CN1256696C - Image processing method of detecting ellipsoid by utilizing restricted random Huff transition - Google Patents
Image processing method of detecting ellipsoid by utilizing restricted random Huff transition Download PDFInfo
- Publication number
- CN1256696C CN1256696C CN 200410017365 CN200410017365A CN1256696C CN 1256696 C CN1256696 C CN 1256696C CN 200410017365 CN200410017365 CN 200410017365 CN 200410017365 A CN200410017365 A CN 200410017365A CN 1256696 C CN1256696 C CN 1256696C
- Authority
- CN
- China
- Prior art keywords
- image
- curve
- oval
- ellipse
- node
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to an image processing method for elliptical detection by using restrictive random Hough transformation, which comprises that the noise influence of an image to be tested is removed by median filtering and an image edge is picked up by using Canny operators, and the image is thinned based on a thinning method of a mould plate to make the image only contain the width of a single pixel; then, the image is converted into vector graphics to obtain the detailed information of every closed curve or arbitrary separate curve on the image, and by using the information, restrictive random Hough transformation is used so as to detect the curves in the image one by one and verify whether each curve is an ellipse or not. In the process of detection, random select sampling is carried out within the range of points on every curve to obtain detected ellipse parameter sets, and every effective ellipse parameter is taken out from the parameter sets. The present invention purposefully samples within each curve segment, the present invention not only reduces the range of sampling points, but also reduces the threshold value of each aspect in the process of detecting the ellipse, and thus, the present invention is favorable to fast and accurately detect the ellipse.
Description
Technical field
The present invention relates to a kind of image processing method that utilizes restricted randomized hough transform to carry out oval detection, be used for Flame Image Process, computer vision and industrial automation check.Belong to the computerized information technical field of image processing.
Background technology
Detect ellipse quickly and accurately at computer vision and area of pattern recognition, particularly industry is made, biomedicine, and aspects such as Automated inspection and assembling have a wide range of applications.Since Hough proposed Hough transformation in 1962, but it just develops into a kind of method that detection of straight lines and circle/ellipse wait other figure very soon.But traditional Hough transformation has several bigger defectives: 1, calculated amount is big; 2, committed memory is big; 3, the parameter of Ti Quing is restricted by the quantized interval of parameter space.If Hough transformation is directly used in oval the detection, because ellipse has 5 free parameters, need to accumulate at five dimension parameter spaces, cause this way because of calculated amount and the excessive reality that do not conform to of memory demand.
In order to overcome above-mentioned defective, Xu etc. have proposed randomized hough transform (Xu L, O ja E.Randomizedhough transform (RHT): basic mechanisms, algorithms and computationalcomplexities, Computer Vision Graphic Image Process:Image understanding, 1993,57 (2): 131-154.).Randomized hough transform is chosen not the point that point-blank several points are mapped to parameter space randomly at image space, constitutes manyly to one mapping, calculates the elliptic parameter that satisfies selected point then.The method has been avoided the huge calculated amount of standard Hough transformation one to many mappings, has certain superiority.Yet randomized hough transform detects performance when processing has a plurality of circles or ellipse not good, during for complicated image, when particularly being subjected to the image of noise, because aimless stochastic sampling can be introduced a large amount of invalid samplings and useless accumulation, the randomness of testing result is very big, time loss is too much, and algorithm performance is reduced greatly.
Summary of the invention
The objective of the invention is at the deficiencies in the prior art, a kind of image processing method that utilizes restricted randomized hough transform to carry out oval detection is proposed, effectively reduce the sampling of Null Spot, improve accuracy of detection oval in the image, significantly reduce the processing time simultaneously.
For realizing such purpose, in the technical scheme of the present invention, at first carry out the image pre-service, image to be detected is removed noise effect and adopted Canny operator extraction image border by medium filtering, and adopt thinning method to make it comprise single pixel wide image thinning based on template; Adopt clustering method to be converted into vector graphics the image after the refinement then, the information of every closed curve or any independent curve and curve sum on the vector graphics document image comprise every curve terminal, nodal point number on every curve, the node coordinate, node connects information such as table; Application limitations randomized hough transform in vector graphics, come every in image curve is detected one by one and verifies whether satisfy elliptic equation, in testing process, only in image, select sampling at random in the scope of the point on every curve, be used for oval the detection and handle.Detected all curves, the elliptic parameter collection that is all detected is concentrated each the oval parameter that is finally detected from elliptic parameter at last.
