CN106023151A - Traditional Chinese medicine tongue manifestation object detection method in open environment - Google Patents
Traditional Chinese medicine tongue manifestation object detection method in open environment Download PDFInfo
- Publication number
- CN106023151A CN106023151A CN201610300494.4A CN201610300494A CN106023151A CN 106023151 A CN106023151 A CN 106023151A CN 201610300494 A CN201610300494 A CN 201610300494A CN 106023151 A CN106023151 A CN 106023151A
- Authority
- CN
- China
- Prior art keywords
- image
- tongue
- segmentation
- value
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
- A61B5/0077—Devices for viewing the surface of the body, e.g. camera, magnifying lens
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Theoretical Computer Science (AREA)
- Pathology (AREA)
- Quality & Reliability (AREA)
- General Physics & Mathematics (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a traditional Chinese medicine tongue manifestation object detection method in an open environment, and relates to medical image processing. The method comprises the following steps: 1, inputting an image A acquired in the open environment; 2, obtaining a correction image B by performing color correction on the acquired image A; 3, performing image segmentation on the correction image B obtained in the second step; 4, performing area feature determination on an image C obtained in the third step; and 5, performing texture feature determination on a candidate tongue body area D obtained in the fourth step. First of all, preprocessing of the color correction is performed on the image so that an influence brought by an external light source color temperature is reduced; then multiple communicating areas are obtained by segmenting the image; the candidate tongue body area is obtained by performing the feature determination on each communicating area; and finally, determination is carried out through comparing area texture features, and thus whether the candidate tongue body area is a tongue manifestation is determined. The purpose of performing object detection on the tongue manifestation in the image is finally realized, i.e., whether the tongue manifestation exists in a current picture is determined, and if so, where a tongue manifestation object is is determined.
Description
Technical field
The present invention relates to Medical Image Processing, especially relate to Tongue object detection method under a kind of open environment.
Background technology
The tongue image that picture of the tongue presents after i.e. people stretches out tongue, the acquisition of Traditional Chinese Medicine picture of the tongue relies primarily on doctor's naked eyes and sees
Examining, in recent years, the development of information technology has promoted Tongue analysis to objectify, digitized and the process of automatization.Both at home and abroad
Scholar has carried out many useful explorations to this, and develops some Analysis of Lingual Picture systems, and (Jiang depends on to achieve preferable effect
I. computerized Evolution of Tongue Inspection of TCM system [J]. China's combination of Chinese and Western medicine magazine, 2000,20 (2): 145-147;Cai Yihang, Liu Changjiang,
Shen Lansun. the design [J] of novel tongue image analysis. observation and control technology, 2005,24 (5): 34-36.), but the bat of these systems
Take the photograph what environment was usually fixed, i.e. shooting, collecting in the environment of airtight, illumination are stablized, and these equipment and instruments are the most stupid
Weight, the most portable, price costly, have certain limitation (Liu Feng. in auxiliary Chinese patent medicine using system of Chinese medicine
Analysis of Lingual Picture research [D]. Xiamen: Xiamen University, master thesis, 2007.).Along with smart mobile phone, panel computer etc. are mobile
Popularizing of equipment, carries out tongue image acquisition by mobile device, it is thus achieved that personal health information gradually becomes under open natural environment
It it is a developing direction.But thing followed problem is, owing to gathering picture of the tongue under open environment, there is light source color temperature, light
The impact of many uncertain factors such as power, shooting angle, equipment difference so that the final image obtained is with fixed environment phase
Ratio, often there is larger difference, follow-up analysis brought difficulty in the picture of the tongue image of collection, even due to photographer,
The image gathered may not be available for the picture of the tongue image analyzed.Therefore, when under open environment, collection picture of the tongue is analyzed,
The serious forgiveness of enhancing system and robustness seem particularly significant.Before picture of the tongue is analyzed, image is carried out picture of the tongue target inspection
Survey and be favorably improved the follow-up analysis speed of system and accuracy rate, it helps doctor of traditional Chinese medicine observes tongue intuitively, accurately, easily
As, improve the speed of diagnosis.Whether picture of the tongue target detection purpose is to judge to exist in the image gathered suitably to be available for follow-up point
The picture of the tongue of analysis, if there is picture of the tongue in the picture gathered, need to detect which position at image.The standard of the target detection of picture of the tongue
Really property directly affects the serious forgiveness of whole system.
At present, the research that launches for Tongue target detection under open environment is the most few, connects most with the inventive method
A kind of based on mobile terminal the Tongue that near technology designs for Xiamen Qiangben Science and Technology Co., Ltd analyzes system (Wang Bo
Bright. a kind of Tongue based on mobile terminal analyzes system: Chinese invention patent is open, 2014200304657 [P] .2014-
12-03.), the picture of the tongue to mobile device collection detects, analyzes.Although this system achieves picture of the tongue detection function, but also deposits
In place of some shortcomings:
1, needing target cutting fritter to be detected, be transformed into hough space ballot, algorithm is complicated, and detection speed is slow;And
Mobile device is used to carry out tongue image acquisition, detection under open environment, often higher to the requirement of real-time of processing procedure;
2, processing procedure needs to describe picture of the tongue model by more parameter, more complicated, it is achieved relatively difficult;
3, need the training mass data of supervision, need to mark manually foreground and background, thus set up detection
Model, training process difficulty is big;
When 4, carrying out data training, it is desirable to sample is the most, high to the dependency of data set.
Summary of the invention
Present invention aims to existing collection during picture of the tongue under open environment and there is many weak points,
As: the impact of many uncertain factors such as light source color temperature, light intensity, shooting angle, equipment difference so that the final figure obtained
As compared with fixed environment, often there is larger difference in the picture of the tongue image of collection, follow-up analysis is brought difficulty, even due to
The reason of photographer, may not be available for the problem such as picture of the tongue image analyzed in the image of collection.The present invention provides one to open
Putting Tongue object detection method under environment, it is suitable that the method can quickly and accurately judge whether exist in the image gathered
The picture of the tongue being available for subsequent analysis, if gather image in there is picture of the tongue, it is judged that picture of the tongue is in the position of image.
