CN107346545A - Improved confinement growing method for the segmentation of optic cup image - Google Patents
Improved confinement growing method for the segmentation of optic cup image Download PDFInfo
- Publication number
- CN107346545A CN107346545A CN201710363040.6A CN201710363040A CN107346545A CN 107346545 A CN107346545 A CN 107346545A CN 201710363040 A CN201710363040 A CN 201710363040A CN 107346545 A CN107346545 A CN 107346545A
- Authority
- CN
- China
- Prior art keywords
- region
- image
- pixel
- seed point
- gray
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Abstract
Improved confinement growing method for the segmentation of optic cup image, it is characterised in that：In this method, first, signature analysis is carried out to eyeground major physiological structure, green channel is have chosen for segmentation object and area-of-interest is extracted roughly according to threshold method（ROI）；Secondly, it is contemplated that the shortcomings of traditional algorithm of region growing is inaccurate, adaptivity is poor in selected seed point, the standard that this method is used as selected seed point by the geometric center and combination center brightness that calculate ROI are improved；Finally, mean filter is carried out to eye fundus image with 5*5 templates, seed merging is carried out to eye fundus image using mountain valley difference criterion and 8 neighborhood connectivity criterias, is finally accurately partitioned into optic cup.The application algorithmic stability compared with traditional algorithm is reliable, there is higher segmentation sensitivity, specificity and accuracy.
Description
Technical field：
The present invention relates to a kind of improved confinement growing method for the segmentation of optic cup image.
Background technology：
Cup area, cup disc area in eye fundus image than etc. parameter turn into the direction of many people research, Fig. 1 is a width eye
Base map picture, there it can be seen that optic disk is approximate circle, there is physiological depression, referred to as optic cup at one in its center, optic cup is to be located at
The most bright simply connected region in optic disk center, its region between optic disk are referred to as disk edge.
For eye fundus image, there has been proposed the dividing method of a variety of optic cups in recent years.Domestic scholars Zheng Shan et al. utilizes eye
The multiphase active contour model under elliptical shape this feature extraction ellipse restriction is presented in optic disk and optic cup in base map picture, realizes
The Accurate Segmentation of optic cup.Zhao Qiu bonuses determine initial profile line with mathematical morphology and Otsu threshold method, improve the correct of segmentation
Rate.Foreign scholar Irene Fond ó n et al. obtain optic cup edge pixel using random forest sorting algorithm.Ashish Issac
Et al. optic cup is split according to the characteristic of image, calculate the average value and standard deviation of image.Dwarikanath Mahapatra etc.
People[5]Specialist field model (FOE) segmentation optic cup is proposed, independent of manual drawing.Hanan Alghmdi et al. use super picture
Plain clustering algorithm, super-pixel feature is extracted, for training grader.Nataraj A.Vijapur et al. research emphasis is base
Correspond to eye fundus image in Pearson came (Pearson-R) coefficient, calculate the perpendicular diameter after optic cup and optic disk segmentation, and compared
Value is with this determination CDR.But algorithm of region growing is almost calculated in optic cup segmentation without reference to the growth of, foreign scholar application region
Method splits the image in different research backgrounds, for example Mahaman Sani Chaibou et al. emphasis is selecting seed region, makes
Information fast propagation is simultaneously performed, and is identified image and is ensured correctly to split.
Application region growth algorithm segmentation optic cup：
Algorithm of region growing has some in image on the picture of certain similar characteristic (such as gray feature, textural characteristics etc.)
Element condenses together to form specific region.Algorithm of region growing is a kind of serial region segmentation method, behind the step of need before
Face result support, if processing pixel order is different, segmentation result may also be different.
Traditional area growth algorithm:
Since traditional region-growing method be similar to those pre defined attributes the neighbour of seed choosing one " seed point "
Domain pixel is attached on each seed, and this process is repeated, untill meeting end condition set in advance.Traditional area
Growth method includes 3 steps:
(1) suitable seed point is selected；
(2) criterion that potting gum can be come in growth course is determined；
(3) condition that growth can be allowed to stop is made.
Algorithm flow is as shown in Figure 2.
The segmentation result of the algorithm depends greatly on selection order and the position of initial seed point, seed point
Selection is sometimes incorrect, causes regional choice mistake.
The content of the invention：
Goal of the invention：
The present invention provides a kind of improved confinement growing method for the segmentation of optic cup image, and the purpose is to overcome conventional lack
Fall into.
Technical scheme：
Improved confinement growing method for the segmentation of optic cup image, it is characterised in that：In this method, first, to eyeground master
Want physiological structure to carry out signature analysis, have chosen green channel for segmentation object and region of interest is extracted roughly according to threshold method
Domain (ROI)；Secondly, it is contemplated that the shortcomings of traditional algorithm of region growing is inaccurate, adaptivity is poor in selected seed point, this
The standard that method is used as selected seed point by the geometric center and combination center brightness that calculate ROI is improved；Finally, 5* is used
5 templates carry out mean filter to eye fundus image, and seed is carried out to eye fundus image using mountain valley difference criterion and 8 neighborhood connectivity criterias
Merge, be finally accurately partitioned into optic cup.
First, G passages and gray processing are extracted, and ROI region is extracted roughly according to its grey level histogram threshold application method；
Then the geometric center in ROI region is calculated and according to the pixel value Selection Center of gray-scale map；Picture around Correlation Centre successively
Element, if center pixel value is less than the pixel of surrounding point, seed point is substituted by the big value of pixel value, obtains pixel value maximum
As seed point, seed point region of search is defined as using optic disk central point as circle the i.e. most bright spot of point, is straight using 0.5 times of disc diameter
The circle in footpath；First with 5*5 templates to eye fundus image mean filter before optic cup is split, improve seed point positioning performance；Finally, root
Pixel, which is found, according to difference between gray-scale map mountain valley as threshold value and in 8 connected regions collects the most pixel of number as optic cup
Split alternative area, realize local auto-adaptive region growing, so as to obtain optic cup region, realize the automatic detection of optic cup and automatic
Extraction.