Ellipse detection method of the present invention mainly comprises following concrete steps:
1. image pre-service
The present invention at first carries out the image pre-service.For original image is the situation of gray level image, earlier it is carried out medium filtering and handles, and adopts the Canny operator to carry out rim detection then, thereby makes it become binary map, and promptly the pixel value on the edge is 1 (white), and other pixel value is 0 (black).If original image itself is exactly a two-value, so it is carried out Filtering Processing.Then the clean bianry image that obtains after the Filtering Processing is carried out Refinement operation based on template again, make refinement after image only have single pixel wide.
2. translated image is a vector graphics
Image after the refinement is adopted the figure clustering method, from top to bottom, from left to right, each pixel is transformed into predefined graphical nodes.Graphical nodes with seven attributes describe locations of pixels and with the relation of other pixel.These attributes comprise node ID, node place curve number, node location (x, y), type, chained list, deleted marker, link curve table.Node ID is a sequence node number in the expression set of node, the curve at curve number record node place, node place, node location is x, the y coordinate of node in image, the node type attribute record is the number of other node of being connected of node therewith, chained list then writes down the situation that links to each other of node and other node, and deleted marker represents whether this node is effective in next step is handled, if it is 1, represent useless, directly the deletion get final product.The link curve table then represented when node type genotype value greater than 2 the time, the crossing curve number at the node place, it is used to handle the curve intersection situation.Further merge then and the analyzed pattern node, the node that will belong to same independence or discrete curve is classified as one group, with this entire image is represented with one group of sets of curves.By above processing, obtain the details of every curve on the image and the integrated information of entire image, promptly obtain the vector graphics of image.
3. the ellipse in the application limitations randomized hough transform detected image in vector graphics
According to the vector figure data that obtains, whether utilize restricted randomized hough transform to detect this curve in order one by one to every in vector graphics curve is oval, its concrete grammar is: 3 of picked at random are come the fitting parameter curve in the coordinate range at every given curve place, i.e. the fitted ellipse equation.If these 3 are satisfied elliptic equation, just they are added in the totalizer, and give a certain weights according to fitting degree.Repeat to select carry out fit procedure and surpass predetermined threshold at 3 up to detecting the pixel point set number that satisfies elliptic equation, from totalizer, select then have best weights parameter set in order to this curve in the presentation video.If the parameter set in the totalizer is to detect the parameter set that obtains similar, the value of average this parameter set then is with the raw parameter collection in this new parameters sets replacement totalizer.Find out after the parameter set, also need parameter set taken and go checking in the image, promptly calculate how many points in the original image and drop on really on the ellipse by this parameter set definition, if satisfy predetermined threshold, then be actual parameter, otherwise be invalid.After having detected a curve as stated above, move to rapidly and carry out the next round detection in next bar curve ranges.By this constantly detection method among a small circle movably, and the ellipse of finishing entire image detects, the parameter set that all that obtain detecting are oval.
4. the oval parameter of each that from parameter set, is finally detected
Each elliptic parameter comprises p, q, r1, five parameters of r2 and θ, p wherein, q are oval by the coordinate after the former heart skew, and r1 is oval x axle radius, r2 is oval y axle radius, and θ is oval along the clockwise deflection angle of x axle, has just determined position and the shape that each is oval by these five parameters.
Method of the present invention is simply effective, and its key is to grasp the global information of image by treating the pre-service of detected image, and then adopts the randomized hough transform of improved to carry out fast detecting.Utilize the present invention to carry out ellipse and detect, in each segment of curve, sample targetedly, detect and avoided in global scope, blindly carrying out stochastic sampling point, thereby reduced choosing of Null Spot, reduced computing time, improved accuracy of detection simultaneously.Method of the present invention detects cell in medical image experimental applications shows, can successfully detect the elliptical erythrocyte in the image in the short period of time, and it is detected as power and is better than classic method, and false drop rate also reduces greatly.
Description of drawings
The medical image original to be detected that Fig. 1 adopts for the embodiment of the invention.
Fig. 2 is the connected pixel number of the present invention's definition when carrying out the image thinning operation.
Fig. 3 is the template of the present invention's definition when carrying out the image thinning operation.
Fig. 4 calculates the elliptical center synoptic diagram for utilizing randomized hough transform.
Fig. 5 utilizes randomized hough transform to detect the figure as a result of elliptical erythrocyte in the original image.
Embodiment
In order to understand technical scheme of the present invention better, be described in further detail below in conjunction with accompanying drawing.
With Fig. 1 is example, and the present invention detects the elliptical erythrocyte in the medical image, is undertaken by following concrete steps:
1, image pre-service.Image 1 is a gray level image, earlier it is carried out medium filtering and handles, and adopts the Canny operator to carry out rim detection then, thereby makes it become binary map, and promptly the pixel value on the edge is 1 (white), and other pixel value is 0 (black).If original image is not a gray level image, be two-value, so it is carried out Filtering Processing.Then carry out Refinement operation based on template, refinement need be used this definition of pixel Betti number: the change number (from 1 to 0) of numerical value when the Betti number of a pixel is defined as its 8 neighborhoods of clockwise visit, referring to Fig. 2, template is referring to Fig. 3, and its refinement step is as follows:
(1) in original image, finds the pixel of matching template M1;
(2) if center pixel is not a terminal point, and to have Betti number be 1, and then this pixel of mark for future use;
(3) repeat (1), (2) step makes all pixels all carry out the M1 template matches;
(4) repeat (1), (2), (3) step makes it to M2, M3, and the M4 template is all taked the same treatment method;
(5) if any pixel is marked as back usefulness, then changing its pixel value is 0;
(6) if in (5) step, there is any pixel value to take place to change, then begins whole process repeated, otherwise then finish from the first step.
By above method, finish the image pre-service.
2, the image after the refinement is converted into vector graphics.At first, seven attributes of definition graphical nodes, scan the image after the refinement then from left to right, from top to bottom, when running into pixel, (refer to that here pixel value is 1 white pixel, below all to represent this meaning), writing down this pixel is 0 as a node and initialization node types attribute, and the sequence number of giving this node place curve simultaneously is 1.Continue scanning, when running into a new pixel again, check the left of current pixel point, the upper left side, top and top-right neck contact are if arbitrary abutment points is a pixel, then type attribute is increased 1, note the node number of this abutment points simultaneously and deposit in the chained list attribute of front contact.And the type attribute of abutment points also will increase 1, and notes present node and number deposit in current its chained list attribute, thereby generates a two-way diagrammatic representation.In addition, if one or more identical curve number attributes that have are arranged in the abutment points, then copy this curve number in the curve number attribute of present node.If abutment points has different curve number attributes, the curve number that then copies the abutment points of present node top (also can be other three directions) arrives present node, simultaneously this curve number is deposited in the table of equal value with aftertreatment.If near present node, do not find abutment points, think that then this node is an initial node, give its new place curve number and node number, and the place curve number also is deposited in the table of equal value.After having scanned all pixels, handle the curve number of record in the table of equal value, eliminate their numerical value gap.Rescan image, come the place curve number attribute of pixel in the alternative image successively with the curve number of ascending arrangement in the table of equal value.When handling all pixels, rescan image, and to remove delete property be 1 pixel.Thereby the image after the refinement is converted into the figure with certain topological structure, and this figure comprises node sum in the image, independent curve sum, node coordinate, information such as relativeness between contained node number of curve and node.
3, the ellipse in the application limitations randomized hough transform detected image in vector graphics.The randomized hough transform algorithm is as follows:
(1) select on the image border 3 as x1 at random, x2, x3 sees Fig. 4.
(2) seek elliptical center.This realizes by following method: obtain the tangent line tan1 by 3, tan2, tan3; Obtain the intersection point t1 of tan1 and tan2 and the intersection point t2 of tan2 and tan3; Obtain x1x2 mid point m1 and x2x3 mid point m2.Like this, elliptical center just drops on the intersection point O of t1m1 and t2m2, as Fig. 4.If can not find the center, then turn back to (1).
(3) found elliptical center after, the elliptical center point is moved to initial point (0,0).
(4) according to ax
2+ 2bxy+cy
2=1, by separate following equation (x1=(and x1, y1) x2=(x2, y2) x3=(x3, y3)), calculating parameter a, b, c, is back to (1) if do not separate.
(5) checking ac-b
2＞0, if true, then be oval, otherwise be not oval.
(6) calculate elliptic parameter p, q, r1, r2 and θ, p wherein, q be oval coordinate after being shifted by the former heart, and r1 is oval x axle radius, and r2 is oval y axle radius, and θ is that ellipse is along the clockwise deflection angle of x axle.These parameters satisfy elliptic equation:
(7) the parameter set chained list got of search, if the value of its value of a parameter set near the parameter set of newly obtaining arranged in the parameter set chained list, the value in just average these two parameter sets, and with the number increase of parameter set chained list.Each parameter set has a so-called counter, and it has found the value of what similar (equally) in order to explanation.If do not find similar value, just be added to this parameter set in the parameter set chained list and value that the parameter set chained list is set is 1.