The present invention comprises the following steps:
1) the image A that input gathers under open environment;
2) the image A gathered is carried out color correction, obtain correction chart as B;
In step 2) in, the concrete grammar that the described image A to gathering carries out color correction is as follows:
(1) under standard photoenvironment, gather picture of the tongue image S1, calculate tri-Color Channel averages of RGB of picture of the tongue image S1
Respectively with overall average K of picture of the tongue image S1sRatio ccr、αg、αb:
Wherein, overall average K of picture of the tongue image S1s=(Ravgs+Gavgs+Bavgs)/3;Ravgs、Gavgs、BavgsIt is respectively mark
The average of tri-Color Channels of RGB of picture of the tongue image S1 is gathered under quasi-photoenvironment;It can be D65 that described standard illumination gathers environment
Light source, colour temperature is 6500K;
(2) adjust the average of tri-Color Channels of RGB of image A as the following formula, obtain correction chart as B;
K=(Ravg+Gavg+Bavg)/3
Wherein, K is the overall average of image A, Ravg、Gavg、BavgRespectively tri-Color Channels of RGB of image A is equal
Value;Rd、Gd、BdFor correction chart as the value of tri-Color Channels of RGB of each pixel of B, Rs、Gs、BsFor each pixel of image A
The value of tri-Color Channels of RGB, αr、αg、αbBy step 2) (1st) ratio partly tried to achieve;
3) to step 2) correction chart that obtains carries out image segmentation as B, and concrete grammar is as follows:
(1) correction image B is transformed into gray space image fB1, (sees document: OHTSU according to maximum variance between clusters
N.A threshold selection method from gray-level histograms[J].System Man&
Cybernetics IEEE Transactions on, 1979,9 (1): 62-66.) gray space image fB1 is carried out threshold value to divide
Cut, obtain splitting image B1 ', and use morphology operations to smooth connected domain segmentation image B1 ', obtain image B1;
In step 3) in (1st) part, described correction image B is transformed into gray space image fB1, according between maximum kind
Variance method carries out Threshold segmentation to gray space image fB1, obtains splitting image B1 ', and segmentation image B1 ' is used morphology
Computing smooths connected domain, obtains specifically comprising the following steps that of image B1
A) correction image B is transformed into gray space image fB1;
B) for gray space image fB1, if span G=[0, L-1] of gray scale G of gray space image fB1, respectively
The probability that gray value occurs is Pi, threshold value is T, and threshold value T is divided into f after gray space image fB1 is carried out binaryzation0And f1: f0=
[0, T], f1=[T+1, L-1], f0And f1Probability be respectivelyAnd α1=1-α0, average gray value is respectivelyWithThen f0And f1Maximum between-cluster variance be: g2(T)=α0(μ0-
μ)2+α1(μ1-μ)2=α0α1(μ0-μ1)2, wherein μ=Σ iPi, obtain threshold value T when g takes maximum, to gray space image
FB1 carries out Threshold segmentation, must split RGB value f of image B1 ' pixelX, y(r, g, b):
Wherein, (x, y) represents the value of gray space image fB1 pixel to fB1, and T is threshold value;
C) use the closed operation in morphology operations to smooth connected domain segmentation image B1 ', depend on according to following two formula
Secondary calculating, obtains image B1;Pixel value g1 in image B1 (x, y) be:
G1 (x, y)=erode (dilate (f1 (x, y), element))
G1 (x, y)=bitwise_not (g1 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B1 ', element is defined as the structural elements in morphology operations to f1
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;
Bitwise_not is defined as the inversion operation of each pixel to image.
(2) correction image B is transformed into hsv color spatial image fB2, the H passage of image fB2 is carried out Threshold segmentation,
Obtain splitting image B2 ', and use morphology operations to smooth connected domain segmentation image B2 ', obtain image B2;
In step 3) in (2nd) part, described correction image B is transformed into hsv color spatial image fB2, to image fB2
H passage carry out Threshold segmentation, obtain split image B2 ', and to segmentation image B2 ' use morphology operations smooth connected domain,
Obtain specifically comprising the following steps that of image B2
A) correction image B is transformed into hsv color spatial image fB2;
B) use following formula that image fB2 is carried out hue threshold segmentation, obtain splitting RGB value f of image B2 ' pixelX, y
(r, g, b):
Wherein, hX, yRepresent H passage pixel value in image fB2, T1And T2Represent the threshold value set;
C) use the closed operation in morphology operations to smooth connected domain segmentation image B2 ', depend on according to following two formula
Secondary calculating, obtains image B2;Pixel value g2 in image B2 (x, y) be:
G2 (x, y)=erode (dilate (f2 (x, y), element))
G2 (x, y)=bitwise_not (g2 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B2 ', element is defined as the structural elements in morphology operations to f2
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;
Bitwise_not is defined as the inversion operation of each pixel to image.
(3) use RGB trichroism component variance method (see document: Jiang Yiwu. computerized Evolution of Tongue Inspection of TCM system [J]. in China
Doctor trained in Western medicine combines magazine, and 2000,20 (2): 145-147) correction chart is carried out Threshold segmentation as B, obtain splitting image B3 ', and right
Segmentation image B3 ' uses morphology operations to smooth connected domain, obtains image B3.
In step 3) in (3rd) part, described employing RGB trichroism component variance method carries out Threshold segmentation to correction chart as B,
Obtain splitting image B3 ', and use morphology operations to smooth connected domain segmentation image B3 ', obtain image B3 concrete step
Rapid as follows:
A) for correction chart as B, it is assumed that its size is m × n, the RGB value of each pixel in correction image B is carried out
Normalization operates, and its span is [0,1], uses following formula to correction chart as each pixel in B calculates RGB tri-colouring component
(m, n), and splits as B variance gate to correction chart, obtains splitting RGB value f of image B3 ' pixelM, n(r, g, b):
Gate (m, n)=(rM, n-gM, n)+(bM, n-gM, n)×6+(rM, n+gM, n+bM, n)/3
Wherein, rM, nRepresent that correction chart is as pixel (m, the value of R passage n), g in BM, nRepresent that correction chart is as picture in B
Vegetarian refreshments (m, the value of G passage n), bM, nRepresent that correction chart is as pixel (m, the value of channel B n) in B;
B) use the closed operation in morphology operations to smooth connected domain segmentation image B3 ', calculate according to formula below,
To image B3;Pixel value g3 in image B3 (x, y) be:
G3 (x, y)=erode (dilate (f3 (x, y), element))
Wherein, (x, y) for the pixel value in segmentation image B3 ', element is defined as the structural elements in morphology operations to f3
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations.
(4) image B1, image B2, tri-images of image B3 are carried out logic "and" operation, obtain image C.
