CN102043951B - Joint finger segmentation method - Google Patents
Joint finger segmentation method Download PDFInfo
- Publication number
- CN102043951B CN102043951B CN2010106174117A CN201010617411A CN102043951B CN 102043951 B CN102043951 B CN 102043951B CN 2010106174117 A CN2010106174117 A CN 2010106174117A CN 201010617411 A CN201010617411 A CN 201010617411A CN 102043951 B CN102043951 B CN 102043951B
- Authority
- CN
- China
- Prior art keywords
- fingerprint
- image
- angle
- threshold value
- carries out
- 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
Abstract
The invention relates to a Mean shift algorithm and elliptic fitting-based joint finger segmentation system, and belongs to the domain of image processing. The system executes the following steps of: de-noising and filtering images to remove interference of an image acquisition process; performing mean shift adaptive operation on the graphs of each area, finding out a central magnetic resonance spectrum (MRS) of each area, performing interpolation operation on the graphs with similar fingerprint connection, performing elliptic fitting operation on the images, calculating an average elliptic deflection angle alpha, rotating the MRS by using the angle, and meanwhile finding out the optimal radius of each ellipse; removing the centers of non-fingerprint areas by using a trapezoidal method; and for over rotation of the MRS, rotating the MRS to the original angle, processing the fingerprint center by using the elliptic fitting algorithm again, performing elliptic regularization, and drawing a rectangle, wherein the area in the rectangle is the fingerprint area. Based on the image time domain, the system has low calculation quantity, can process the condition of finger deflection, and has low error and distortion.
Description
Technical field
The present invention relates to a kind of will the couplet refer to that figure carries out the image processing method of dividing processing, be the basis of follow-up many fingers fingerprint recognition promptly and prepare based on the fingerprint segmenting system of mean shift algorithm and ellipse fitting.
Background technology
Constantly perfect along with Flame Image Process and pattern-recognition each side technology, fingerprint recognition system is widely used in people's production and life, and in every field, plays an increasingly important role gradually.Single fingerprint identification technology is relatively ripe.But, according to the coupling index principle in the fingerprint recognition comparison process, certain mistake record rate and mistake in single fingerprint recognition process, can occur and refuse rate, life and production are caused difficulty or loss.And the multimode recognition methods that refers to fingerprint can improve the fingerprint recognition precision not increasing significantly under the collecting device condition of cost.The fingerprint that collects at the scene in addition also often occurs with the form of many fingers fingerprint.
Present domestic processing joins and refers to that drawing method has the couplet based on frequency-domain analysis to refer to figure front and back background separation algorithm, " Shanghai Communications University's journal ", 2010 08 phases.This algorithm is to be based upon on the basis of frequency domain, and noise reduction is better, but calculated amount is bigger.
The Chinese invention patent of University of Science and Technology of China Electronics; Application number 200810045690; " based on gradient projection and morphologic fingerprint image dividing method ", this patent is based on gradient projection and morphologic fingerprint image dividing method, is based upon on the gradient map of fingerprint; Calculated amount is smaller, and also more level and smooth through the fingerprint prospect profile after the morphological operation.Treatment effect for single fingerprint graph is better, but tilts or join to refer to that the situation of figure has certain shortcoming for finger.
Summary of the invention
The invention provides a kind of fingerprint dividing method, can in the short period of time fingerprint be referred to extract the figure from couplet based on mean shift algorithm and ellipse fitting.
Technical scheme of the present invention comprises the steps:
Step 1. pair original fingerprint image carries out noise reduction filtering to be handled, and removes the noise that factor such as environment in the image acquisition process causes.
The threshold value of step 1-1 computed image (Threshold).A Low (lower limit) and a Up (upper limit) have been added.As Low＞threshold value＞Up, carry out filtering, when threshold value is not, do not carry out filtering operation in this scope.
Step 1-2 carries out nonlinear filtering and strengthens: carry out threshold value earlier and take advantage of 1.5 times operation; When the gray-scale value of the current process points of image during greater than the threshold value that has been reinforced, with 1.5 times on duty of current gray level, otherwise with 0.6 times on duty of current gray level.
Step 1-3 adopts gaussian filtering.
Step 1-4 carries out former figure compensation to image, filtered gradation of image value is added 0.7 times of former figure gray-scale value.