(8) behind the certain number of times of process, analytical parameters collection chained list.Get each its cumulative number and satisfy the value of the parameter set of certain threshold value, and verify them.Here the checking of saying refers to removes to calculate and drops on really on the ellipse by this parameter set definition for how many points in the two-value original image.This kind searched need certain tolerance, and pixel count is by r1 and r2 standardize (saying exactly, should be that the oval pixel count found gone up is divided by by the pixel number on the rectangle of 2*r1 and 2*r2 decision).After gained standardization numerical value satisfies certain threshold value, can think oval and exist really.
(9), then the point on these ellipses is removed from former figure, with simplified image if detect ellipse.Emptying the parameter set chained list then begins another and takes turns detection (promptly turning to (1)).
According to the method described above, treated image being carried out ellipse as input detects.When carrying out the ellipse detection, according to the information that obtains curve, pixel not at random on a large scale in the sampling, but loop iteration carries out among a small circle, in the time of for the first time, select on article one curve 3 as alternative giving the randomized hough transform normal process, and carry out randomized hough transform and verify.In case after the checking, can jump out and continue to handle the second curve at once.In like manner handle all curves on the image.Deposit in the parameter set detecting successful elliptic parameter.By the method, image is had a few all and is handled checking by the order of curve number.Because the present invention has dwindled range of choice by every curve of circular treatment when getting at random, got rid of Null Spot simultaneously, like this, just saved detection time, improved accuracy of detection.Simultaneously, because the point of selecting itself is exactly on a curve, or be oval, or be not, judge that than being easier to therefore, some corresponding threshold parameters can suitably be established low, neither influence accuracy of detection, also reduced detection time simultaneously.
4, the oval parameter of each that from parameter set, is finally detected.Each elliptic parameter comprises p, and q, r1, r2 and θ, p wherein, q are that oval r1 is oval x axle radius by the coordinate after the former heart skew, and r2 is oval y axle radius, and θ is oval along the clockwise deflection angle of x axle.Thereby position and shape that each is oval have been determined.By parameter ellipse is repainted one time, draw the ellipse that detects, as Fig. 5.
Claims (1)
1, a kind of image processing method that utilizes restricted randomized hough transform to carry out oval detection is characterized in that comprising following concrete steps:
1) image pre-service: for gray level image, carry out medium filtering earlier and carry out the Canny operator edge detection, make image become binary map, for bianry image, only carry out medium filtering, adopt then based on the method for template bianry image is carried out refinement, make refinement after image have only single pixel wide;
2) image after the conversion refinement is a vector graphics: adopt clustering method, each pixel of image after the refinement is transformed into graphical nodes, the graphical nodes node ID, node place curve number, node location, type, chained list, deleted marker and link curve table seven attributes describe locations of pixels and with the relation of other pixel, further merge then and the analyzed pattern node, the node that will belong to same independence or discrete curve is classified as one group, with this entire image is represented with one group of sets of curves, obtain the details of every curve on the image and the integrated information of entire image, promptly obtain the vector graphics of image;
3) ellipse in the application limitations randomized hough transform detected image in vector graphics: according to the vector figure data that obtains, 3 of picked at random are come the fitting parameter curve in the coordinate range at every given curve place, if these 3 are satisfied elliptic equation, just they are added in the totalizer, and give a certain weights according to fitting degree, repeat to select 3 execution fit procedure to surpass predetermined threshold up to detecting the pixel point set number that satisfies elliptic equation, from totalizer, select then have best weights parameter set in order to this curve in the presentation video, if the parameter set in the totalizer is to detect the parameter set that obtains similar, the value of then average this parameter set, with the raw parameter collection in this new parameters sets replacement totalizer, the parameter set of finding out is taken gone checking in the image at last; After having detected a curve as stated above, move to rapidly and carry out the next round detection in next bar curve ranges, by this constantly detection method among a small circle movably, the ellipse of finishing entire image detects the parameter set that all that obtain detecting are oval;