4) to step 3) the image C that obtains carries out provincial characteristics judgement, and concrete grammar is as follows:
(1) for each connected domain in image C, the convex closure (S of this connected domain is calculatedi);
(2) each convex closure (S is calculatedi) areaArea area with image CCRatio:Delete the area ratio connected domain less than 0.02;
(4) each connected domain convex closure (S is calculatedi) barycenter (Ci) and image C center (C0) Euclidean distance:
Wherein, Cix、CiyRepresent convex closure (S respectivelyi) abscissa of barycenter and vertical coordinate, C0xAnd C0yRepresent respectively in image C
The abscissa of the heart and vertical coordinate;
(4) each connected domain convex closure (S is calculatedi) length-width ratio scale=w/h of minimum enclosed rectangle;
Wherein w represents the width of minimum enclosed rectangle, and h represents the length of minimum enclosed rectangle;
(5) each connected domain convex closure (S is calculatedi) 7 Hu not bending moment mi(i ∈ [1,7] (sees document: Hu M.Visual
Pattern Recognition by Moment Invariants[J].Information Theory Ire
Transactions on,1962,8(2):179-187.);
Described Hu not bending moment is defined as follows:
For image f that size is M × N (x, y), f (x, two dimension (p+q) rank square y) is defined as:
(p+q) rank central moment is defined as accordingly:
Wherein
By ηpqThe normalization central moment represented is defined as:
Wherein γ=(p+q)/2+1
The linear combination of Hu normalization central moment constructs has translation, flexible, 7 Hu not bending moments of invariable rotary
(6) one normal shape picture of the tongue image (S of artificial selection0), calculate picture of the tongue image (S0) 7 Hu not bending moment Mi(i
∈ [1,7]), calculate matching degreeDescribed normal shape picture of the tongue image (S0) can be traditional Chinese medical science doctor
The raw normal shape picture of the tongue image rule of thumb selected with relevant professional knowledge;
(7) each connected domain convex closure (Si is calculated successively according to following three formula)Similarity score, select similarity
The maximum of score, the connected domain convex closure (S that the maximum of this similarity score is correspondingi) it is the candidate's tongue body region D selected:
Wherein, area is by step 4) (2nd) ratio partly tried to achieve;Scale is step 4) (4th) part tried to achieve
Length-width ratio;Dis is step 4) (3rd) Euclidean distance partly tried to achieve;Match is step 4) (6th) part tried to achieve
Matching degree;
5) to step 4) candidate's tongue body region D of obtaining carries out textural characteristics judgement, and concrete grammar is as follows:
(1) level of candidate's tongue body region D, vertical, the associating of three kinds of gray level co-occurrence matrixes of angular shift amount are calculated respectively
Probability density distribution;
(2) the joint probability density distribution setting gray level co-occurrence matrixes is designated as [Pmn] L × L, wherein L is gray scale span, m
=[0, L-1], n=[0, L-1];According to joint probability density distribution be calculated 6 textural characteristics (see document:
HARALICK R M,SHANMUGAM K,DINSTEIN I.Textural features for image
classification[J].Systems Man&Cybernetics IEEE Transactions on,2010,smc-3(6):
610 621.), described 6 textural characteristics include that angle second moment, contrast, entropy, unfavourable balance square, intermediate value are relevant with gray scale;
Wherein,
(3) to step 5) (1st) part obtain three kinds of gray level co-occurrence matrixes joint probability density distribution respectively according to
Step 5) (2nd) method 6 textural characteristics of calculating partly, there are 18 texture characteristic amount f of candidate's tongue body region Di(i
∈[0.17]);
(4) the picture of the tongue image gathered under artificial selection's some width standard photoenvironment, under each width standard photoenvironment
The picture of the tongue image gathered, is partitioned into tongue body region SD therein by hand, obtains several tongue body regions SD, to each tongue body district
Territory SD calculates its level, vertical, the joint probability density distribution of three kinds of gray level co-occurrence matrixes of angular shift amount respectively;To these three
Joint probability density is distributed respectively according to step 5) (2nd) method calculating textural characteristics partly, each tongue body region SD is altogether
Obtain 18 texture characteristic amounts;18 texture characteristic amounts calculated to these tongue body regions SD, seek each texture characteristic amount
Meansigma methods Fi(i ∈ [0,17]);
(5) calculation procedure 5) (3rd) the texture characteristic amount f partly tried to achievei(i ∈ [0,17]) and step 5) (4th) part
Meansigma methods F of the texture characteristic amount tried to achieveiThe texture similarity of (i ∈ [0,17]), uses following equation to calculate, if result E is little
In set threshold value T3, then be judged as picture of the tongue, the image A otherwise gathered is not available for analyze picture of the tongue:
The feature of Tongue image under open environment that the present invention is directed to carries out picture of the tongue target detection, first carries out image
The pretreatment of color correction, reduces the impact brought because of external light source colour temperature;Then image is split, obtain multiple connection
Region;And each connected region is carried out feature judgement, obtain candidate's tongue body region;Carry out finally by comparison domain textural characteristics
Judge, it is judged that whether this candidate's tongue body region is picture of the tongue.The present invention is finally reached and the picture of the tongue in image is carried out target detection
Purpose, i.e. judges either with or without picture of the tongue in photo current, if it has, picture of the tongue target is at which.
Color correction process uses the gray world algorithm improved, and sets corresponding for the distinctive color characteristic of picture of the tongue image
Parameter value;Picture of the tongue partitioning portion uses the segmentation of maximum between-cluster variance, hue threshold and the segmentation of RGB tri-colouring component difference to combine
Method carry out image segmentation;Provincial characteristics judgment part judges by contrasting the shape facility in each region, obtains tongue body
Region;The textural characteristics of picture of the tongue judges that then using gray level co-occurrence matrixes to be characterized detects, and is carried out the tongue body region obtained
Judge comparison, be finally reached the purpose of picture of the tongue target detection.
Relative to prior art, the invention have the benefit that
1, the gray world algorithm that the present invention improves according to picture of the tongue characteristic use carries out color correction to image, decrease because of
Gather the impact that environmental colors colour cast is brought.
2, the present invention uses the segmentation of maximum between-cluster variance, hue threshold and RGB tri-colouring component difference to split the side combined
Method carries out image segmentation;Provincial characteristics judgement is carried out according to picture of the tongue shape, position;Textural characteristics is used to detect;These sides
The picture of the tongue feature that method is both under open environment carries out processing, analyzing, and these methods complement each other, and being used in combination with can
More accurately identify picture of the tongue target, there is the advantage that recognition accuracy is high.
3, the method operand that the present invention uses is little, program realizes simple, it is not necessary to training mass data, algorithm complex
Low, processing speed is fast, the highest to the degree of dependence of data set.
The feature of picture of the tongue image under open environment that the present invention is directed to divides multiple step to carry out processing, analyzing, and reaches picture of the tongue mesh
The purpose of mark detection.The image segmentation of the present invention, provincial characteristics judge, textural characteristics judges the mutually the most auxiliary phase of several step
Becoming, image segmentation utilizes the color characteristic of image to carry out coarse sizing, and provincial characteristics judges to utilize the shape of image, space characteristics to enter
Row fine screening, textural characteristics is judged to utilize the textural characteristics of image finally to judge, is entered by the different characteristic combining image
Row contrast, the accuracy rate of identification is higher, fast operation, it is achieved simple, achieves new technique effect.Relative to immediate
Prior art, has obvious distinguishing characteristics, and has obvious advantage: the method for the present invention need not hough space and throws
Ticket, algorithm complex is low, and operand is little, it is not required that carries out the training mass data having supervision, therefore realizes relatively simple, right
The degree of dependence of data set is the highest, more superior in accuracy rate with processing speed, has more practicality.