(1), pre-process：
(1) G passages and gray processing, are extracted：
Eye fundus image is the coloured image of RGB patterns, and we extract the green G passages conduct in three passages of RGB models
Process object, formula (1) are the method for extraction G passages：
G be extraction after green component, R be extraction before image red component, G be extraction before image green component, B
For the blue component of image before extraction；
(2) ROI, is extracted：
ROI is area-of-interest, in the method as optic disk region, and it is approximately circular, and this method applies threshold first
Value method extracts roughly optic disk region；
(3) ROI geometric centers, are calculated：
Obtain bianry image (Ib) after, operation and closed operation are opened by two continuous morphology, obtains one continuously
Region；Whether the region of a closure can be obtained in normal eye fundus image or in abnormal eye fundus image；
(2), selected seed point：
After obtaining ROI, formula is split according to seed point algorithm of region growing and calculates the position of seed point, and marks kind
Sub- point；
Center of fiqure is chosen shown in formula such as formula (5)：
Wherein,It is the area of ROI region, X is integrand, and μ is the seed point coordinates chosen after being computed；
(3) region growing：
(1) 5*5 templates mean filter：
The groundwork of this part is smooth eyeground gray level image, strengthens the positioning performance of seed point；We are using 5*5's
Mean filter goes smooth eye fundus image to eliminate spot and noise；Shown in concrete operations such as formula (6)：
Wherein, m is that the pixel total number including current pixel is included in the template, and point (x, y) is the seat where this pixel
Cursor position；G (x, y) is smooth later area grayscale value, and f (x, y) is the gray value at any point in template；
(2) optic cup is split：
We make same class potting gum into the region comprising a seed point using three standards；
First standard：Absolute grayscale difference between any pixel and seed point is necessarily less than T, and T values are by retina eye
Bottom image grey level histogram obtains, and what T values represented is eyeground grey level histogram between 255 and first mountain valley from the right number
Difference；
Second standard：Comprising optic cup region, necessarily at least 8 neighborhoods connect pixel in optic cup region or ROI region
Logical；
3rd standard：Using seed point p as the center of circle, 0.5 times of optic disk radius is radius as seed growth region, and is surpassed
Go out the limit of consideration not as seed growth region of this scope；
Three seed points more than grow criterion, after initial seed point determination, judge kind according to criterion
Son point peripheral region pixel whether meet grow criterion, if meeting criterion set in advance, by property it is identical and
Pixel in threshold range is classified as the number of pixels increase of the same area, i.e. seed point region, and this region area is continuous
Expand；Whether eligible the region others pixel is continued search for, until search finishes；The region constantly grows, finally
It is partitioned into optic cup region；Its concrete operations is：
The gray scale in eye fundus image grey level histogram by calculating local neighborhood is worth to threshold value, its growth standard function
For：
Wherein, μ is the average of Gray Histogram value, and σ is the variance of Gray Histogram value, and c is coefficient factor, and x is pixel
The gray value of point.
After segmentation optic cup：
(1) seed point p selected before is marked；
(2) seed point p and field pixel p around it are calculatediThe distance between d；
(3) according to distance d, whether this pixel where judgment threshold TP can be integrated into growth district：
TP=| MM×M-MN×N|, 0 ＜ d≤D (7)
If TP is less than or equal to corresponding threshold value, the pixel is integrated into seed point region；This paper threshold values are bases
In grey level histogram between mountain valley difference (255-240=15) and set；
Wherein MM×M,MN×NPixel P is represented respectivelyiThe gray average and seed point p of regioniGray value, D represent away from
From critical value, this paper critical distances are 0.5 times of optic disk radiuses.
In the step of extracting G passages and gray processing, it is processed for improving processing speed, Wo Men for gray level image
Gray processing processing is carried out after extracting G channel components.
The grey level histogram of eye fundus image is utilized in the step of extracting ROI, background is dark optic disk optic cup in the picture
Region is light；Light pixel in object produces right peak on the histogram, and substantial amounts of gray level produces in background
The left peak of histogram；Number of pixels of the object boundary nearby with gray level between two peak values is relatively fewer, so as to generate
Paddy between two peaks；Selection paddy will obtain rational border as gray threshold；Optic disk may determine that according to grey level histogram
Threshold value.
Threshold value selection is 180 and then a closed operation is carried out on the basis of being 180 in threshold value in this method, and elimination is tiny not
The region of closure, the region for eventually forming a closing is ROI region.
Opening Operation Definition in the step of calculating ROI geometric centers is：
B is the element of an optic disk planform, 30 pixels of radius；
IBIt is original-gray image；
Closed operation is defined as：
B is the element of an optic disk planform, radius 20；
ICIt is the image after opening operation processing；
Morphologic corrosion and expansion respectively with Θ andRepresent, tiny and independent cell can be removed by opening operation
Domain, corresponding to bright lesion region, such as the druse in eye fundus image；Opening operation, to eliminate its medium and small and isolated
Structural element；The optimum size of structural element is drawn by our all analysis of experimental data, including different illuminations
Tiny changes with some；Closed operation has filled up breach tiny in ROI region and has been connected to area tiny in optic disk region
Domain.
The step of selected seed point, this part of selected seed point include following six step：
(1) eye fundus image ROI gray-scale map is converted into by bianry image by the method for threshold value；
(2) remove tiny region with morphology closed operation and a continuous ROI is only provided；
(3) ROI geometric center is calculated；
(4) if geometric center pixel is 255, it is seed point to automatically select geometric center；
(5) if geometric center pixel is not 255, using geometric center as round dot, search model is used as using 0.5 times of optic disk radius
Enclose, 5 length in pixels are scale, every time 45 ° of rotation, and obtain the maximum point of gray value on scale, i.e., shown 49 in Figure 5
A maximum point of radian alternately seed point is chosen in point；
(6) by this rule, alternative seed point search is carried out in the whole field of search；Finally selected in all alternative seed points
It is seed point to take the maximum point of a gray value；
After meeting conditions above this can be marked to put the seed point as Region growing segmentation algorithm automatically.