The judgement again of step 1-5 threshold value: through Low＞threshold value＞Up, whether judgment threshold meets the requirements, if do not meet, then repeats above-mentioned steps once more.If meet, then enter into step 2.
Confirming of each regional center of step 2. and each regional deflection angle, and refer to situation with this company of having judged whether.
Step 2-1 carries out vertical projection to image, and removing pixel is zero zone, calculates initial radium R
0Use Mean shift (hereinafter to be referred as MS) algorithm to confirm each regional center O again
i(i=1,2,3 ...).
Step 2-2 is to the result points O of step 2-1
iCompare any 2 O in result points
ImWith O
InDistance during less than threshold value threshold_ distance, judge that then type fingerprint occurring connects a situation, jump out step 2, get into step 3.
When step 2-3 connected when type of appearance fingerprint not, the result points of the Mean Shift process that step 2-1 is obtained was carried out the ellipse fitting operation, calculated major axes orientation anglei (i=1,2,3 of the ellipse that each zone forms ...); Through adaptive process, find final radius R adius.
The rotation of step 3. image and each regional center and separation type of having fingerprint connect two fingerprints of situation.This step makes system can handle the situation of finger deflection.
Step 3-1 is for the image that type fingerprint connects situation occurring.
Step 3-1-1 carries out the ellipse fitting operation, obtains oval angle, removes the oval angle at fingerprint tie point place; Get rid of the big point of error; Calculate the mean value (angle_average1) of oval angle.
Step 3-1-2 is rotated above-mentioned MS result points if angle_average1 carries out the rotation of image greater than threshold value threshold_angle simultaneously, and angle is angle_average1;
Step 3-1-3 connects situation for type fingerprint occurring; From step 3-1-2, can find the gauge point that satisfies " connection criterion " among the MS result, get 2 average X coordinate, as the initial fuzzy rope point of receiving; Utilize vertical half projection algorithm then, calculate fingerprint tie point accurately; Inserting gray scale at the tie point place is that 0 radius is the black picture element band of Radius; Call " Mean Shift algorithm ", image is carried out definite operation at fingerprint radius R adius and all fingerprints and type fingerprint center.
Step 3-2 is for the image that type fingerprint connects situation not occurring.
Step 3-2-1 calculates the mean value (angle_average2) of oval angle.
Step 3-2-2 is rotated above-mentioned MS result points if angle_average2 carries out the rotation of image greater than threshold value threshold_angle simultaneously, and angle is angle_average2
Image after step 4. pair step 3 rotation carries out logical fingerprint and judges that this step can be removed the interference central point of non-finger-print region, makes image only stay the central point of four fingerprints at last.
Step 4-1 removes non-fingerprint noise spot with trapezoidal method, does trapezoidally at each peak, removes trapezoidal interior central point.。
Step 4-2 judges the right-hand man according to the height relationships between the finger.
Step 5. is confirmed final finger-print region for the image after the above-mentioned processing.This step can refer to that from couplet " circle " comes out the figure with four fingerprints.
Step 5-1 is to removing noise spot four fingerprint center F1 afterwards, and F2, F3, F4 are that initial radium carries out the ellipse fitting operation with Radius, find out the corresponding transverse W of each central point, the deflection angle α of minor axis L and fingerprint.
Step 5-2 is to the flexible within the specific limits search of each oval W and L, and the scope of W is 1.0-1.1, the scope of L be 0.8-1.1 work as certain pixel around when average gray value is greater than threshold value in the certain limit, stop to move, write down this point and be a1, a2, b1, b2; Use the congruent triangles principle and can calculate four summit c1 of rectangle, c2, c3, c4 makes rectangle, is finger-print region.
Described threshold value is whether judgment variable meets the requirements of metric.
It is that two fingerprints connect together or close proximity that described type of fingerprint connects.
Described ellipse fitting operation is that utilization matrix knowledge fits to ellipse with a certain image-region, can calculate long axis of ellipse thus, minor axis and major axis angle.
Described adaptive process is to be the loop iteration process, when satisfying condition, jumps out circulation.
Described vertical half projection algorithm be with a certain straight line as cut-off rule, with image segmentation two parts up and down, the pixel on the top of image is carried out the operation of vertical projection to transverse axis, obtain the corresponding pixel number of each horizontal ordinate greater than threshold value.