The result obtained by the present invention, is conducive to improving follow-up inspection of the tongue and analyzes the serious forgiveness of system, robustness and system
Analysis speed, also allow for doctor of traditional Chinese medicine and observe picture of the tongue intuitively, accurately, easily, improve the speed of diagnosis.The present invention's is direct
Purpose is not to obtain disease or the diagnostic result of health status.Even if obtaining result by the present invention, figure also can only be obtained
Whether there is picture of the tongue, and the conclusion of the position of picture of the tongue in Xiang, diagnostic result can not be immediately arrived at, need follow-up inspection of the tongue to divide
Analysis system or the further of doctor of traditional Chinese medicine are analyzed and judge.Method the most disclosed by the invention, its direct purpose is not diagnosis,
The present invention is not belonging to the diagnostic method of disease.
Accompanying drawing explanation
Fig. 1 is the image gathered under open environment;
Fig. 2 is step 3) image segmentation result;
Fig. 3 is step 4) screen the candidate's tongue body region obtained.
Detailed description of the invention
With detailed description of the invention, technical solution of the present invention is described in further detail below in conjunction with the accompanying drawings:
The embodiment of the present invention comprises the steps:
1) the image A that input gathers under open environment, as shown in Figure 1;
2) the image A gathered is carried out color correction, obtain correction chart as B;
In step 2) in, the concrete grammar that the described image A to gathering carries out color correction is as follows:
(1) under standard photoenvironment, gather picture of the tongue image S1, calculate tri-Color Channel averages of RGB of picture of the tongue image S1
Respectively with overall average K of picture of the tongue image S1sRatio ccr、αg、αb:
Wherein, overall average K of picture of the tongue image S1s=(Ravgs+Gavgs+Bavgs)/3;Ravgs、Gavgs、BavgsIt is respectively mark
The average of tri-Color Channels of RGB of picture of the tongue image S1 is gathered under quasi-photoenvironment;It can be D65 that described standard illumination gathers environment
Light source, colour temperature is 6500K;
In the present embodiment, the value of parameters is as follows: αr=1.09, αg=0.95, αb=0.94.
(2) adjust the average of tri-Color Channels of RGB of image A as the following formula, obtain correction chart as B;
K=(Ravg+Gavg+Bavg)/3
Wherein, K is the overall average of image A, Ravg、Gavg、BavgRespectively tri-Color Channels of RGB of image A is equal
Value;Rd、Gd、BdFor correction chart as the value of tri-Color Channels of RGB of each pixel of B, Rs、Gs、BsFor each pixel of image A
The value of tri-Color Channels of RGB, αr、αg、αbBy step 2) (1st) ratio partly tried to achieve;
3) to step 2) correction chart that obtains carries out image segmentation as B, and concrete grammar is as follows:
(1) correction image B is transformed into gray space image fB1, according to maximum variance between clusters to gray space image
FB1 carries out Threshold segmentation, obtains splitting image B1 ', and uses morphology operations to smooth connected domain segmentation image B1 ', obtains
Image B1;
In step 3) in (1st) part, described correction image B is transformed into gray space image fB1, according between maximum kind
Variance method carries out Threshold segmentation to gray space image fB1, obtains splitting image B1 ', and segmentation image B1 ' is used morphology
Computing smooths connected domain, obtains specifically comprising the following steps that of image B1
A) correction image B is transformed into gray space image fB1;
B) for gray space image fB1, if span G=[0, L-1] of gray scale G of gray space image fB1, respectively
The probability that gray value occurs is Pi, threshold value is T, and threshold value T is divided into f after gray space image fB1 is carried out binaryzation0And f1: f0=
[0, T], f1=[T+1, L-1], f0And f1Probability be respectivelyAnd α1=1-α0, average gray value is respectivelyWithThen f0And f1Maximum between-cluster variance be: g2(T)=α0(μ0-
μ)2+α1(μ1-μ)2=α0α1(μ0-μ1)2, whereinObtain threshold value T when g takes maximum, to gray space figure
As fB1 carries out Threshold segmentation, RGB value f of image B1 ' pixel must be splitX, y(r, g, b):
Wherein, (x, y) represents the value of gray space image fB1 pixel to fB1, and T is threshold value;
In the present embodiment, T=90.
C) use the closed operation in morphology operations to smooth connected domain segmentation image B1 ', depend on according to following two formula
Secondary calculating, obtains image B1;Pixel value g1 in image B1 (x, y) be:
G1 (x, y)=erode (dilate (f1 (x, y), element))
G1 (x, y)=bitwise_not (g1 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B1 ', element is defined as the structural elements in morphology operations to f1
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;
Bitwise_not is defined as the inversion operation of each pixel to image.
In the present embodiment, element is defined as the oval structure of [11 × 11].
(2) correction image B is transformed into hsv color spatial image fB2, the H passage of image fB2 is carried out Threshold segmentation,
Obtain splitting image B2 ', and use morphology operations to smooth connected domain segmentation image B2 ', obtain image B2;
In step 3) in (2nd) part, described correction image B is transformed into hsv color spatial image fB2, to image fB2
H passage carry out Threshold segmentation, obtain split image B2 ', and to segmentation image B2 ' use morphology operations smooth connected domain,
Obtain specifically comprising the following steps that of image B2
A) correction image B is transformed into hsv color spatial image fB2;
B) use following formula that image fB2 is carried out hue threshold segmentation, obtain splitting RGB value f of image B2 ' pixelX, y
(r, g, b):
Wherein, hX, yRepresent H passage pixel value in image fB2, T1And T2Represent the threshold value set;
In the present embodiment, T1=7, T2=29.
C) use the closed operation in morphology operations to smooth connected domain segmentation image B2 ', depend on according to following two formula
Secondary calculating, obtains image B2;Pixel value g2 in image B2 (x, y) be:
G2 (x, y)=erode (dilate (f2 (x, y), element))
G2 (x, y)=bitwise_not (g2 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B2 ', element is defined as the structural elements in morphology operations to f2
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;
Bitwise_not is defined as the inversion operation of each pixel to image.
In the present embodiment, element is defined as the oval structure of [11 × 11]
(3) use RGB trichroism component variance method that as B, correction chart is carried out Threshold segmentation, obtain splitting image B3 ', and right
Segmentation image B3 ' uses morphology operations to smooth connected domain, obtains image B3.