Advantageous effect：
The application should in this way, to 15 glaucoma eye fundus images in high-resolution eye fundus image (HRF) database
Detected one by one with 15 healthy eyes eye fundus images, rate of accuracy reached to 93.3%, test result indicates that, the algorithm can it is quick,
Optic cup that effectively automatic detection goes out in eye fundus image simultaneously correctly splits it, and the algorithm is steady compared with traditional algorithm
It is fixed reliable, there are higher segmentation sensitivity, specificity and accuracy.
Brief description of the drawings：
Fig. 1 is background technology eyeground figure；
Fig. 2 is this method flow chart；
Fig. 3 is each passage extractions of colored eye fundus image RGB and gray processing figure；
Fig. 4 is the grey level histogram of eye fundus image；
Fig. 5 is that rotary scale determines seed point diagram；
Fig. 6 is neighborhood histogram analysis figure；
Fig. 7 is the segmentation effect figure of this method；
Fig. 8 is scholar's Freehandhand-drawing optic cup region；
Fig. 9 is accuracy comparison chart；
Figure 10 is specific comparison chart；
Figure 11 is sensitiveness comparison chart.
Embodiment：
The present invention provides a kind of improved confinement growing method for the segmentation of optic cup image, and this method is the segmentation after improving
Method:
First, G passages and gray processing are extracted, and ROI region is extracted roughly according to its grey level histogram threshold application method；
Then the geometric center in ROI region is calculated and according to the pixel value Selection Center of gray-scale map.Picture around Correlation Centre successively
Element, if center pixel value is less than the pixel of surrounding point, seed point is substituted by the big value of pixel value, obtains pixel value maximum
As seed point, seed point region of search is defined as using optic disk central point as circle the i.e. most bright spot of point, is straight using 0.5 times of disc diameter
The circle in footpath.First with 5*5 templates to eye fundus image mean filter before optic cup is split, improve seed point positioning performance；Finally, root
Pixel, which is found, according to difference between gray-scale map mountain valley as threshold value and in 8 connected regions collects the most pixel of number as optic cup
Split alternative area, realize local auto-adaptive region growing, so as to obtain optic cup region, realize the automatic detection of optic cup and automatic
Extraction.
Pretreatment
(1) G passages and gray processing are extracted
Eye fundus image is the coloured image of RGB patterns.By the comparison to red, green, blue triple channel image, it is found that three lead to
The brightness of road image reduces successively, red channel brightness of image highest, and optic cup and optic disk color contrast are low；The spy of green channel
Point is that image is more clear, and the contrast of optic cup and optic disk is big, and the optic cup outline effect that it is included is best；Blue channel image
Dark, contrast and definition are relatively low, and also many noises.Therefore we are extracted in three passages of RGB models
Green G passages are as process object.Formula (1) is the method for extraction G passages.
G be extraction after green component, R be extraction before image red component, G be extraction before image green component, B
For the blue component of image before extraction.
In order to improve processing speed, we are handled for gray level image, it is therefore desirable to laggard in extraction G channel components
The processing of row gray processing.Fig. 3 is the design sketch after extraction R, G, B triple channel and gray processing.
(2) ROI is extracted
ROI is area-of-interest, is optic disk region herein, and it is approximately circular.Threshold application method first herein
Rough extraction optic disk region.The grey level histogram of a certain eye fundus image is as shown in figure 4, background is dark, optic disk in the picture
Optic cup region is light.Light pixel in object produces right peak on the histogram, and substantial amounts of gray level is produced in background
Life is at the left peak of histogram.Number of pixels of the object boundary nearby with gray level between two peak values is relatively fewer, so as to produce
The paddy between two peaks is given birth to.Selection paddy will obtain rational border as gray threshold.It may determine that according to grey level histogram
The threshold value of optic disk, threshold value selection herein is 180 and then carries out a closed operation on the basis of being 180 in threshold value, is eliminated tiny
Inc region.The region for eventually forming a closing is ROI region.
(3) ROI geometric centers are calculated
Obtain bianry image (Ib) after, operation and closed operation are opened by two continuous morphology, a company can be obtained
Continuous region.Whether the area of a closure can be obtained in the eye fundus image of health or in ill eye fundus image
Domain.
Opening Operation Definition is：
B is the element of an optic disk planform, radius 30.
IBIt is original-gray image.
Closed operation is defined as：
B is the element of an optic disk planform, radius 20.
ICIt is the image after opening operation processing.
Morphologic corrosion and expansion respectively with Θ andRepresent, tiny and independent cell can be removed by opening operation
Domain, corresponding to bright lesion region, such as the druse in eye fundus image.Opening operation, to eliminate its medium and small and isolated
Structural element.The optimum size of structural element is drawn by our all analysis of experimental data, including different illuminations
Tiny changes with some.Closed operation has filled up breach tiny in ROI region and has been connected to area tiny in optic disk region
Domain.
1.2.2 selected seed point
After obtaining ROI, the position of seed point is calculated herein according to seed point algorithm of region growing segmentation formula, and mark
Go out seed point.
Center of fiqure is chosen shown in formula such as formula (5)：
Wherein,It is the area of ROI region.
This part of selected seed point includes following six step：
(1) eye fundus image ROI gray-scale map is converted into by bianry image by the method for threshold value.
(2) remove tiny region with morphology closed operation and a continuous ROI is only provided.
(3) ROI geometric center is calculated.
(4) if geometric center pixel is 255, it is seed point to automatically select geometric center.
(5) if geometric center pixel is not 255, using geometric center as round dot, search model is used as using 0.5 times of optic disk radius
Enclose, 5 length in pixels are scale, every time 45 ° of rotation, and obtain the maximum point of gray value on scale, i.e., shown 49 in Figure 5
A maximum point of radian alternately seed point is chosen in point.
(6) by this rule, alternative seed point search is carried out in the whole field of search.Finally selected in all alternative seed points
It is seed point to take the maximum point of a gray value.
After meeting conditions above this can be marked to put the seed point as Region growing segmentation algorithm automatically.
