CN105095907B - A kind of heterogeneous fibre identification method of cotton based on RBF neural - Google Patents
A kind of heterogeneous fibre identification method of cotton based on RBF neural Download PDFInfo
- Publication number
- CN105095907B CN105095907B CN201410198601.8A CN201410198601A CN105095907B CN 105095907 B CN105095907 B CN 105095907B CN 201410198601 A CN201410198601 A CN 201410198601A CN 105095907 B CN105095907 B CN 105095907B
- Authority
- CN
- China
- Prior art keywords
- image
- gray level
- cotton
- rbf neural
- identification method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
- Preliminary Treatment Of Fibers (AREA)
Abstract
The invention discloses a kind of heterogeneous fibre identification method of cotton based on RBF neural, the image of detected sample is gathered in the heterogeneous fibre identification method of the cotton, and the image to collecting is pre-processed and binary conversion treatment, and window division is carried out to the image after binary conversion treatment, target window is determined according to the image after binary conversion treatment, and extract the feature vector of each target window, grader is used as using trained RBF neural, differentiate whether the corresponding pixel of each target window is hetero fibre according to each feature vector, and mark hetero fibre.The present invention is used as grader using trained RBF neural, pattern-recognition is carried out to cotton hetero fibre, compared with the heterogeneous fibre identification method of the existing cotton based on colour recognition, substantially increase the heterogeneous fiber recognition of cotton and differentiate precision, reduce probability of miscarriage of justice.
Description
Technical field
The present invention relates to the heterogeneous fibre identification technical field of cotton, more particularly to a kind of cotton based on RBF neural
Hetero fibre discrimination method.
Background technology
Cotton hetero fibre, also known as " three ", refer to be mixed into non-woolliness fiber in cotton, including chemical fibre, silk, fiber crops,
The impurity such as hair and plastics.Divided by shape, " three " can be divided into block impurity, strip impurity and Filamentous impurity three classes.
" three " source is extremely wide, and each flow of the cotton from picking to processing can be mixed into " three ".Mixed in cotton
For " three " entered although proportion very little, harm is very big.In cotton processing, " three " mixed in cotton are easy to
Broken to be divided into the small fault of the thinner fiber of countless smallers, quantity is multiplied, and random distribution is in cotton, thus is difficult to
Remove.The small fault of these fibers easily makes Yarn break, reduces production efficiency and causes cloth dyeing uneven, influences outside cloth
See.In view of the above problems, cotton spinning enterprise is had to, a large amount of manpowers of tissue packet-by-packet open cotton bag before cotton feeds intake, and block-by-block is torn
Loosely, pick, also picked clearly at any time in the various processes of spinning, therefore cost increases, inefficiency, causes China's cotton spinning by root
Quality reduces, and competitiveness weakens.Therefore, it is in current cotton spinning technical field to realize automatic sorting and reject cotton hetero fibre
One of hot spot of research.How the key technology of automatic sorting and rejecting cotton hetero fibre accurately identifies and position " three ",
It is the necessary condition for implementing to improve gined cotton quality.
In recent years, with the fast development of computer image processing technology, realizing the identification of " three " and being accurately positioned is
Feasible.Had much using the method for machine vision sorting " three " at present, mainly by gathering cotton samples pictures,
Then image steganalysis cotton hetero fibre is utilized, and according to recognition result into being positioned.Existing cotton hetero fibre
It is more in identification and location technology to be completed using colour recognition, as Publication No. 201269857 Chinese patent application in disclose
One kind is based on the colored and black and white identification heterogeneous fiber recognition method of cotton.The recognition methods efficiency is low, the feelings judged by accident easily occurs
Condition, accuracy of identification need to be further improved.
The content of the invention
In view of the deficiencies of the prior art, the present invention provides a kind of heterogeneous fibre identification of the cotton based on RBF neural
Method.