In step 3) in (3rd) part, described employing RGB trichroism component variance method carries out Threshold segmentation to correction chart as B,
Obtain splitting image B3 ', and use morphology operations to smooth connected domain segmentation image B3 ', obtain image B3 concrete step
Rapid as follows:
A) for correction chart as B, it is assumed that its size is m × n, the RGB value of each pixel in correction image B is carried out
Normalization operates, and its span is [0,1], uses following formula to correction chart as each pixel in B calculates RGB tri-colouring component
(m, n), and splits as B variance gate to correction chart, obtains splitting RGB value f of image B3 ' pixelM, n(r, g, b):
Gate (m, n)=(rM, n-gM, n)+(bM, n-gM, n)x6+(rM, n+gM, n+bM, n)/3
Wherein, rM, nRepresent that correction chart is as pixel (m, the value of R passage n), g in BM, nRepresent that correction chart is as picture in B
Vegetarian refreshments (m, the value of G passage n), bM, nRepresent that correction chart is as pixel (m, the value of channel B n) in B;
B) use the closed operation in morphology operations to smooth connected domain segmentation image B3 ', calculate according to formula below,
To image B3;Pixel value g3 in image B3 (x, y) be:
G3 (x, y)=erode (dilate (f3 (x, y), element))
Wherein, (x, y) for the pixel value in segmentation image B3 ', element is defined as the structural elements in morphology operations to f3
Element;Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations.
In the present embodiment, element is defined as the oval structure of [11 × 11]
(4) image B1, image B2, tri-images of image B3 are carried out logic "and" operation, obtain image C, such as Fig. 2 institute
Show.
4) to step 3) the image C that obtains carries out provincial characteristics judgement, and concrete grammar is as follows:
(1) for each connected domain in image C, the convex closure (S of this connected domain is calculatedi);
(2) each convex closure (S is calculatedi) areaArea area with image CCRatio:Delete the area ratio connected domain less than 0.02;
(5) each connected domain convex closure (S is calculatedi) barycenter (Ci) and image C center (C0) Euclidean distance:
Wherein, Cix、CiyRepresent convex closure (S respectivelyi) abscissa of barycenter and vertical coordinate, C0xAnd C0yRepresent respectively in image C
The abscissa of the heart and vertical coordinate;
(4) each connected domain convex closure (S is calculatedi) length-width ratio scale=w/h of minimum enclosed rectangle;
Wherein w represents the width of minimum enclosed rectangle, and h represents the length of minimum enclosed rectangle;
(5) each connected domain convex closure (S is calculatedi) 7 Hu not bending moment mi(i ∈ [1,7]);
Described Hu not bending moment is defined as follows:
For image f that size is M × N (x, y), f (x, two dimension (p+q) rank square y) is defined as:
(p+q) rank central moment is defined as accordingly:
Wherein
By ηpqThe normalization central moment represented is defined as:
Wherein γ=(p+q)/2+1
The linear combination of Hu normalization central moment constructs has translation, flexible, 7 Hu not bending moments of invariable rotary:
(6) one normal shape picture of the tongue image (S of artificial selection0), calculate picture of the tongue image (S0) 7 Hu not bending moment Mi(i
∈ [1,7]), calculate matching degreeDescribed normal shape picture of the tongue image (S0) can be traditional Chinese medical science doctor
The raw normal shape picture of the tongue image rule of thumb selected with relevant professional knowledge;
In the present embodiment, MiThe value of (i ∈ [1,7]) be respectively as follows: [0.167538,0.00155933,0.00032405,
6.71027×10-6,-3.12169×10-10,-2.63866×10-7,-2.15107×10-11]
(7) each connected domain convex closure (S is calculated successively according to following three formulai) similarity score, select similarity
The maximum of score, the connected domain convex closure (S that the maximum of this similarity score is correspondingi) it is the candidate's tongue body region D selected,
As shown in Figure 3:
Wherein, area is step 4) (2nd) ratio partly tried to achieve;Scale is step 4) (4th) part tried to achieve
Length-width ratio;Dis is step 4) (3rd) Euclidean distance partly tried to achieve;Match is step 4) (6th) part tried to achieve
Degree of joining;
5) to step 4) candidate's tongue body region D of obtaining carries out textural characteristics judgement, and concrete grammar is as follows:
(1) level of candidate's tongue body region D, vertical, the associating of three kinds of gray level co-occurrence matrixes of angular shift amount are calculated respectively
Probability density distribution;
(2) the joint probability density distribution setting gray level co-occurrence matrixes is designated as [Pmn]L×L, wherein L is gray scale span, m
=[0, L-1], n=[0, L-1];It is calculated 6 textural characteristics, described 6 textural characteristics according to joint probability density distribution
Relevant with gray scale including angle second moment, contrast, entropy, unfavourable balance square, intermediate value;
Wherein,
(3) to step 5) (1st) part obtain three kinds of gray level co-occurrence matrixes joint probability density distribution respectively according to
Step 5) (2nd) method 6 textural characteristics of calculating partly, there are 18 texture characteristic amount f of candidate's tongue body region Di(i
∈ [0,17];
(4) the picture of the tongue image gathered under artificial selection's some width standard photoenvironment, under each width standard photoenvironment
The picture of the tongue image gathered, is partitioned into tongue body region SD therein by hand, obtains several tongue body regions SD;To each tongue body district
Territory SD calculates its level, vertical, the joint probability density distribution of three kinds of gray level co-occurrence matrixes of angular shift amount respectively;To these three
Joint probability density is distributed respectively according to step 5) (2nd) method calculating textural characteristics partly, each tongue body region SD is altogether
Obtain 18 texture characteristic amounts;18 texture characteristic amounts calculated to these tongue body regions SD, seek each texture characteristic amount
Meansigma methods Fi(i ∈ [0,17];
(5) calculation procedure 5) (3rd) the texture characteristic amount f partly tried to achievei(i ∈ [0,17]) and step 5) (4th) part
Texture characteristic amount meansigma methods F tried to achieveiThe texture similarity of (i ∈ [0,17]), uses following equation to calculate, if result E is less than
Set threshold value T3, then be judged as picture of the tongue, the image A otherwise gathered is not available for analyze picture of the tongue:
In the present embodiment, standard picture of the tongue characteristic vector meansigma methods Fi(value of i ∈ [0,17] is respectively as follows:
[0.00280036,59.4779,0.428998,2.9444,136.397,939.903,0.00197886,
68.4362,0.348336,3.0673,136.616,1336.41,0.00292386,38.6503,0.442026,2.9221,
136.406,901.906], the value of threshold value T3 is 4.7.
Claims (10)
1. Tongue object detection method under an open environment, it is characterised in that comprise the following steps:
1) the image A that input gathers under open environment;
2) the image A gathered is carried out color correction, obtain correction chart as B;
3) to step 2) correction chart that obtains carries out image segmentation as B, obtains image C;
4) to step 3) the image C that obtains carries out provincial characteristics judgement, obtains candidate's tongue body region D;
5) to step 4) candidate's tongue body region D of obtaining carries out textural characteristics judgement.