Region growing：
(1) 5*5 templates mean filter
The groundwork of this part is smooth eyeground gray level image, in order to strengthens the positioning performance of seed point.This
Text has used 3*3,5*5 and 7*7 mean filter to be smoothed eye fundus image respectively, reduces noise.Use 3*3
The image of mean filter is because template is too small without too big effect, and 7*7 template can make edge transition smooth, cause side
Boundary obscures, so finally we use 5*5 mean filter to go smooth eye fundus image to eliminate spot and noise.Concrete operations
As shown in formula (6)：
Wherein, m is that the pixel total number including current pixel is included in the template.
(2) optic cup is split
We make same class potting gum into the region comprising a seed point using three standards.
First standard：Absolute grayscale difference between any pixel and seed point is necessarily less than T, and T values are by retina eye
Bottom image grey level histogram obtains, and what T values represented is eyeground grey level histogram between 255 and first mountain valley from the right number
Difference.From first mountain valley value of the right number it is 240 in figure, then T=255-240 as shown in intensity histogram Fig. 4.
Second standard：Comprising optic cup region, necessarily at least 8 neighborhoods connect pixel in optic cup region or ROI region
Logical.
3rd standard：Using seed point p as the center of circle, 0.5 times of optic disk radius is radius as seed growth region, and is surpassed
Go out the limit of consideration not as seed growth region of this scope.
Three seed points more than grow criterion, after initial seed point determination, judge kind according to criterion
Son point peripheral region pixel whether meet grow criterion, if meeting criterion set in advance, by property it is identical and
Pixel in threshold range is classified as the number of pixels increase of the same area, i.e. seed point region, and this region area is continuous
Expand.Whether eligible the region others pixel is continued search for, until search finishes.The region constantly grows, finally
It is partitioned into optic cup region.
Gray scale in eye fundus image grey level histogram of the algorithm by calculating local neighborhood is worth to threshold value, and it grows mark
Quasi-function is：
Wherein, μ is the average of Gray Histogram value, and σ is the variance of Gray Histogram value.C is coefficient factor.
For eyeground optic cup this institutional framework, data message is reflected on the grey level histogram of eyeground, such as Fig. 6 institutes
Show, left figure is eye fundus image grey level histogram in Fig. 6, and for ease of finding threshold value, right figure is by the gray-scale map after histogram-fitting.
We effectively can extract threshold value data in histogram, carry out adaptive region growth.
Improved algorithm of region growing step is herein：
(4) seed point p selected before is marked；
(5) seed point p and field pixel p around it are calculatediThe distance between d；
(6) according to distance d, whether this pixel where judgment threshold TP can be integrated into growth district：
TP=| MM×M-MN×N|, 0 ＜ d≤D (7)
If TP is less than or equal to corresponding threshold value, the pixel is integrated into seed point region.This paper threshold values are bases
In grey level histogram between mountain valley difference (255-240=15) and set.
Wherein MM×M,MN×NPixel P is represented respectivelyiThe gray average and seed point p of regioniGray value, D represent away from
From critical value, this paper critical distances are 0.5 times of optic disk radiuses.
Experimental result contrasts with algorithm：
It is herein 3504 to 30 width resolution ratio in eyeground high-resolution HRF (High Resolution Fundus) database
The JPG colours eye fundus image (wherein 15 width are glaucoma image, and 15 width are healthy eye fundus image) of × 2336 pixels carries out algorithm
Checking.These images have different colors, brightness and quality.For there is the eye fundus image of difference, by expert to wherein should
The relevant position of difference carries out manual mark, and gained annotation results are used for the performance for evaluating automatic detection algorithm.
Split-run test result：
Fig. 7 is the last segmentation result of this paper algorithms.The optic disk area that rough split plot design is partitioned into is shown in wherein figure a
Domain, figure b are the complete i.e. ROI in optic disk region of closure formed after morphology closed operation, and figure c is improved according to this method
The seed point that criterion afterwards adaptively selects, figure d are the images after 5*5 mean filter template convolutions, and figure e is application enhancements
The optic cup region that algorithm of region growing afterwards is partitioned into, figure f are the tiny breach in optic cup region after segmentation is filled up in closed operation
Image.As can be seen that having finally given the optic cup region of complete closure using this method during from this.
This paper and other algorithm of region growing ratios：
In traditional optic cup partitioning algorithm, conventional is Threshold Segmentation Algorithm and edge detection method.In threshold value point
In cutting, it is crucial the problem of be threshold value selection, threshold value selection can by image grey level histogram obtain, according to each gray level
Distribution situation select one or more threshold values, to determine the tonal range in each region, but such a method is cumbersome, needs sometimes
Will manually selected threshold, it is impossible to reach good adaptive performance.Edge detection algorithm extraction optic cup is needed in erasing blood
Applied mathematics morphology progress canny rim detections, complex disposal process, real-time are poor on the basis of pipe.Table 1 is to difference
Algorithm performance comparison data result, it can be seen that the algorithm of region growing after improving is in terms of accuracy rate, positive predictive value
It is improved and run time has also shortened.
The algorithms of different performance evaluation of table 1.
Algorithm is evaluated：
Because the edge that different ophthalmology scholars draws can be variant, therefore for the segmentation of optic cup optic disk also without a kind of visitor
The evaluation criteria of sight.
Fig. 8 is the optic cup region of certain expert's Freehandhand-drawing, herein in this, as correct standard, for evaluation algorithms performance.In table 2
Data be 10 eye fundus images chosen from the high-definition picture HRF databases of eyeground as sample, using this paper algorithms
The result split.
The coefficient correlation and cup disc ratio of the optic disk of table 2. and optic cup
Table 2.The correlation coefficient of the optic disc,optic cup and
CDR
OD, OC in table are respectively the pixel region area value of optic disk and optic cup.
As can be seen that relative error control exists in algorithm partition data comparing result from the expert in table 2 and after improvement
Within 4%, accuracy rate is of a relatively high.