A kind of heterogeneous fibre identification method of cotton based on RBF neural, including:
(1) image of detected sample is gathered, and the image to collecting pre-processes, and obtains pretreated figure
Picture;
(2) binaryzation is carried out to pretreated image, obtains binary image;
(3) binary image is divided into several windows, to be used as target comprising window of the pixel value as 0 pixel
Window;
(4) according to the pixel value of all pixels point in binary image in each target window, using gray scale symbiosis square
Battle array analytic approach obtains the gray level co-occurrence matrixes of each target window;
(5) each gray level co-occurrence matrixes are directed to, calculate the characteristic value of the gray level co-occurrence matrixes, and the feature to be calculated
It is worth the feature vector as the gray level co-occurrence matrixes;
(6) each gray level co-occurrence matrixes pair will be differentiated in the feature vector input grader of each gray level co-occurrence matrixes respectively
Whether the target window answered is hetero fibre, and the window that identification result is hetero fibre is marked in binary image,
The grader is trained RBF neural.
More images that detected sample is obtained by CCD in the present invention.In step (6) according to the identification result of grader into
Line flag, completes the positioning of the hetero fibre to identifying.
The present invention the heterogeneous fibre identification method of cotton in gather detected sample image, and the image to collecting into
Row pretreatment and binary conversion treatment, and window division is carried out to the image after binary conversion treatment, after binary conversion treatment
Image pixel, determines target window, and extracts the feature vector of each target window, using trained RBF neural as
Grader, differentiates whether the corresponding pixel of each target window is hetero fibre according to each feature vector, and marks heterogeneous
Fiber.
RBF neural (radial basis function neural network, Radial Basis Function neural network)
Multidimensional nonlinear mapping ability it is strong, can arbitrary accuracy approach arbitrary nonlinear function, and with global approximation capability, with
The precision for being used as grader, the heterogeneous fiber recognition of cotton being greatly improved of trained RBF neural, with existing base
Compare in the heterogeneous fibre identification method of the cotton of colour recognition, substantially increase the heterogeneous fiber recognition of cotton and differentiate precision, reduce
Probability of miscarriage of justice.
RBF neural in the present invention has the characteristics that convergence speed is fast, and each radial basis function is obtained by training
Center, variance and hidden layer lead the weights of output layer.According to the difference of radial basis function center choosing method, there is following instruction
Practice method:Randomly select center method, Self-organizing Selection Center method and have supervision Selection Center method etc..It is preferably, of the invention
It is middle that RBF neural is trained using the method based on OCA clusters (i.e. the objective clusters of OCA), i.e., obtained by OCA clustering procedures
Take the center of radial basis function.OCA clustering procedures are based on Self-organizing data mining thought GMDH (data packet processing method), can
Adaptively determine cluster number and Different categories of samples, process are simply easily achieved.
Preprocessing process in the step (1) includes the following steps:
(1-1) carries out gray processing to the image collected using weighted mean method and handles to obtain gray level image;
(1-2) carries out light correction by grey level histogram regulationization to gray level image;
Gray level image after (1-3) corrects light using gaussian filtering carries out denoising.
Weights when weighted mean method handles image gray processing in step (1-1) can be adjusted according to actual conditions.It is logical
Pretreatment is crossed, image is switched into gray level image, then carries out light correction, the uneven influence to image of light is eliminated, goes forward side by side
Row denoising, eliminates influence of noise, improves the precision of subsequent treatment.
Two are carried out to pretreated image using based on otsu algorithms and local Threshold Segmentation Algorithm in the step (2)
Value, it is specific as follows:
(2-1) determines the n neighborhood territory pixels point of current pixel point, and calculate and work as to any one pixel in gray level image
The average pixel value of the corresponding all n neighborhood territory pixels points of preceding pixel point;
(2-2) determines the optimal threshold of gray level image using otsu algorithms, will to any one pixel in gray level image
The corresponding average pixel value of current pixel point meets condition:
pm> dm- 5, and pm> T,
The pixel value for then making current pixel point is 255, and otherwise, the pixel value for making the current pixel point is 0, pmFor current picture
The pixel value of vegetarian refreshments, dmFor the corresponding average pixel value of current pixel point, T is the optimal threshold of the gray level image.
Why conventional threshold values algorithm cannot effectively split the heterogeneous fibre image of cotton, there is following reason:1. it cannot be filled
Divide the Pixel Information using image neighborhood pixels point, thus cannot effectively detect Filamentous cotton hetero fibre;2. some occurs
The neighborhood of pixel is fully fallen in target or background area just, causes block and strip cotton hetero fibre core wrong
Think cotton.