2. Tongue object detection method under a kind of open environment, it is characterised in that in step 2) in,
The concrete grammar that the described image A to gathering carries out color correction is as follows:
(1) under standard photoenvironment, gather picture of the tongue image S1, calculate tri-Color Channel averages of RGB of picture of the tongue image S1 respectively
Overall average K with picture of the tongue image S1sRatio ccr、αg、αb:
Wherein, overall average K of picture of the tongue image S1s=(Ravgs+Gavgs+Bavgs)/3;Ravgs、Gavgs、BavgsIt is respectively standard illumination
The average of tri-Color Channels of RGB of picture of the tongue image S1 is gathered under environment;
(2) adjust the average of tri-Color Channels of RGB of image A as the following formula, obtain correction chart as B;
K=(Ravg+Gavg+Bavg)/3
Wherein, K is the overall average of image A, Ravg、Gavg、BavgIt is respectively the average of tri-Color Channels of RGB of image A;Rd、
Gd、BdFor correction chart as the value of tri-Color Channels of RGB of each pixel of B, Rs、Gs、BsRGB for each pixel of image A
The value of three Color Channels, αr、αg、αbBy step 2) (1st) ratio partly tried to achieve.
3. Tongue object detection method under a kind of open environment, it is characterised in that in step 2) the
(1), in part, it is D65 light source that described standard illumination gathers environment, and colour temperature is 6500K.
4. Tongue object detection method under a kind of open environment, it is characterised in that in step 3) in,
Described to step 2) correction chart that the obtains concrete grammar that carries out image segmentation as B is as follows:
(1) correction image B is transformed into gray space image fB1, according to maximum variance between clusters, gray space image fB1 is entered
Row threshold division, obtains splitting image B1 ', and uses morphology operations to smooth connected domain segmentation image B1 ', obtain image
B1;
(2) correction image B is transformed into hsv color spatial image fB2, the H passage of image fB2 is carried out Threshold segmentation, obtains
Segmentation image B2 ', and use morphology operations to smooth connected domain segmentation image B2 ', obtain image B2;
(3) use RGB trichroism component variance method that as B, correction chart is carried out Threshold segmentation, obtain splitting image B3 ', and to segmentation
Image B3 ' uses morphology operations to smooth connected domain, obtains image B3;
(4) image B1, image B2, tri-images of image B3 are carried out logic "and" operation, obtain image C.
5. Tongue object detection method under a kind of open environment, it is characterised in that in step 3) the
(1), in part, described correction image B is transformed into gray space image fB1, according to maximum variance between clusters to gray space figure
As fB1 carries out Threshold segmentation, obtain splitting image B1 ', and use morphology operations to smooth connected domain segmentation image B1 ',
To specifically comprising the following steps that of image B1
A) correction image B is transformed into gray space image fB1;
B) for gray space image fB1, if span G=[0, L-1] of gray scale G of gray space image fB1, each gray scale
The probability that value occurs is Pi, threshold value is T, and threshold value T is divided into f after gray space image fB1 is carried out binaryzation0And f1: f0=[0,
T],f1=[T+1, L-1], f0And f1Probability be respectivelyAnd α1=1-α0, average gray value is respectivelyWithThen f0And f1Maximum between-cluster variance be: g2(T)=α0(μ0-
μ)2+α1(μ1-μ)2=α0α1(μ0-μ1)2, wherein μ=Σ iPi, obtain threshold value T when g takes maximum, to gray space image
FB1 carries out Threshold segmentation, must split RGB value f of image B1 ' pixelX, y(r, g, b):
Wherein, (x, y) represents the value of gray space image fB1 pixel to fB1, and T is threshold value;
C) use the closed operation in morphology operations to smooth connected domain segmentation image B1 ', count successively according to following two formula
Calculate, obtain image B1;Pixel value g1 in image B1 (x, y) be:
G1 (x, y)=erode (dilate (f1 (x, y), element))
G1 (x, y)=bitwise_not (g1 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B1 ', element is defined as the structural element in morphology operations to f1;
Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;bitwise_
Not is defined as the inversion operation of each pixel to image.
6. Tongue object detection method under a kind of open environment, it is characterised in that in step 3) the
(2) in part, described correction image B is transformed into hsv color spatial image fB2, the H passage of image fB2 is carried out threshold value and divides
Cut, obtain splitting image B2 ', and use morphology operations to smooth connected domain segmentation image B2 ', obtain the concrete step of image B2
Rapid as follows:
A) correction image B is transformed into hsv color spatial image fB2;
B) use following formula that image fB2 is carried out hue threshold segmentation, obtain splitting RGB value f of image B2 ' pixelX, y(r,
G, b):
Wherein, hX, yRepresent H passage pixel value in image fB2, T1And T2Represent the threshold value set;
C) use the closed operation in morphology operations to smooth connected domain segmentation image B2 ', count successively according to following two formula
Calculate, obtain image B2;Pixel value g2 in image B2 (x, y) be:
G2 (x, y)=erode (dilate (f2 (x, y), element))
G2 (x, y)=bitwise_not (g2 (x, y))
Wherein, (x, y) for the pixel value in segmentation image B2 ', element is defined as the structural element in morphology operations to f2;
Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations;bitwise_
Not is defined as the inversion operation of each pixel to image.
7. Tongue object detection method under a kind of open environment, it is characterised in that in step 3) the
(3) in part, described employing RGB trichroism component variance method carries out Threshold segmentation to correction chart as B, obtains splitting image B3 ', and
To segmentation image B3 ' use morphology operations smooth connected domain, obtain image B3 specifically comprise the following steps that
A) for correction chart as B, it is assumed that its size is m × n, the RGB value of each pixel in correction image B is carried out normalizing
Changing operation, its span is [0,1], uses following formula to correction chart as each pixel in B calculates the trichroism component variance of RGB
(m, n), and splits as B gate to correction chart, obtains splitting RGB value f of image B3 ' pixelM, n(r, g, b):
Gate (m, n)=(rM, n-gM, n)+(bM, n-gM, n)×6+(rM, n+gM, n+bM, n)/3
Wherein, rM, nRepresent that correction chart is as pixel (m, the value of R passage n), g in BM, nRepresent that correction chart is as pixel in B
(m, the value of G passage n), bM, nRepresent that correction chart is as pixel (m, the value of channel B n) in B;
B) use the closed operation in morphology operations to smooth connected domain segmentation image B3 ', calculate according to formula below, obtain figure
As B3;Pixel value g3 in image B3 (x, y) be:
G3 (x, y)=erode (dilate (f3 (x, y), element))
Wherein, (x, y) for the pixel value in segmentation image B3 ', element is defined as the structural element in morphology operations to f3;
Dilate is defined as the expansive working in morphology operations;Erode is defined as the etching operation in morphology operations.