It is to be based respectively on image level and phase to the relevant evaluation index of optic cup automatic detection algorithm performance in the application
Close what peculiar zone level defined.
Wherein TP is kidney-Yang, and FP is false sun, and TN is Kidney-Yin, and FN is false cloudy.Specific explanations are:It is detected as including for algorithm
The eye fundus image of big optic cup, it is whether consistent with expert's artificial judgment result according to it, it is referred to as kidney-Yang and vacation sun；For calculating
Whether method is detected as normal eye fundus image, identical with expert's artificial judgment result identical according to it, is referred to as Kidney-Yin and false
It is cloudy.By algorithm to the testing result in peculiar region in retina compared with the artificial judgment result of ophthalmology scholar when, due to algorithm
What is finally provided is all that it is judged as abnormal region, therefore only exists kidney-Yang, false sun, and the vacation of missing inspection is cloudy, does not deposit now
Kidney-Yin.Select accuracy (accuracy), specific (specificity), the sensitiveness of image level respectively herein
(sensitivity), and peculiar zone level sensitivity evaluation algorithm performance.The definition of these indexs such as formula (8)~
(10) defined.
Herein from the accuracy of image, specificity, the broad aspect of sensitiveness three to the algorithm of region growing after improvement, threshold value
Method, edge detection method and traditional algorithm of region growing have carried out Performance Evaluation contrast respectively.Comparing result such as Fig. 9, Figure 10, figure
Shown in 11.
Fig. 9, Figure 10, Figure 11 reflect the results of property of the performance indications of three big characteristics, i.e. optic cup segmentation.Can by contrast
To find out：The accurate performance of four kinds of algorithms is respectively from high to low：Algorithm of region growing, threshold method, traditional area after improvement
Growth algorithm and edge detection algorithm.Specific performance is respectively from high to low：Algorithm of region growing, threshold method after improvement, side
Edge detection algorithm and traditional area growth algorithm.And for sensitive property, sensitiveness is higher, and antijamming capability is lower, leads to
Cross in table it can be seen that sensitiveness from high to low be respectively：Traditional area growth algorithm, edge detection algorithm, threshold method and change
Algorithm of region growing after entering.Compared by four kinds of algorithm performances in three tables, it can be seen that the region after this paper is improved
Growth algorithm combination property is higher.
In summary, it can be seen that the region growing calculation after improving by splitting the Contrast on effect of optic cup to algorithms of different
Method can be accurately positioned seed point location, the segmentation criterion after improvement, and region is merged and can extract optic cup after regular
Region.Obtained optic cup region is accurate, it is complete, without hole, in cutting procedure, the adaptive selected seed point of algorithm, universality compared with
It is good.The algorithm is applied in 30 width images of HRF databases, optic cup splits accuracy rate more than 0.93.
Claims (9)
1. the improved confinement growing method for the segmentation of optic cup image, it is characterised in that：It is first, main to eyeground in this method
Physiological structure carries out signature analysis, have chosen green channel for segmentation object and extracts area-of-interest roughly according to threshold method
(ROI)；Secondly, it is contemplated that the shortcomings of traditional algorithm of region growing is inaccurate, adaptivity is poor in selected seed point, we
The standard that method is used as selected seed point by the geometric center and combination center brightness that calculate ROI is improved；Finally, 5*5 is used
Template carries out mean filter to eye fundus image, and seed is carried out to eye fundus image using mountain valley difference criterion and 8 neighborhood connectivity criterias
Merge, be finally accurately partitioned into optic cup.
2. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：First,
G passages and gray processing are extracted, and ROI region is extracted roughly according to its grey level histogram threshold application method；Then ROI areas are calculated
Geometric center in domain and according to the pixel value Selection Center of gray-scale map；Pixel around Correlation Centre successively, if center pixel
Value is less than the pixel of surrounding point, then seed point is substituted by the big value of pixel value, and it is that most bright spot is made to obtain the maximum point of pixel value
For seed point, seed point region of search is defined as using optic disk central point as circle, the circle using 0.5 times of disc diameter as diameter；Dividing
5*5 templates are first used to improve seed point positioning performance eye fundus image mean filter before cutting optic cup；Finally, according to gray-scale map mountain
As threshold value and in 8 connected regions, searching pixel collects the most pixel of number as optic cup segmentation candidate area to difference between paddy
Domain, local auto-adaptive region growing is realized, so as to obtain optic cup region, realize the automatic detection of optic cup and automatically extract.
3. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：This method
The step of it is as follows：(1), pre-process：
(1) G passages and gray processing, are extracted：
Eye fundus image is the coloured image of RGB patterns, and the green G passages that we are extracted in three passages of RGB models are used as processing
Object, formula (1) are the method for extraction G passages：
<mrow>
<mi>g</mi>
<mo>=</mo>
<mfrac>
<mi>G</mi>
<mrow>
<mi>R</mi>
<mo>+</mo>
<mi>G</mi>
<mo>+</mo>
<mi>B</mi>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
G is the green component after extracting, and R is the red component of image before extraction, and G is the green component of image before extraction, and B is to carry
The blue component of image before taking；
(2) ROI, is extracted：
ROI is area-of-interest, in the method as optic disk region, and it is approximately circular, this method threshold application method first
Rough extraction optic disk region；
(3) ROI geometric centers, are calculated：
Obtain bianry image (Ib) after, operation and closed operation are opened by two continuous morphology, obtains a continuous region；
Whether the region of a closure can be obtained in normal eye fundus image or in abnormal eye fundus image；
(2), selected seed point：
After obtaining ROI, formula is split according to seed point algorithm of region growing and calculates the position of seed point, and marks seed
Point；
Center of fiqure is chosen shown in formula such as formula (5)：
Wherein,It is the area of ROI region, X is integrand, and μ is the seed point coordinates chosen after being computed；
(3) region growing：
(1) 5*5 templates mean filter：
The groundwork of this part is smooth eyeground gray level image, strengthens the positioning performance of seed point；We use 5*5 average
Wave filter goes smooth eye fundus image to eliminate spot and noise；Shown in concrete operations such as formula (6)：
<mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>m</mi>
</mfrac>
<mi>&Sigma;</mi>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, m is that the pixel total number including current pixel is included in the template, and point (x, y) is the coordinate bit where this pixel
Put；G (x, y) is smooth later area grayscale value, and f (x, y) is the gray value at any point in template；
(2) optic cup is split：
We make same class potting gum into the region comprising a seed point using three standards；
First standard：Absolute grayscale difference between any pixel and seed point is necessarily less than T, and T values are by retina eyeground figure
As grey level histogram obtains, what T values represented is difference of the eyeground grey level histogram between 255 and first mountain valley from the right number
Value；
Second standard：Comprising optic cup region, pixel is necessarily at least the connection of 8 neighborhoods in optic cup region or ROI region；
3rd standard：Using seed point p as the center of circle, 0.5 times of optic disk radius is radius as seed growth region, and exceeds this
The limit of consideration not as seed growth region of scope；
Three seed points more than grow criterion, after initial seed point determination, judge seed point according to criterion
Whether the pixel of peripheral region meets to grow criterion, if meeting criterion set in advance, property is identical and in threshold value
In the range of pixel be classified as the same area, i.e., the number of pixels increase of seed point region, this region area constantly expand；
Whether eligible the region others pixel is continued search for, until search finishes；The region constantly grows, and is finally partitioned into
Optic cup region；Its concrete operations is：
The gray scale in eye fundus image grey level histogram by calculating local neighborhood is worth to threshold value, and its growth standard function is：
<mrow>
<mi>h</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>;</mo>
<mi>&mu;</mi>
<mo>,</mo>
<mi>&sigma;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mn>1</mn>
<mo>,</mo>
<mi>&mu;</mi>
<mo>-</mo>
<mi>c</mi>
<mi>&sigma;</mi>
<mo><</mo>
<mi>x</mi>
<mo><</mo>
<mi>&mu;</mi>
<mo>+</mo>
<mi>c</mi>
<mi>&sigma;</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>0</mn>
<mo>,</mo>
<mi>o</mi>
<mi>t</mi>
<mi>h</mi>
<mi>e</mi>
<mi>r</mi>
<mi>w</mi>
<mi>i</mi>
<mi>s</mi>
<mi>e</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, μ is the average of Gray Histogram value, and σ is the variance of Gray Histogram value, and c is coefficient factor, and x is pixel
Gray value.
4. the improved confinement growing method according to claim 3 for the segmentation of optic cup image, it is characterised in that：Segmentation regards
After cup：
(1) seed point p selected before is marked；
(2) seed point p and field pixel p around it are calculatediThe distance between d；
(3) according to distance d, whether this pixel where judgment threshold TP can be integrated into growth district：
TP=| MM×M-MN×N|, 0 ＜ d≤D (7)
If TP is less than or equal to corresponding threshold value, the pixel is integrated into seed point region；This paper threshold values are according to gray scale
In histogram between mountain valley difference (255-240=15) and set；
Wherein MM×M,MN×NPixel P is represented respectivelyiThe gray average and seed point p of regioniGray value, D represent distance face
Dividing value, this paper critical distances are 0.5 times of optic disk radiuses.
5. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：Extract G
In the step of passage and gray processing, it is processed for improving processing speed for gray level image, we are in extraction G channel components
Gray processing processing is carried out afterwards.
6. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：Extraction
The grey level histogram of eye fundus image is utilized in the step of ROI, background is dark in the picture, and optic disk optic cup region is light color
's；Light pixel in object produces right peak on the histogram, and substantial amounts of gray level is produced on a left side for histogram in background
Peak；Object boundary nearby has that the number of pixels of gray level between two peak values is relatively fewer, so as to generate between two peaks
Paddy；Selection paddy will obtain rational border as gray threshold；The threshold value of optic disk is may determine that according to grey level histogram.
7. the improved confinement growing method according to claim 6 for the segmentation of optic cup image, it is characterised in that：This method
Middle threshold value selection is 180 and then carries out a closed operation on the basis of being 180 in threshold value, eliminates tiny inc region, most
The region for forming a closing afterwards is ROI region.
8. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：Calculate
Opening Operation Definition in the step of ROI geometric centers is：
B is the element of an optic disk planform, 30 pixels of radius；
IBIt is original-gray image；
Closed operation is defined as：
B is the element of an optic disk planform, radius 20；
ICIt is the image after opening operation processing；
Morphologic corrosion and expansion respectively with Θ andRepresent, tiny and independent zonule can be removed by opening operation, corresponding
In bright lesion region, such as the druse in eye fundus image；Open operation and eliminate its medium and small and isolated structure
Element；The optimum size of structural element is drawn by our all analysis of experimental data, including different illuminations and some
Tiny change；Closed operation has filled up breach tiny in ROI region and has been connected to region tiny in optic disk region.