Due to each pixel n neighborhood territory pixels point of use, image neighborhood pixels infomation detection wire vent can be made full use of
Shape cotton hetero fibre.Since otsu threshold methods (otsu algorithms) are global thresholds, work as all n for some pixel occur
Neighborhood territory pixel point fully falls in situation in target area (hetero fibre region) or background area (non-) hetero fibre region just
When, block and strip cotton hetero fibre can be effectively detected according to global threshold.And due to the width of cotton filiform impurity
Generally only 1~2 pixel, so all n neighborhoods for being not in some pixel entirely fall in Filamentous cotton hetero fibre
In region.
The present invention combines the good global segmentation ability of otsu algorithms and the segmentation thought of local threshold, overcomes classic map
As partitioning algorithm cannot make full use of image neighborhood pixels information and when the neighborhood of some pixel fully falls in target or the back of the body just
The shortcomings that segmentation errors are caused during scene area, realizes effective detection of block, strip and Filamentous cotton hetero fibre.
Preferably, the n is 3~5.
The value of n is directly related to the precision of discriminating, and n values are bigger, differentiates that precision is higher, but n, which crosses conference, to be caused to differentiate
Inefficiency, therefore preferably, n=3.
Preferably, the window in the step (3) is rectangular window.
Further, the size of the rectangular window is 10 × 10~16 × 16.
The shapes and sizes of window can be set according to actual conditions, and ensure that whole image can be divided by all windows.Window
Mouth size is in units of pixel.
Using first before the gray level co-occurrence matrixes of each target window of gray level co-occurrence matrixes analytic approach acquisition in the step (4)
It it is 8 grades or 16 grades by the gray-scale compression of binary image.
By compressing number of greyscale levels, calculation amount is reduced, improves the efficiency of gray feature matrix extraction.
Characteristic value in the step (5) includes angular second moment, contrast, entropy and inverse difference moment.
Wherein, energy f1(angular second moment), according to its formula:
It is calculated, energy is the quadratic sum of each element in gray level co-occurrence matrixes, reflects image texture fineness degree and ash
Spend the degree that is evenly distributed.f1Bigger, energy is bigger, and texture is thicker;Conversely, f1Smaller, energy is smaller, and texture is thinner.
Contrast f2(the moment of inertia), according to formula:
Calculate, contrast reflects the clarity of image.f2Greatly, the ditch of texture is deeper, and effect is more clear;Conversely, f2More
Small, ditch is more shallow, and effect is fuzzyyer.
Entropy f3According to formula:
It is calculated, entropy reflects the complexity and non-uniformity of image texture.f3Bigger, texture is more complicated, gray scale
Element difference is smaller in co-occurrence matrix;Conversely, f3Smaller, texture is simpler, and element difference is bigger.
Inverse difference moment f4, according to formula:
Calculate, inverse difference moment reflects the homogeney of image texture, i.e. change size between image texture different zones.
Wherein, i, j are the coordinate of each element in gray level co-occurrence matrixes, and L is the sum of the row or column of gray level co-occurrence matrixes,
(row, column number is identical in the present invention),For the value of j-th of element of the i-th row in gray level co-occurrence matrixes.
The present invention carries out pattern-recognition using trained RBF neural as grader, to cotton hetero fibre, with
The heterogeneous fibre identification method of the existing cotton based on colour recognition compares, and substantially increases the heterogeneous fiber recognition of cotton and differentiates essence
Degree, reduces probability of miscarriage of justice.And use the Threshold Segmentation Algorithm based on otsu algorithms and local threshold, it is possible to achieve to block, bar
Shape is precisely separating with Filamentous cotton hetero fibre, and Filamentous cotton hetero fibre cannot effectively be split by overcoming classical partitioning algorithm
The shortcomings that.
Embodiment
Below in conjunction with specific embodiment, the present invention is described in detail.