8. Tongue object detection method under a kind of open environment, it is characterised in that in step 4) in,
Described to step 3) to carry out the concrete grammar of provincial characteristics judgement as follows for the image C that obtains:
(1) for each connected domain in image C, the convex closure (S of this connected domain is calculatedi);
(2) each convex closure (S is calculatedi) areaArea area with image CCRatio:Delete the area ratio connected domain less than 0.02;
(3) each connected domain convex closure (S is calculatedi) barycenter (Ci) and image C center (C0) Euclidean distance:
Wherein, Cix、CiyRepresent convex closure (S respectivelyi) abscissa of barycenter and vertical coordinate, C0xAnd C0yRepresent image C center respectively
Abscissa and vertical coordinate;
(4) each connected domain convex closure (S is calculatedi) length-width ratio scale=w/h of minimum enclosed rectangle;
Wherein w represents the width of minimum enclosed rectangle, and h represents the length of minimum enclosed rectangle;
(5) each connected domain convex closure (S is calculatedi) 7 Hu not bending moment mi(i ∈ [1,7]);
Described Hu not bending moment is defined as follows:
For image f that size is M × N (x, y), f (x, two dimension (p+q) rank square y) is defined as:
(p+q) rank central moment is defined as accordingly:
Wherein
By ηpqThe normalization central moment represented is defined as:
Wherein γ=(p+q)/2+1
The linear combination of Hu normalization central moment constructs has translation, flexible, 7 Hu not bending moments of invariable rotary
(6) one normal shape picture of the tongue image (S of artificial selection0), calculate picture of the tongue image (S0) 7 Hu not bending moment Mi(i ∈ [1,
7]), matching degree is calculated
(7) each connected domain convex closure (S is calculated successively according to following three formulai) similarity score, select similarity score
Maximum, the connected domain convex closure (S that the maximum of this similarity score is correspondingi) it is the candidate's tongue body region D selected:
Wherein, area is by step 4) (2nd) ratio partly tried to achieve;Scale is step 4) (4th) length partly tried to achieve
Wide ratio;Dis is step 4) (3rd) Euclidean distance partly tried to achieve;Match is step 4) (6th) coupling partly tried to achieve
Degree.
9. Tongue object detection method under a kind of open environment, it is characterised in that in step 4) the
(6) in part, described normal shape picture of the tongue image (S0) it is that doctor of traditional Chinese medicine is rule of thumb normal with what relevant professional knowledge selected
Shape picture of the tongue image.
10. Tongue object detection method under a kind of open environment, it is characterised in that in step 5)
In, described to step 4) to carry out the concrete grammar of textural characteristics judgement as follows for candidate's tongue body region D of obtaining:
(1) level of candidate's tongue body region D, vertical, the joint probability of three kinds of gray level co-occurrence matrixes of angular shift amount are calculated respectively
Density Distribution;
(2) the joint probability density distribution setting gray level co-occurrence matrixes is designated as [Pmn]L×L, wherein L is gray scale span, m=[0,
L-1], n=[0, L-1];Being calculated 6 textural characteristics according to joint probability density distribution, described 6 textural characteristics include angle
Second moment, contrast, entropy, unfavourable balance square, intermediate value are relevant with gray scale;
Wherein,
(3) to step 5) joint probability density of three kinds of gray level co-occurrence matrixes that obtains of (1st) part is distributed respectively according to step
5) method of (2nd) part calculates 6 textural characteristics, there are 18 texture characteristic amount f of candidate's tongue body region Di(i ∈ [0,
17]);
(4) the picture of the tongue image gathered under artificial selection's some width standard photoenvironment, gathers under each width standard photoenvironment
Picture of the tongue image, be by hand partitioned into tongue body region SD therein, obtain several tongue body regions SD, to each tongue body region SD
Calculate its level, vertical, the joint probability density distribution of three kinds of gray level co-occurrence matrixes of angular shift amount respectively;These three is combined
Probability density distribution is respectively according to step 5) (2nd) method calculating textural characteristics partly, each tongue body region SD there are
18 texture characteristic amounts;18 texture characteristic amounts calculated to these tongue body regions SD, seek the flat of each texture characteristic amount
Average fi(i ∈ [0,17]);
(5) calculation procedure 5) (3rd) the texture characteristic amount f partly tried to achievei(i ∈ [0,17]) and step 5) (4th) part tries to achieve
Meansigma methods F of texture characteristic amountiThe texture similarity of (i ∈ [0,17]), uses following equation to calculate, if result E is less than setting
Threshold value T3, then be judged as picture of the tongue, the image A otherwise gathered is not available for analyze picture of the tongue:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610300494.4A CN106023151B (en) | 2016-05-09 | 2016-05-09 | Tongue object detection method under a kind of open environment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610300494.4A CN106023151B (en) | 2016-05-09 | 2016-05-09 | Tongue object detection method under a kind of open environment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106023151A true CN106023151A (en) | 2016-10-12 |
CN106023151B CN106023151B (en) | 2019-02-01 |
Family
ID=57099693
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610300494.4A Active CN106023151B (en) | 2016-05-09 | 2016-05-09 | Tongue object detection method under a kind of open environment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106023151B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106845313A (en) * | 2016-12-28 | 2017-06-13 | 广州智慧城市发展研究院 | A kind of binary processing method of Quick Response Code |
CN107194937A (en) * | 2017-05-27 | 2017-09-22 | 厦门大学 | Tongue image partition method under a kind of open environment |
CN108303420A (en) * | 2017-12-30 | 2018-07-20 | 上饶市中科院云计算中心大数据研究院 | A kind of domestic type sperm quality detection method based on big data and mobile Internet |
CN108388874A (en) * | 2018-03-05 | 2018-08-10 | 厦门大学 | Prawn morphological parameters method for automatic measurement based on image recognition and cascade classifier |
CN108629780A (en) * | 2018-04-23 | 2018-10-09 | 闽江学院 | Tongue image dividing method based on color decomposition and threshold technology |
CN108830874A (en) * | 2018-04-19 | 2018-11-16 | 麦克奥迪(厦门)医疗诊断***有限公司 | A kind of number pathology full slice Image blank region automatic division method |
CN109636864A (en) * | 2018-12-19 | 2019-04-16 | 新绎健康科技有限公司 | A kind of tongue dividing method and system based on color correction Yu depth convolutional neural networks |
WO2020038312A1 (en) * | 2018-08-20 | 2020-02-27 | 深圳市前海安测信息技术有限公司 | Multi-channel tongue body edge detection device and method, and storage medium |
CN112489053A (en) * | 2020-11-26 | 2021-03-12 | 深圳市艾合芯科智慧医疗有限公司 | Tongue image segmentation method and device and storage medium |
CN113160203A (en) * | 2021-04-30 | 2021-07-23 | 湄洲湾职业技术学院 | Artificial intelligence tongue picture coating color recognition system |
CN114596621A (en) * | 2022-05-10 | 2022-06-07 | 慧医谷中医药科技(天津)股份有限公司 | Tongue picture data processing method and system based on machine vision |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104021566A (en) * | 2014-06-24 | 2014-09-03 | 天津大学 | GrabCut algorithm-based automatic segmentation method of tongue diagnosis image |
CN104933723A (en) * | 2015-07-21 | 2015-09-23 | 闽江学院 | Tongue image segmentation method based on sparse representation |
-
2016
- 2016-05-09 CN CN201610300494.4A patent/CN106023151B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104021566A (en) * | 2014-06-24 | 2014-09-03 | 天津大学 | GrabCut algorithm-based automatic segmentation method of tongue diagnosis image |
CN104933723A (en) * | 2015-07-21 | 2015-09-23 | 闽江学院 | Tongue image segmentation method based on sparse representation |
Non-Patent Citations (4)
Title |
---|
BOB ZHANG 等: "Tongue Color Analysis for Medical Application", 《EVIDENCE-BASED COMPLEMENTARY AND ALTERNATIVE MEDICINE》 * |
刘齐 等: "自然环境下舌诊图像偏色检测及其颜色校正方法", 《厦门大学学报(自然科学版)》 * |
许家佗 等: "基于差分统计方法的舌象纹理特征的分析与识别", 《上海中医药大学学报》 * |
陈碧黎: "复杂环境下中医舌诊关键技术研究", 《道客巴巴——HTTP://XUESHU.BAIDU.COM/S?WD=PAPERURI%3A%28B4F6E7D78DAABB6F1AA68B3191B62A46%29&FILTER=SC_LONG_SIGN&TN=SE_XUESHUSOURCE_2KDUW22V&SC_VURL=HTTP%3A%2F%2FWWW.DOC88.COM%2FP-3197552581670.HTML&IE=UTF-8&SC_US=3834659808226783687》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106845313A (en) * | 2016-12-28 | 2017-06-13 | 广州智慧城市发展研究院 | A kind of binary processing method of Quick Response Code |
CN107194937A (en) * | 2017-05-27 | 2017-09-22 | 厦门大学 | Tongue image partition method under a kind of open environment |
CN107194937B (en) * | 2017-05-27 | 2020-04-24 | 厦门大学 | Traditional Chinese medicine tongue picture image segmentation method in open environment |
CN108303420A (en) * | 2017-12-30 | 2018-07-20 | 上饶市中科院云计算中心大数据研究院 | A kind of domestic type sperm quality detection method based on big data and mobile Internet |
CN108388874B (en) * | 2018-03-05 | 2020-03-31 | 厦门大学 | Prawn morphological parameter automatic measurement method based on image recognition and cascade classifier |
CN108388874A (en) * | 2018-03-05 | 2018-08-10 | 厦门大学 | Prawn morphological parameters method for automatic measurement based on image recognition and cascade classifier |
CN108830874A (en) * | 2018-04-19 | 2018-11-16 | 麦克奥迪(厦门)医疗诊断***有限公司 | A kind of number pathology full slice Image blank region automatic division method |
CN108629780A (en) * | 2018-04-23 | 2018-10-09 | 闽江学院 | Tongue image dividing method based on color decomposition and threshold technology |
CN108629780B (en) * | 2018-04-23 | 2020-01-03 | 闽江学院 | Tongue image segmentation method based on color decomposition and threshold technology |
WO2020038312A1 (en) * | 2018-08-20 | 2020-02-27 | 深圳市前海安测信息技术有限公司 | Multi-channel tongue body edge detection device and method, and storage medium |
CN109636864A (en) * | 2018-12-19 | 2019-04-16 | 新绎健康科技有限公司 | A kind of tongue dividing method and system based on color correction Yu depth convolutional neural networks |
CN112489053A (en) * | 2020-11-26 | 2021-03-12 | 深圳市艾合芯科智慧医疗有限公司 | Tongue image segmentation method and device and storage medium |
CN112489053B (en) * | 2020-11-26 | 2021-07-13 | 深圳市艾合芯科智慧医疗有限公司 | Tongue image segmentation method and device and storage medium |
CN113160203A (en) * | 2021-04-30 | 2021-07-23 | 湄洲湾职业技术学院 | Artificial intelligence tongue picture coating color recognition system |
CN113160203B (en) * | 2021-04-30 | 2023-06-23 | 湄洲湾职业技术学院 | Artificial intelligent tongue picture and tongue fur color recognition system |
CN114596621A (en) * | 2022-05-10 | 2022-06-07 | 慧医谷中医药科技(天津)股份有限公司 | Tongue picture data processing method and system based on machine vision |
CN114596621B (en) * | 2022-05-10 | 2022-07-29 | 慧医谷中医药科技(天津)股份有限公司 | Tongue picture data processing method and system based on machine vision |
Also Published As
Publication number | Publication date |
---|---|
CN106023151B (en) | 2019-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106023151A (en) | Traditional Chinese medicine tongue manifestation object detection method in open environment | |
Adegun et al. | Deep learning-based system for automatic melanoma detection | |
CN109523535B (en) | Pretreatment method of lesion image | |
Krithika et al. | An individual grape leaf disease identification using leaf skeletons and KNN classification | |
CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
CN115345819A (en) | Gastric cancer image recognition system, device and application thereof | |
CN109145921A (en) | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster | |
CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
CN107194937A (en) | Tongue image partition method under a kind of open environment | |
CN108647625A (en) | A kind of expression recognition method and device | |
CN107292299B (en) | Side face recognition methods based on kernel specification correlation analysis | |
Singh et al. | Human pose estimation using convolutional neural networks | |
Casanova et al. | Texture analysis using fractal descriptors estimated by the mutual interference of color channels | |
CN109902585A (en) | A kind of three modality fusion recognition methods of finger based on graph model | |
Hu et al. | Estimation of tea leaf blight severity in natural scene images | |
CN109325408A (en) | A kind of gesture judging method and storage medium | |
CN109035227A (en) | The system that lung tumors detection and diagnosis is carried out to CT image | |
CN102567734A (en) | Specific value based retina thin blood vessel segmentation method | |
CN110021019A (en) | A kind of thickness distributional analysis method of the AI auxiliary hair of AGA clinical image | |
CN110032932A (en) | A kind of human posture recognition method based on video processing and decision tree given threshold | |
CN111062936B (en) | Quantitative index evaluation method for facial deformation diagnosis and treatment effect | |
CN111857334A (en) | Human body gesture letter recognition method and device, computer equipment and storage medium | |
Li et al. | Sublingual vein extraction algorithm based on hyperspectral tongue imaging technology | |
Jenifa et al. | Classification of cotton leaf disease using multi-support vector machine | |
CN106529441A (en) | Fuzzy boundary fragmentation-based depth motion map human body action recognition method |
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 |