4) from parameter set, obtain final detected p, q, r1, five elliptic parameters of r2 and θ, thereby determine each oval position and shape, p wherein, q be oval coordinate after being offset by the former heart, r1 is oval x axle radius, and r2 is oval y axle radius, and θ is oval along the clockwise deflection angle of x axle.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410017365 CN1256696C (en) | 2004-04-01 | 2004-04-01 | Image processing method of detecting ellipsoid by utilizing restricted random Huff transition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410017365 CN1256696C (en) | 2004-04-01 | 2004-04-01 | Image processing method of detecting ellipsoid by utilizing restricted random Huff transition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1564190A CN1564190A (en) | 2005-01-12 |
| CN1256696C true CN1256696C (en) | 2006-05-17 |
Family
ID=34478936
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200410017365 Expired - Fee Related CN1256696C (en) | 2004-04-01 | 2004-04-01 | Image processing method of detecting ellipsoid by utilizing restricted random Huff transition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1256696C (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8861206B2 (en) | 2009-03-13 | 2014-10-14 | Osram Gesellschaft Mit Beschraenkter Haftung | Casing for electronic ballast |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101388077A (en) * | 2007-09-11 | 2009-03-18 | 松下电器产业株式会社 | Target shape detecting method and device |
| CN101546425B (en) * | 2009-05-05 | 2011-04-13 | 广东工业大学 | A vector filtering method of double-threshold type color image |
| CN101777127B (en) * | 2010-03-11 | 2011-11-23 | 成都三泰电子实业股份有限公司 | Human body head detection method |
| CN102236894A (en) * | 2010-04-30 | 2011-11-09 | 西门子公司 | Circle detection method and device |
| CN101882318B (en) * | 2010-06-22 | 2012-04-18 | 重庆大学 | Format conversion method from PLT-format picture into numerical control punching vector graphic element |
| CN101908153B (en) * | 2010-08-21 | 2012-11-21 | 上海交通大学 | Method for estimating head postures in low-resolution image treatment |
| CN102147857A (en) * | 2011-03-22 | 2011-08-10 | 黄晓华 | Image processing method for detecting similar round by using improved hough transformation |
| CN108918532B (en) * | 2018-06-15 | 2021-06-11 | 长安大学 | System and method for detecting damage of expressway traffic sign |
| CN109272518B (en) * | 2018-08-17 | 2020-05-05 | 东南大学 | Morris water maze experiment image analysis system and method |
| CN111563925B (en) * | 2020-05-07 | 2023-08-11 | 大连理工大学 | Ellipse detection acceleration method based on generalized Pascal mapping |
-
2004
- 2004-04-01 CN CN 200410017365 patent/CN1256696C/en not_active Expired - Fee Related
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8861206B2 (en) | 2009-03-13 | 2014-10-14 | Osram Gesellschaft Mit Beschraenkter Haftung | Casing for electronic ballast |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1564190A (en) | 2005-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596066B (en) | Character recognition method based on convolutional neural network | |
| EP1678659B1 (en) | Method and image processing device for analyzing an object contour image, method and image processing device for detecting an object, industrial vision apparatus, smart camera, image display, security system, and computer program product | |
| CN1256696C (en) | Image processing method of detecting ellipsoid by utilizing restricted random Huff transition | |
| Liu et al. | Identification of fork points on the skeletons of handwritten Chinese characters | |
| CN1885317A (en) | Adaptive edge detection method based on morphology and information entropy | |
| CN103235929B (en) | Identification method and identification device on basis of hand vein images | |
| CN113177959B (en) | QR code real-time extraction method in rapid movement process | |
| CN105931252A (en) | Ellipse rapid detection method based on geometric constraint | |
| CN107452035B (en) | Method and apparatus for analyzing lane line image and computer readable medium thereof | |
| CN103020626A (en) | Method and system for detecting elliptical target in image | |
| CN105022989A (en) | Robust extended local binary pattern texture feature extraction method | |
| CN105184225A (en) | Multinational paper money image identification method and apparatus | |
| Rill-García et al. | Pixel-accurate road crack detection in presence of inaccurate annotations | |
| CN104346628A (en) | License plate Chinese character recognition method based on multi-scale and multidirectional Gabor characteristic | |
| Boubaker et al. | New algorithm of straight or curved baseline detection for short arabic handwritten writing | |
| CN105005781A (en) | Robust extended local binary pattern texture feature extracting method | |
| CN106778766A (en) | A kind of rotary digital recognition methods and system based on anchor point | |
| CN113033567A (en) | Oracle bone rubbing image character extraction method fusing segmentation network and generation network | |
| CN115953399B (en) | Industrial part structural defect detection method based on contour features and SVDD | |
| CN111563925A (en) | Ellipse detection acceleration method based on generalized Pascal mapping | |
| CN111160142A (en) | Certificate bill positioning detection method based on numerical prediction regression model | |
| CN116311391A (en) | High-low precision mixed multidimensional feature fusion fingerprint retrieval method | |
| CN1581206A (en) | Finger print identification method | |
| CN115797649A (en) | Crack extraction method under complex background | |
| CN1845175A (en) | Grain surface damage detecting method based on wavelet and co-occurrence matrix |
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 | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20060517 |