A kind of heterogeneous fibre identification method of cotton based on RBF neural, including:
(1) image of detected sample is gathered, and the image to collecting pre-processes, and obtains pretreated figure
Picture.By pretreatment, the uneven precision subsequently differentiated with the influence of noise, raising of light is eliminated, preprocessing process is as follows:
(1-1) carries out gray processing to the image collected using weighted mean method and handles to obtain gray level image;
(1-2) carries out light correction by grey level histogram regulationization to gray level image;
Gray level image after (1-3) corrects light using gaussian filtering carries out denoising.
(2) binaryzation is carried out to pretreated image, obtains binary image, it is specific as follows:
(2-1) determines the n neighborhood territory pixels point of current pixel point, and calculate and work as to any one pixel in gray level image
The average pixel value of the corresponding all n neighborhood territory pixels points of preceding pixel point;
(2-2) determines the optimal threshold of gray level image using otsu algorithms, will to any one pixel in gray level image
The corresponding average pixel value of current pixel point meets condition:
pm> dm- 5, and pm> T,
The pixel value for then making current pixel point is 255, and otherwise, the pixel value for making the current pixel point is 0, pmFor current picture
The pixel value of vegetarian refreshments, dmFor the corresponding average pixel value of current pixel point, T is the optimal threshold of the gray level image.
N=3 in the present embodiment, n neighborhood territory pixel point are 3 × 3 neighborhood territory pixel point of current pixel point.
(3) binary image is divided into several windows, to be used as target comprising window of the pixel value as 0 pixel
Window.
It is the rectangular window that size is 16 × 16 in the present embodiment.
(4) it is 8 grades or 16 grades (being 16 grades in the present embodiment) by the gray-scale compression of binary image, according to compressed
The pixel value of all pixels point in binary image in each target window, is obtained each using gray level co-occurrence matrixes analytic approach
The gray level co-occurrence matrixes of target window.
(5) each gray level co-occurrence matrixes are directed to, according to the characteristic value of the compressed calculating gray level co-occurrence matrixes, and in terms of
Feature vector of the obtained characteristic value as the gray level co-occurrence matrixes.
Characteristic value in the present embodiment includes angular second moment, contrast, entropy and inverse difference moment.Wherein, angular second moment f1(energy
Amount), according to its formula:
It is calculated, angular second moment is the quadratic sum of each element in gray level co-occurrence matrixes, is energy, reflects image line
Manage fineness degree and intensity profile uniformity coefficient.f1Bigger, energy is bigger, and texture is thicker;Conversely, f1Smaller, energy is smaller, texture
It is thinner.
Contrast f2(the moment of inertia), according to formula:
Calculate, contrast reflects the clarity of image.f2Greatly, the ditch of texture is deeper, and effect is more clear;Conversely, f2More
Small, ditch is more shallow, and effect is fuzzyyer.
Entropy f3According to formula:
It is calculated, entropy reflects the complexity and non-uniformity of image texture.f3Bigger, texture is more complicated, gray scale
Element difference is smaller in co-occurrence matrix;Conversely, f3Smaller, texture is simpler, and element difference is bigger.
Inverse difference moment f4, according to formula:
Calculate, inverse difference moment reflects the homogeney of image texture, i.e. change size between image texture different zones.
Wherein, i, j are the coordinate of each element in gray level co-occurrence matrixes, and L is the sum of the row or column of gray level co-occurrence matrixes,
(row, column number is identical in the present invention),For the value of j-th of element of the i-th row in gray level co-occurrence matrixes.
(6) each gray level co-occurrence matrixes pair will be differentiated in the feature vector input grader of each gray level co-occurrence matrixes respectively
Whether the target window answered is hetero fibre, and the window that identification result is hetero fibre, this point are marked in binary image
Class device is trained RBF neural.
The above description is merely a specific embodiment, but protection scope of the present invention is not limited thereto, any
Those familiar with the art the invention discloses technical scope in, the change or replacement that can readily occur in, all should
It is included within the scope of the present invention.