9. the improved confinement growing method according to claim 1 for the segmentation of optic cup image, it is characterised in that：Choose kind
The step of son point, this part of selected seed point include following six step：
(1) eye fundus image ROI gray-scale map is converted into by bianry image by the method for threshold value；
(2) remove tiny region with morphology closed operation and a continuous ROI is only provided；
(3) ROI geometric center is calculated；
(4) if geometric center pixel is 255, it is seed point to automatically select geometric center；
(5) if geometric center pixel is not 255, using geometric center as round dot, using 0.5 times of optic disk radius as hunting zone, 5
Length in pixels is scale, every time 45 ° of rotation, and obtains the maximum point of gray value on scale, i.e., in Figure 5 in shown 49 points
Choose a maximum point of radian alternately seed point；
(6) by this rule, alternative seed point search is carried out in the whole field of search；Finally one is chosen in all alternative seed points
The maximum point of individual gray value is seed point；
After meeting conditions above this can be marked to put the seed point as Region growing segmentation algorithm automatically.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710363040.6A CN107346545A (en) | 2017-05-22 | 2017-05-22 | Improved confinement growing method for the segmentation of optic cup image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710363040.6A CN107346545A (en) | 2017-05-22 | 2017-05-22 | Improved confinement growing method for the segmentation of optic cup image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107346545A true CN107346545A (en) | 2017-11-14 |
Family
ID=60253354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710363040.6A Pending CN107346545A (en) | 2017-05-22 | 2017-05-22 | Improved confinement growing method for the segmentation of optic cup image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107346545A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108444934A (en) * | 2018-01-30 | 2018-08-24 | 四川沐迪圣科技有限公司 | A kind of automatic segmentation of composite material debonding defect and quantization method |
| CN108564564A (en) * | 2018-03-09 | 2018-09-21 | 华南理工大学 | Based on the medical image cutting method for improving fuzzy connectedness and more seed points |
| CN109509199A (en) * | 2018-10-10 | 2019-03-22 | 华南理工大学 | A kind of medical image Organizational Intelligence dividing method based on three-dimensional reconstruction |
| CN109816637A (en) * | 2019-01-02 | 2019-05-28 | 电子科技大学 | The detection method in hard exudate region in a kind of eye fundus image |
| CN109949348A (en) * | 2019-01-22 | 2019-06-28 | 天津大学 | A kind of error hiding minimizing technology based on super-pixel movement statistics |
| CN110009612A (en) * | 2019-03-28 | 2019-07-12 | 安徽皓视光电科技有限公司 | Mobile lens sight glass image partition method based on sub-pixel precision |
| WO2020125318A1 (en) * | 2018-12-19 | 2020-06-25 | 上海鹰瞳医疗科技有限公司 | Glaucoma image recognition method and device and diagnosis system |
| CN113192099A (en) * | 2021-04-30 | 2021-07-30 | 上海联影医疗科技股份有限公司 | Tissue extraction method, device, equipment and medium |
| CN113935991A (en) * | 2021-12-13 | 2022-01-14 | 浙江托普云农科技股份有限公司 | Corn cross section parameter measuring method and system, electronic equipment and storage medium |
| CN113989167A (en) * | 2021-12-27 | 2022-01-28 | 杭州爱科科技股份有限公司 | Contour extraction method, device, equipment and medium based on seed point self-growth |
| CN115170446A (en) * | 2022-09-06 | 2022-10-11 | 聊城市洛溪信息科技有限公司 | Self-adaptive metal plate image enhancement method based on morphological processing |
| CN115187592A (en) * | 2022-09-08 | 2022-10-14 | 山东奥洛瑞医疗科技有限公司 | Image segmentation method suitable for nuclear magnetic resonance image |
| CN115439481A (en) * | 2022-11-09 | 2022-12-06 | 青岛平电锅炉辅机有限公司 | Deaerator welding quality detection method based on image processing |
| CN115880299A (en) * | 2023-03-03 | 2023-03-31 | 山东时代塑胶有限公司 | Quality detection system of lightweight concrete composite self-insulation external wall panel |
| CN116309549A (en) * | 2023-05-11 | 2023-06-23 | 爱尔眼科医院集团股份有限公司 | Fundus region detection method, fundus region detection device, fundus region detection equipment and readable storage medium |
| CN116309593A (en) * | 2023-05-23 | 2023-06-23 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120201460A1 (en) * | 2011-02-08 | 2012-08-09 | Kyungpook National University Industry-Academic Cooperation Foundation | Image segmentation devices and methods based on sequential frame image of static scene |
| CN103093457A (en) * | 2012-12-28 | 2013-05-08 | 中国科学院深圳先进技术研究院 | Detecting system and detecting method for ultrasonic blood vessel boundaries |
| CN104835157A (en) * | 2015-05-04 | 2015-08-12 | 北京工业大学 | Eye fundus image optical cup automatic segmentation method based on improved PDE image repairing |
| CN105303526A (en) * | 2015-09-17 | 2016-02-03 | 哈尔滨工业大学 | Ship target detection method based on coastline data and spectral analysis |
| CN105466950A (en) * | 2016-01-14 | 2016-04-06 | 上海孚兴电子科技有限公司 | Bamboo dice online detection method and system based on vision |
-
2017
- 2017-05-22 CN CN201710363040.6A patent/CN107346545A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120201460A1 (en) * | 2011-02-08 | 2012-08-09 | Kyungpook National University Industry-Academic Cooperation Foundation | Image segmentation devices and methods based on sequential frame image of static scene |
| CN103093457A (en) * | 2012-12-28 | 2013-05-08 | 中国科学院深圳先进技术研究院 | Detecting system and detecting method for ultrasonic blood vessel boundaries |
| CN104835157A (en) * | 2015-05-04 | 2015-08-12 | 北京工业大学 | Eye fundus image optical cup automatic segmentation method based on improved PDE image repairing |
| CN105303526A (en) * | 2015-09-17 | 2016-02-03 | 哈尔滨工业大学 | Ship target detection method based on coastline data and spectral analysis |
| CN105466950A (en) * | 2016-01-14 | 2016-04-06 | 上海孚兴电子科技有限公司 | Bamboo dice online detection method and system based on vision |
Non-Patent Citations (1)
| Title |
|---|
| 刘振宇 等: "改进区域生长算法在视杯图像分割中的应用", 《辽宁大学学报（自然科学版）》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108444934A (en) * | 2018-01-30 | 2018-08-24 | 四川沐迪圣科技有限公司 | A kind of automatic segmentation of composite material debonding defect and quantization method |
| CN108444934B (en) * | 2018-01-30 | 2021-12-10 | 四川沐迪圣科技有限公司 | Automatic segmentation and quantification method for debonding defect of composite material |
| CN108564564A (en) * | 2018-03-09 | 2018-09-21 | 华南理工大学 | Based on the medical image cutting method for improving fuzzy connectedness and more seed points |
| CN109509199B (en) * | 2018-10-10 | 2021-07-20 | 华南理工大学 | Medical image organization intelligent segmentation method based on three-dimensional reconstruction |
| CN109509199A (en) * | 2018-10-10 | 2019-03-22 | 华南理工大学 | A kind of medical image Organizational Intelligence dividing method based on three-dimensional reconstruction |
| WO2020125318A1 (en) * | 2018-12-19 | 2020-06-25 | 上海鹰瞳医疗科技有限公司 | Glaucoma image recognition method and device and diagnosis system |
| CN109816637A (en) * | 2019-01-02 | 2019-05-28 | 电子科技大学 | The detection method in hard exudate region in a kind of eye fundus image |
| CN109949348A (en) * | 2019-01-22 | 2019-06-28 | 天津大学 | A kind of error hiding minimizing technology based on super-pixel movement statistics |
| CN109949348B (en) * | 2019-01-22 | 2023-04-18 | 天津大学 | Method for removing mismatching based on super-pixel motion statistics |
| CN110009612A (en) * | 2019-03-28 | 2019-07-12 | 安徽皓视光电科技有限公司 | Mobile lens sight glass image partition method based on sub-pixel precision |
| CN110009612B (en) * | 2019-03-28 | 2023-09-26 | 安徽皓视光电科技有限公司 | Sub-pixel precision-based mobile phone lens window glass image segmentation method |
| CN113192099B (en) * | 2021-04-30 | 2023-02-17 | 上海联影医疗科技股份有限公司 | Tissue extraction method, device, equipment and medium |
| CN113192099A (en) * | 2021-04-30 | 2021-07-30 | 上海联影医疗科技股份有限公司 | Tissue extraction method, device, equipment and medium |
| CN113935991A (en) * | 2021-12-13 | 2022-01-14 | 浙江托普云农科技股份有限公司 | Corn cross section parameter measuring method and system, electronic equipment and storage medium |
| CN113989167B (en) * | 2021-12-27 | 2022-03-29 | 杭州爱科科技股份有限公司 | Contour extraction method, device, equipment and medium based on seed point self-growth |
| CN113989167A (en) * | 2021-12-27 | 2022-01-28 | 杭州爱科科技股份有限公司 | Contour extraction method, device, equipment and medium based on seed point self-growth |
| CN115170446B (en) * | 2022-09-06 | 2022-11-18 | 聊城市洛溪信息科技有限公司 | Self-adaptive metal plate image enhancement method based on morphological processing |
| CN115170446A (en) * | 2022-09-06 | 2022-10-11 | 聊城市洛溪信息科技有限公司 | Self-adaptive metal plate image enhancement method based on morphological processing |
| CN115187592A (en) * | 2022-09-08 | 2022-10-14 | 山东奥洛瑞医疗科技有限公司 | Image segmentation method suitable for nuclear magnetic resonance image |
| CN115439481A (en) * | 2022-11-09 | 2022-12-06 | 青岛平电锅炉辅机有限公司 | Deaerator welding quality detection method based on image processing |
| CN115439481B (en) * | 2022-11-09 | 2023-02-21 | 青岛平电锅炉辅机有限公司 | Deaerator welding quality detection method based on image processing |
| CN115880299A (en) * | 2023-03-03 | 2023-03-31 | 山东时代塑胶有限公司 | Quality detection system of lightweight concrete composite self-insulation external wall panel |
| CN116309549A (en) * | 2023-05-11 | 2023-06-23 | 爱尔眼科医院集团股份有限公司 | Fundus region detection method, fundus region detection device, fundus region detection equipment and readable storage medium |
| CN116309549B (en) * | 2023-05-11 | 2023-10-03 | 爱尔眼科医院集团股份有限公司 | Fundus region detection method, fundus region detection device, fundus region detection equipment and readable storage medium |
| CN116309593A (en) * | 2023-05-23 | 2023-06-23 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
| CN116309593B (en) * | 2023-05-23 | 2023-09-12 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107346545A (en) | Improved confinement growing method for the segmentation of optic cup image | |
| CN109493954B (en) | SD-OCT image retinopathy detection system based on category distinguishing and positioning | |
| CN104463140B (en) | A kind of colored eye fundus image optic disk automatic positioning method | |
| Xu et al. | Automated optic disk boundary detection by modified active contour model | |
| CN106651888B (en) | Colour eye fundus image optic cup dividing method based on multi-feature fusion | |
| Lu et al. | Automatic optic disc detection from retinal images by a line operator | |
| CN106204555B (en) | A kind of optic disk localization method of combination Gbvs model and phase equalization | |
| Liu et al. | ARGALI: an automatic cup-to-disc ratio measurement system for glaucoma analysis using level-set image processing | |
| CN106530283A (en) | SVM (support vector machine)-based medical image blood vessel recognition method | |
| CN104794721B (en) | A kind of quick optic disk localization method based on multiple dimensioned spot detection | |
| CN108073918B (en) | Method for extracting blood vessel arteriovenous cross compression characteristics of fundus retina | |
| CN110555845A (en) | Fundus OCT image identification method and equipment | |
| Jaafar et al. | Automated detection and grading of hard exudates from retinal fundus images | |
| CN104392240A (en) | Parasite egg identification method based on multi-feature fusion | |
| CN109035227A (en) | The system that lung tumors detection and diagnosis is carried out to CT image | |
| CN107480644A (en) | The positioning of optic disk and dividing method, device and storage medium in eye fundus image | |
| Jaafar et al. | Detection of exudates from digital fundus images using a region-based segmentation technique | |
| Surendiran et al. | Segmentation of optic disc and cup using modified recurrent neural network | |
| CN109523524A (en) | A kind of eye fundus image hard exudate detection method based on integrated study | |
| Di Leo et al. | An improved procedure for the automatic detection of dermoscopic structures in digital ELM images of skin lesions | |
| CN108846827B (en) | Method for rapidly segmenting fundus optic disk based on multiple circles | |
| CN109559321A (en) | A kind of sonar image dividing method and equipment | |
| Mendonça et al. | Segmentation of the vascular network of the retina | |
| CN108320799A (en) | Image analysis and recognition method for lateral flow paper strip disease diagnosis | |
| CN106384343A (en) | Morphological processing based hard exudation area detecting method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171114 |