Described mean Shift algorithm is the step of an iteration, promptly calculates the skew average of current point earlier, moves this to its skew average, then as new starting point, continues to move, up to the end that meets some requirements.Finally obtain regional center.
The invention has the beneficial effects as follows to operate on the time domain basis that calculated amount is little, working time is few; Can connect situation by type of processing fingerprint; Can handle the situation that finger tilts.
Description of drawings
Fig. 1 is that the present invention joins the process flow diagram that refers to dividing method.
Fig. 2 is the effect synoptic diagram (comprising the path) after the mean shift process when not having class fingerprint connection situation.
Fig. 3 is that the noise spot synoptic diagram is not removed in image rotation back when having class fingerprint connection situation.
Fig. 4 is the final effect synoptic diagram of fingerprint extraction when not having class fingerprint connection situation.
Fig. 5 is type of having fingerprint result points effect synoptic diagram (not comprising the path) after the mean shift process for the first time when connecting situation.
Fig. 6 is type of having fingerprint result points effect synoptic diagram of mean shift process for the second time after the rotation interpolation when connecting situation.
Fig. 7 removes the result points synoptic diagram behind the noise spot when being type of having fingerprint connection situation.
The final effect synoptic diagram of fingerprint extraction when Fig. 8 is type of having fingerprint connection situation.
Embodiment
Be described in detail specific embodiment of the present invention below in conjunction with technical scheme and accompanying drawing.
Embodiment 1: do not have type fingerprint and connect situation
1. figure is carried out noise reduction filtering and handle, Low=30 during the threshold value Threshold of computed image wherein, Up=60.
2. image is carried out vertical projection, confirm each regional center with mean shift algorithm, like Fig. 2.
3. the ellipse fitting operation is carried out at each center, calculated the major axes orientation of the ellipse of each zone formation,, find final radius through adaptive process.
4. get rid of the big point of major axes orientation offset error, calculate the mean value of oval angle.
5. mean value and 15 degree that will go up the step relatively, if greater than then with image rotating, be rotated above-mentioned MS result points simultaneously.If less than, then do not rotate.
6. remove non-fingerprint noise spot with trapezoidal method.Wherein trapezoidal trapezoidal drift angle is 95 degree, like Fig. 3.
7. judge the right-hand man.
8. carry out the ellipse fitting operation to removing noise spot four fingerprint centers afterwards, find out the corresponding transverse of each central point, the deflection angle of minor axis and fingerprint.
9. to each long axis of ellipse and minor axis flexible search in 1.0-1.1 and 0.8-1.1 scope respectively, wherein threshold value gets 0.3.Find 4 summits of rectangle.
10. the rectangle that draws, result such as Fig. 4.
Embodiment 2: type of having fingerprint connects situation
1. figure is carried out noise reduction filtering and handle, Low=30 during the threshold value Threshold of computed image wherein, Up=60.
2. image is carried out vertical projection, confirm each regional center with mean shift algorithm, result points such as Fig. 5.
3. the ellipse fitting operation is carried out at each center, calculated the major axes orientation of the ellipse of each zone formation, get rid of type point and the big point of major axes orientation offset error of fingerprint junction, calculate the mean value of oval angle.
4. mean value and 15 degree that will go up the step relatively, if greater than then with image rotating, be rotated above-mentioned MS result points simultaneously.If less than, then do not rotate.
5. calculate fingerprint tie point accurately; Inserting gray scale at the tie point place is that 0 radius is the black picture element band of Radius.
6. reuse " mean shift algorithm ", image is carried out definite operation at fingerprint radius R adius and all fingerprints and type fingerprint center, result points such as Fig. 6.
7. remove non-fingerprint noise spot with trapezoidal method.Wherein trapezoidal trapezoidal drift angle is 95 degree, result points such as Fig. 7.
8. judge the right-hand man.
9. carry out the ellipse fitting operation to removing noise spot four fingerprint centers afterwards, find out the corresponding transverse of each central point, the deflection angle of minor axis and fingerprint.
10. to each long axis of ellipse and minor axis flexible search in 1.0-1.1 and 0.8-1.1 scope respectively, wherein threshold value gets 0.3; Find 4 summits of rectangle.
The rectangle 11. draw, result such as Fig. 8.
Claims (3)
1. a couplet refers to dividing method, is a kind of fingerprint dividing method based on mean shift algorithm and ellipse fitting, it is characterized in that following steps:
Step 1. pair image carries out noise reduction filtering to be handled; Concrete steps are following:
The threshold value of step 1-1. computed image (Threshold), the upper limit of having added a lower limit and the Up of a Low is worked as Low>threshold value Up, carry out filtering, when threshold value not in this scope, do not carry out filtering operation;
Step 1-2. carries out nonlinear filtering and strengthens: carry out threshold value earlier and take advantage of 1.5 times operation; When the gray-scale value of the current process points of image during greater than the threshold value that has been reinforced, with 1.5 times on duty of current gray level, otherwise with 0.6 times on duty of current gray level;
Step 1-3. adopts gaussian filtering;
Step 1-4. carries out former figure compensation to image;
The judgement again of step 1-5. threshold value: whether judgment threshold meets the requirements, through Low>threshold value>Up, if do not meet, then repeat above-mentioned steps 1-2,1-3 and 1-4 once more, carry out the filtering once more of image and strengthen; If meet, then enter into following step 2;
It is definite with each regional deflection angle that step 2. is carried out each regional center, and judged whether to join the situation that refers to this; Concrete steps are following:
Step 2-1. carries out vertical projection to image, finds out initial radium R
0Confirm each regional center O with mean shift algorithm again
i, i=1 wherein, 2,3
Step 2-2. is to the result points O of step 2-1
iCompare any 2 O in result points
ImWith O
InDistance during less than threshold value threshold_ distance, judge that then type fingerprint occurring connects a situation, jump out step 2, get into step 3;
When step 2-3. connected when type of appearance fingerprint not, the result points of the mean shift process that step 2-1 is obtained was carried out the ellipse fitting operation, calculated the major axes orientation anglei of the ellipse that each zone forms, i=1 wherein, 2,3 Through adaptive process, find final radius R adius;
The rotation of step 3. image and each regional center and separation type of having fingerprint connect two fingerprints of situation; Concrete steps are following:
Step 3-1. is for the image that type fingerprint connects situation occurring;
Step 3-1-1. carries out the ellipse fitting operation, obtains oval angle, removes the oval angle at fingerprint tie point place; Get rid of the big point of error; Calculate the mean value angle_average1 of oval angle;
Step 3-1-2. is rotated above-mentioned mean shift result points if angle_average1 greater than threshold value threshold_angle, carries out the rotation of image simultaneously, and angle is angle_average1;
Step 3-1-3. connects situation for type fingerprint occurring, from step 3-1-2, finds the gauge point that satisfies " connection criterion " among the mean shift result; Get 2 average X coordinate,, utilize vertical half projection algorithm then, calculate fingerprint tie point accurately as initially searching for generally a little; Inserting gray scale at the tie point place is that 0 radius is the black picture element band of Radius; Call mean shift algorithm, image is carried out definite operation at fingerprint radius R adius and all fingerprints and type fingerprint center;
Step 3-2. is for the image that type fingerprint connects situation not occurring;
Step 3-2-1. calculates the mean value angle_average2 of oval angle;
Step 3-2-2. is rotated above-mentioned mean shift result points if angle_average2 greater than threshold value threshold_angle, carries out the rotation of image simultaneously, and angle is angle_average2;
Image after step 4. pair step 3 rotation carries out logical fingerprint and judges that concrete steps are following:
Step 4-1. removes non-fingerprint noise spot with trapezoidal method; Do trapezoidally at each peak, removes trapezoidal interior central point;
Step 4-2. judges the right-hand man;
Image after step 5. is handled for step 4 is confirmed final finger-print region;
Step 5-1. is to removing noise spot four fingerprint center F1 afterwards, and F2, F3, F4 carry out the ellipse fitting operation, find out the corresponding transverse W of each central point, the deflection angle α of minor axis L and fingerprint;
Step 5-2., stops to move to each oval W and the flexible search of L when average gray value is greater than threshold value around certain pixel, writes down this point and is a1, a2, b1, b2; Described threshold value is whether judgment variable meets the requirements of metric; Calculate four summit c1 of rectangle, c2, c3, c4 makes rectangle, is finger-print region.
2. the dividing method that refers to according to claim 1 is characterized in that, step 1-4. carries out former figure compensation to image and comprises: filtered gradation of image value adds 0.7 times of former figure gray-scale value.
3. the dividing method that refers to according to claim 1 is characterized in that, step 5-2, and the flexible scope of W is 1.0-1.1, the flexible scope of L is 0.8-1.1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010106174117A CN102043951B (en) | 2010-12-31 | 2010-12-31 | Joint finger segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010106174117A CN102043951B (en) | 2010-12-31 | 2010-12-31 | Joint finger segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102043951A CN102043951A (en) | 2011-05-04 |
| CN102043951B true CN102043951B (en) | 2012-07-25 |
Family
ID=43910079
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010106174117A Expired - Fee Related CN102043951B (en) | 2010-12-31 | 2010-12-31 | Joint finger segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102043951B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102999750B (en) * | 2012-12-31 | 2015-08-12 | 清华大学 | A kind of fingerprint on site Enhancement Method removing background interference |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383006A (en) * | 2007-09-05 | 2009-03-11 | 清华大学深圳研究生院 | Image pre-processing method of oval correction and circle normalization |
| EP2131307A1 (en) * | 2008-05-27 | 2009-12-09 | Siemens AG Österreich | Method for segmenting fingerprint images |
-
2010
- 2010-12-31 CN CN2010106174117A patent/CN102043951B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383006A (en) * | 2007-09-05 | 2009-03-11 | 清华大学深圳研究生院 | Image pre-processing method of oval correction and circle normalization |
| EP2131307A1 (en) * | 2008-05-27 | 2009-12-09 | Siemens AG Österreich | Method for segmenting fingerprint images |
Non-Patent Citations (2)
| Title |
|---|
| Robert H,Surinder R,et al.slap fingerprint segmentation.《Computer Vision Winter Workshop 2009》.2009,第1-3部分. * |
| 高建坡等.基于均值移动和椭圆拟合的人脸跟踪算法.《东南大学学报(自然科学版)》.2007,第37卷(第1期),897-902. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102043951A (en) | 2011-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107679520B (en) | Lane line visual detection method suitable for complex conditions | |
| CN106846359B (en) | Moving target rapid detection method based on video sequence | |
| CN102750708B (en) | Affine motion target tracing algorithm based on fast robust feature matching | |
| CN102999886B (en) | Image Edge Detector and scale grating grid precision detection system | |
| CN103325126B (en) | A kind of video target tracking method under dimensional variation and circumstance of occlusion | |
| US9710725B2 (en) | Method and system for extracting image salient curve | |
| CN101937565B (en) | Dynamic image registration method based on moving target track | |
| CN101916446A (en) | Gray level target tracking algorithm based on marginal information and mean shift | |
| CN102609917B (en) | Image edge fitting B spline generating method based on clustering algorithm | |
| CN102521836A (en) | Edge detection method based on gray-scale image of specific class | |
| CN109858455B (en) | Block detection scale self-adaptive tracking method for round target | |
| CN104036523A (en) | Improved mean shift target tracking method based on surf features | |
| CN104537342B (en) | A kind of express lane line detecting method of combination ridge border detection and Hough transformation | |
| CN108986038A (en) | A kind of wheel-hub contour detection method based on Improved Hough Transform | |
| CN112464847B (en) | Human body action segmentation method and device in video | |
| CN109685827B (en) | Target detection and tracking method based on DSP | |
| CN103077531A (en) | Grayscale target automatic tracking method based on marginal information | |
| CN104809736A (en) | Medical tomographic image closed skeleton outline calculation method based on priori knowledge | |
| CN105719275A (en) | Parallel combination image defect segmentation method | |
| CN110298344A (en) | A kind of positioning of instrument knob and detection method based on machine vision | |
| CN111161276B (en) | Iris normalized image forming method | |
| CN102043951B (en) | Joint finger segmentation method | |
| CN202815864U (en) | Gesture identification system | |
| CN105809703A (en) | Adhesion hemocyte image segmentation method based on improved fractional differential and graph theory | |
| Wang et al. | Lane detection algorithm based on density clustering and RANSAC |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120725 Termination date: 20141231 |
|
| EXPY | Termination of patent right or utility model |