Claims (7)
1. a kind of heterogeneous fibre identification method of cotton based on RBF neural, it is characterised in that including:
(1) image of detected sample is gathered, and the image to collecting pre-processes, and obtains pretreated image;
(2) binaryzation is carried out to pretreated image, obtains binary image;
(3) binary image is divided into several windows, to be used as target window comprising window of the pixel value as 0 pixel
Mouthful;
(4) according to the pixel value of all pixels point in binary image in each target window, using gray level co-occurrence matrixes point
Analysis method obtains the gray level co-occurrence matrixes of each target window;Described obtains each target window using gray level co-occurrence matrixes analytic approach
Mouthful gray level co-occurrence matrixes before first by the gray-scale compression of binary image be 8 grades or 16 grades;
(5) each gray level co-occurrence matrixes are directed to, calculate the characteristic value of the gray level co-occurrence matrixes, and the characteristic value to be calculated is made
For the feature vector of the gray level co-occurrence matrixes;
(6) it will differentiate that each gray level co-occurrence matrixes are corresponding in the feature vector input grader of each gray level co-occurrence matrixes respectively
Whether target window is hetero fibre, and the window that identification result is hetero fibre is marked in binary image,
The grader is trained RBF neural.
2. the heterogeneous fibre identification method of cotton as claimed in claim 1 based on RBF neural, it is characterised in that described
Preprocessing process in step (1) includes the following steps:
(1-1) carries out gray processing to the image collected using weighted mean method and handles to obtain gray level image;
(1-2) carries out light correction by grey level histogram regulationization to gray level image;
Gray level image after (1-3) corrects light using gaussian filtering carries out denoising.
3. the heterogeneous fibre identification method of cotton as claimed in claim 1 or 2 based on RBF neural, it is characterised in that institute
State in step (2) and binaryzation is carried out to pretreated image using based on otsu algorithms and local Threshold Segmentation Algorithm, specifically
It is as follows:
(2-1) determines the n neighborhood territory pixels point of current pixel point, and calculate current picture to any one pixel in gray level image
The average pixel value of the corresponding all n neighborhood territory pixels points of vegetarian refreshments;
(2-2) determines the optimal threshold of gray level image using otsu algorithms, to any one pixel in gray level image, if currently
The corresponding average pixel value of pixel meets condition:
pm>dm- 5, and pm>T,
The pixel value for then making current pixel point is 255, and otherwise, the pixel value for making the current pixel point is 0, pmFor current pixel point
Pixel value, dmFor the corresponding average pixel value of current pixel point, T is the optimal threshold of the gray level image.
4. the heterogeneous fibre identification method of cotton as claimed in claim 3 based on RBF neural, it is characterised in that described
N be 3~5.
5. the heterogeneous fibre identification method of cotton as claimed in claim 4 based on RBF neural, it is characterised in that described
Window in step (3) is rectangular window.
6. the heterogeneous fibre identification method of cotton as claimed in claim 5 based on RBF neural, it is characterised in that described
The size of rectangular window is 10 × 10~16 × 16.
7. the heterogeneous fibre identification method of cotton as claimed in claim 6 based on RBF neural, it is characterised in that described
Characteristic value in step (5) includes angular second moment, contrast, entropy and inverse difference moment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410198601.8A CN105095907B (en) | 2014-05-12 | 2014-05-12 | A kind of heterogeneous fibre identification method of cotton based on RBF neural |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410198601.8A CN105095907B (en) | 2014-05-12 | 2014-05-12 | A kind of heterogeneous fibre identification method of cotton based on RBF neural |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105095907A CN105095907A (en) | 2015-11-25 |
CN105095907B true CN105095907B (en) | 2018-05-11 |
Family
ID=54576292
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410198601.8A Active CN105095907B (en) | 2014-05-12 | 2014-05-12 | A kind of heterogeneous fibre identification method of cotton based on RBF neural |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105095907B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105624846A (en) * | 2016-02-17 | 2016-06-01 | 江南大学 | Method for determining positions of heterogeneous fibers in raw cotton |
