | Linear Algebra Technologies Ltd | Corner detection |
GB2514397B (en) * | 2013-05-23 | 2017-10-11 | Linear Algebra Tech Ltd | Corner detection |
CN105787912A (en) * | 2014-12-18 | 2016-07-20 | 南京大目信息科技有限公司 | Classification-based step type edge sub pixel localization method |
CN105787912B (en) * | 2014-12-18 | 2021-07-30 | 南京大目信息科技有限公司 | Classification-based step type edge sub-pixel positioning method |
CN104992144A (en) * | 2015-06-11 | 2015-10-21 | 电子科技大学 | Method for distinguishing transmission line from road in remote sensing image |
CN104992144B (en) * | 2015-06-11 | 2018-05-29 | 电子科技大学 | The differentiating method of power transmission line and highway in remote sensing images |
CN106682678A (en) * | 2016-06-24 | 2017-05-17 | 西安电子科技大学 | Image angle point detection and classification method based on support domain |
CN106682678B (en) * | 2016-06-24 | 2020-05-01 | 西安电子科技大学 | Image corner detection and classification method based on support domain |
CN107909555A (en) * | 2017-11-27 | 2018-04-13 | 北京大恒图像视觉有限公司 | A kind of gridding noise elimination method for keeping acutance |
CN107909555B (en) * | 2017-11-27 | 2020-06-02 | 北京大恒图像视觉有限公司 | Sharpness-keeping grid noise elimination method |
CN113888456A (en) * | 2020-07-01 | 2022-01-04 | 长春工业大学 | Corner detection algorithm based on contour |
CN113888456B (en) * | 2020-07-01 | 2024-05-24 | 长春工业大学 | Corner detection method based on contour |
CN112308797A (en) * | 2020-10-30 | 2021-02-02 | 维沃移动通信有限公司 | Corner detection method and device, electronic equipment and readable storage medium |
CN112308797B (en) * | 2020-10-30 | 2024-02-02 | 维沃移动通信有限公司 | Corner detection method and device, electronic equipment and readable storage medium |
CN113643176A (en) * | 2021-07-28 | 2021-11-12 | 沈阳先进医疗设备技术孵化中心有限公司 | Rib display method and device |
CN113643176B (en) * | 2021-07-28 | 2024-05-28 | 东软医疗***股份有限公司 | Rib display method and device |
CN116228634A (en) * | 2022-12-07 | 2023-06-06 | 辉羲智能科技(上海)有限公司 | Distance transformation calculation method, application, terminal and medium for image detection |
CN116228634B (en) * | 2022-12-07 | 2023-12-22 | 辉羲智能科技(上海)有限公司 | Distance transformation calculation method, application, terminal and medium for image detection |
CN117114971A (en) * | 2023-08-01 | 2023-11-24 | 北京城建设计发展集团股份有限公司 | Pixel map-to-vector map conversion method and system |
CN117114971B (en) * | 2023-08-01 | 2024-03-08 | 北京城建设计发展集团股份有限公司 | Pixel map-to-vector map conversion method and system |
Also Published As
Publication number | Publication date |
---|---|
CN101794446B (en) | 2011-12-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101794446B (en) | Line search type detection method of image corner point | |
CN101887586B (en) | Self-adaptive angular-point detection method based on image contour sharpness | |
CN108986126A (en) | The center of circle detection method of RANSAC algorithm is detected and improved based on Gauss curve fitting sub-pixel edge | |
CN107248159A (en) | A kind of metal works defect inspection method based on binocular vision | |
CN105096299B (en) | Polygon detecting method and polygon detecting device | |
CN101499168B (en) | Structured light strip center extraction method based on ridge line tracing and Hessian matrix | |
CN106530347A (en) | Stable high-performance circle feature detection method | |
CN109859226A (en) | A kind of detection method of the X-comers sub-pix of figure segmentation | |
CN109816051B (en) | Hazardous chemical cargo feature point matching method and system | |
CN108830899A (en) | A kind of center of circle detection method that the quick self-adapted angle point based on geometrical characteristic combines | |
CN104715491A (en) | Subpixel edge detection method based on one-dimensional gray moment | |
CN109902576A (en) | A kind of training method and application of head shoulder images classifier | |
CN104282001A (en) | Method for enhancing image feature two-value descriptor performance | |
CN106339710A (en) | Edge width defect detection method and edge width defect detection device | |
CN109325958A (en) | A kind of offshore ship detection method for refining and improve generalised Hough transform based on profile | |
CN102289677A (en) | Method for analyzing image based on principal component analysis and method applicable to detection of defects of fabric | |
CN110390338A (en) | A kind of SAR high-precision matching process based on non-linear guiding filtering and ratio gradient | |
CN108734059A (en) | indoor mobile robot object identification method | |
CN108629227A (en) | The method and system on left and right vehicle wheel boundary are determined in the picture | |
CN107480710A (en) | Feature Points Matching result processing method and device | |
Kruglov | The algorithm of the roundwood volume measurement via photogrammetry | |
CN103955929B (en) | Image local edge pattern and non-edge mode judging method and judgment means | |
CN106373117B (en) | A kind of quickly more circle detection methods | |
CN105205826B (en) | A kind of SAR image azimuth of target method of estimation screened based on direction straight line | |
Hao et al. | Circle area detection based on convolutional neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20100804 Assignee: JIANGSU YUYUE MEDICAL EQUIPMENT&SUPPLY Co.,Ltd. Assignor: Southeast University Contract record no.: 2013320000104 Denomination of invention: Line search type detection method of image corner point Granted publication date: 20111214 License type: Exclusive License Record date: 20130314 |
|
LICC | Enforcement, change and cancellation of record of contracts on the licence for exploitation of a patent or utility model | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20111214 |