CN105760877B (en) * | 2016-02-19 | 2019-03-05 | 天纺标检测认证股份有限公司 | A kind of woollen and cashmere recognizer based on gray level co-occurrence matrixes model |
CN109063821A (en) * | 2018-07-06 | 2018-12-21 | 新疆大学 | A kind of prediction technique and system based on cotton fiber color grade |
CN109543532B (en) * | 2018-10-19 | 2023-07-28 | 兰波(苏州)智能科技有限公司 | Fiber classification method based on classification standard and human behavior |
CN109492544B (en) * | 2018-10-19 | 2023-01-03 | 兰波(苏州)智能科技有限公司 | Method for classifying animal fibers through enhanced optical microscope |
CN113249949B (en) * | 2021-05-27 | 2023-05-05 | 孔华 | Cotton outlet quality judging system and method for automatic cotton ejection structure |
CN114897891B (en) * | 2022-07-12 | 2022-09-09 | 南通恒立机械设备有限公司 | Mixing uniformity detection method and system for spiral-bar mixer |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN201269857Y (en) * | 2008-10-22 | 2009-07-08 | 北京中棉机械成套设备有限公司 | Foreign fiber detection and metering device in lint cotton |
CN102660815A (en) * | 2012-03-30 | 2012-09-12 | 浙江理工大学 | Cotton foreign fiber automatic removing method and system based on image acquisition and DSP (digital signal processing) algorithm |
CN202717892U (en) * | 2012-03-30 | 2013-02-06 | 浙江理工大学 | Foreign fiber automatic cotton foreign fiber removal device based on ultraviolet detection mechanism |
-
2014
- 2014-05-12 CN CN201410198601.8A patent/CN105095907B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN201269857Y (en) * | 2008-10-22 | 2009-07-08 | 北京中棉机械成套设备有限公司 | Foreign fiber detection and metering device in lint cotton |
CN102660815A (en) * | 2012-03-30 | 2012-09-12 | 浙江理工大学 | Cotton foreign fiber automatic removing method and system based on image acquisition and DSP (digital signal processing) algorithm |
CN202717892U (en) * | 2012-03-30 | 2013-02-06 | 浙江理工大学 | Foreign fiber automatic cotton foreign fiber removal device based on ultraviolet detection mechanism |
Also Published As
Publication number | Publication date |
---|---|
CN105095907A (en) | 2015-11-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105095907B (en) | A kind of heterogeneous fibre identification method of cotton based on RBF neural | |
CN106407917B (en) | The retinal vessel extracting method and system distributed based on Dynamic Multi-scale | |
Gulhane et al. | Detection of diseases on cotton leaves and its possible diagnosis | |
Savkare et al. | Automatic system for classification of erythrocytes infected with malaria and identification of parasite's life stage | |
CN106529602B (en) | A kind of millimeter-wave image automatic target recognition method and device | |
CN106980852B (en) | Based on Corner Detection and the medicine identifying system matched and its recognition methods | |
Renugambal et al. | Application of image processing techniques in plant disease recognition | |
CN106127735B (en) | A kind of facilities vegetable edge clear class blade face scab dividing method and device | |
CN105701812B (en) | Visual identification system suitable for cotton picking robot | |
CN105844213B (en) | Green fruit recognition method | |
CN116958572B (en) | Leaf disease and pest area analysis method in fruit tree breeding | |
CN106096667A (en) | Bill images sorting technique based on SVM | |
CN103218615B (en) | Face judgment method | |
Feng et al. | A separating method of adjacent apples based on machine vision and chain code information | |
CN115082465B (en) | Wool and cashmere classification method based on scanning electron microscope image processing | |
Ji et al. | In-field automatic detection of maize tassels using computer vision | |
CN109325498A (en) | The Vein extraction algorithm of Canny operator is improved based on window dynamic threshold | |
CN108961301A (en) | It is a kind of based on the unsupervised Chaetoceros image partition method classified pixel-by-pixel | |
CN104899595A (en) | Male and female silkworm chrysalis sorting and counting device based on SIFT (Scale Invariant Feature Transform) feature image | |
Septiarini et al. | Pixel quantification and color feature extraction on leaf images for oil palm disease identification | |
CN109472779A (en) | A kind of yarn appearance characteristic parameter extraction and analysis method based on morphosis | |
CN105631816B (en) | A kind of iris image noise classification detection method | |
CN110288616A (en) | A method of based on dividing shape and RPCA to divide hard exudate in eye fundus image | |
CN102831379A (en) | Face image recognition method and device | |
Manjaramkar et al. | Connected component clustering based hemorrhage detection in color fundus images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |