CN109919929A - A kind of fissuring of tongue feature extracting method based on wavelet transformation - Google Patents
A kind of fissuring of tongue feature extracting method based on wavelet transformation Download PDFInfo
- Publication number
- CN109919929A CN109919929A CN201910169550.9A CN201910169550A CN109919929A CN 109919929 A CN109919929 A CN 109919929A CN 201910169550 A CN201910169550 A CN 201910169550A CN 109919929 A CN109919929 A CN 109919929A
- Authority
- CN
- China
- Prior art keywords
- image
- tongue
- pixel
- wavelet
- gradient
- 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
Abstract
The invention discloses a kind of fissuring of tongue feature extracting method based on wavelet transformation, comprising the following steps: median filtering is acted on into original tongue image, with the noise jamming factor in smooth original tongue image;Smoothed out tongue image is split using Canny edge detection operator;High and low frequency tongue image component map is obtained using wavelet decomposition to the tongue image after segmentation;High-low frequency weight figure is merged with Wavelet Fusion technology;Clearly fissuring of tongue feature is reconstructed by wavelet inverse transformation.Wavelet function is carried out decomposition and reconstruction by the present invention, and is applied in the processing to bifid tongue print image, including tongue image denoising;By the way that the processing technique of wavelet transformation is added, the feature extracting and matching of bifid tongue print image can be made more accurate.
Description
Technical field
The invention belongs to lunar physics technique fields, specifically, being related to a kind of fissuring of tongue feature extraction based on wavelet transformation
Method.
Background technique
Chinese medicine thinks that tongue picture variation is able to reflect the viscera lesions situation of patient body, by the examination to patient's tongue picture,
Doctor can have found patient a part dysfunction in time, play preferable booster action to the diagnosis of some diseases.In tradition
In Chinese medicine, doctor checks patient by " four methods of diagnosis ", with this come judge the qi and bloods of patient's internal organs, yin-yang physiology and
Pathological state." four methods of diagnosis " are divided into prestige, hear, ask, cutting four kinds of modes in tcm diagnosis, wherein lingual diagnosis is the important set of Chinese medicine diagnosis
At part, play an important role in clinical disease diagnosis.Lingual diagnosis is a kind of physiology and pathology form by observing tongue body
Come find human body cacosplanchnia reason physiological change examination method, doctor by tongue picture observation can make phase to the state of an illness of patient
The diagnosis and assessment answered have important value in traditional Chinese medicine.By long-term clinical practice, lingual diagnosis is as a result proved
Diseased region and the severity of disease can accurately be identified.Therefore, have for the research of lingual diagnosis this theory of medicine
Highly important meaning can provide important evidence for the diagnosis and treatment of clinical disease
Occur crackle of various shapes, dehiscence furrow on lingual surface, it is deep mixed, it is seen that in totaglossa, also seen in blade or
The tip of the tongue, tongue side etc..Main deficiency of YIN-blood, spleen deficiency are wet to invade.Crackle is a kind of disease occurred in Mucosa of lingual dorsum, and fissured tongue is distributed in
The whole surface of tongue is a kind of ill tongue, mainly based on tongue first half or the tip of the tongue both side edges.The research significance of tongue picture for
, there is very far-reaching influence in the problems such as human health.Crackle is a kind of disease occurred in Mucosa of lingual dorsum, and fissured tongue is distributed in tongue
Whole surface, be a kind of ill tongue, consultation rate is lower, mainly based on tongue first half or the tip of the tongue both side edges.It is clinical special
Point is exactly that Mucosa of lingual dorsum deep mixed, regular or irregular dehiscence furrow occurs.The reason of fissured tongue occurs generally speaking has at 4 points,
It is deficiency of Yin opinion, deficiency of blood opinion, deficiency of vital energy opinion and deficiency of yang opinion respectively.Common card type includes syndrome of yin deficiency of liver and kidney, stomach-Yin virtual loss is demonstrate,proved, liver blood loses
Asthenic symptoms, syndrome of deficiency of both qi and yin, syndrome of stagnation of liver qi and spleen deficiency, heart-fire hyperactivity card, syndrome of deficiency of spleen qi and stomach qi, syndrome of yang deficiency of spleen and kidney etc..
Bifid tongue print image has the reflectivity of cracks light extremely low, and crackle gray scale is far below background gray scale, and crackle
Locate the strong defect of grey scale change.Currently, the research method in relation to fissuring of tongue substantially can be divided into two major classes.Wherein, a kind of method
It is mainly based upon the gray scale of tongue image, color information and Threshold segmentation is carried out to fissuring of tongue, but do not consider fissuring of tongue feature, thus
Be difficult to it is accurate, be completely partitioned into fissuring of tongue.The another kind of line method for detecting that is mainly based upon is split fissuring of tongue, can be with
It is divided into 3 groups, i.e. the line method for detecting based on profile, the line method for detecting based on center line and the line detecting side based on region
Method.Although the method based on line detecting considers fissuring of tongue textural characteristics, all there is a degree of deficiency, such as: it is based on
The line method for detecting of profile is using first derivative, to noise-sensitive, and cannot often be closed in practical applications
Profile;Line method for detecting based on center line is generally adopted by second dervative, equally to noise-sensitive, and position of center line
Extraction often there is large error;And the line method for detecting based on region is often by open grain, the pseudo- crackle on tongue fur
It splits, needs to remove extra texture manually, can just be partitioned into crackle.
Summary of the invention
In view of this, the present invention provides a kind of fissuring of tongue feature extracting method based on wavelet transformation.
In order to solve the above-mentioned technical problem, the invention discloses a kind of fissuring of tongue feature extraction side based on wavelet transformation
Method, comprising the following steps:
S1, median filtering is acted on into original tongue image, with the noise jamming factor in smooth original tongue image;
S2, smoothed out tongue image is split using Canny edge detection operator;
S3, high and low frequency tongue image component map is obtained using wavelet decomposition to the tongue image after segmentation;
S4, high-low frequency weight figure is merged with Wavelet Fusion technology;
S5, clearly fissuring of tongue feature is reconstructed by wavelet inverse transformation.
Optionally, median filtering is acted on into original tongue image in the step S1 specifically:
With two-dimentional sleiding form, pixel in plate is ranked up according to the size of pixel value, generates monotone increasing or decline
It is 2-D data sequence;
Two dimension median filter exports
G (x, y)=med { f (x-k, y-l), (k, l ∈ W) } (1)
Wherein f (x, y), g (x, y) are respectively original image and treated image, and k, l are respectively the field of pixel x, y
Pixel;W is two dimension pattern plate, is 3 × 3 regions.
Optionally, smoothed out tongue image is split specifically using Canny edge detection operator in the step S2
Are as follows:
S2.1, noise is filtered out with smoothed image using Gaussian filter;
S2.2, the gradient intensity of each pixel and direction in image are calculated;
S2.3, using non-maxima suppression, to eliminate edge detection bring spurious response;
S2.4, true and potential edge is determined using dual threshold detection.
Optionally, noise is filtered out with smoothed image using Gaussian filter in the step S2.1 specifically: in order to flat
Sliding image carries out convolution using Gaussian filter and image, and size is the generation side of the Gaussian filter core of (2k+1) x (2k+1)
Formula is given by:
Wherein, σ is variance, and k is the dimension of determining nuclear matrix, and i, j are the pixel values of image.
Optionally, the step S2.2 falls into a trap the gradient intensity of each pixel and direction in nomogram picture specifically:
Edge in image can be directed toward all directions, therefore Canny algorithm is come in detection image using Sobel operator
Horizontal, vertical and diagonal edge;The operator of edge detection returns to horizontal GxWith vertical GyThe first derivative values in direction, thus
To determine the gradient G and direction theta of pixel;
Sobel warp factor are as follows:
The operator includes two group 3 × 3 of matrix, respectively horizontal direction and vertical direction, it is made plane volume with image
Product, obtains the brightness difference approximation of horizontal direction and vertical direction respectively;Original image, G are represented with IxAnd GyRespectively represent water
Square to and vertical direction edge detection gray value of image, formula is as follows:
The horizontal direction of each of image pixel and vertical gray value are combined by following formula, to calculate point ash
The size of degree:
Gradient direction is calculated with following formula:
θ=arc tan (Gy/Gx) (6)
Wherein, G is gradient intensity, and θ indicates that gradient direction, theta indicate gradient direction, and arctan is arctan function;
Sobel operator according to above and below pixel, left and right neck point intensity-weighted it is poor, edge reach this phenomenon of extreme value inspection
Survey edge.
Optionally, non-maxima suppression is applied in the step S2.3, to eliminate edge detection bring spurious response tool
Body are as follows:
During non-maxima suppression, image is handled using 3 × 3 moving window, center pixel gradient value
It is compared with other pixel gradient values in field, if center pixel value is not the maximum of field pixel, the picture
Vegetarian refreshments is assigned a value of 0, edge that the is on the contrary then pixel being considered as image;Non- maximum suppression is carried out to pixel each in gradient image
The step of processed, is:
A) by one (0,45,90,135,180,225,270,315) that its gradient direction is approximately in following values, i.e., on
Lower left and right and 45 ° of directions;
B) compare the gradient intensity of the pixel of the pixel and the positive negative direction of its gradient direction;
C) retain if the pixel gradient intensity maximum, otherwise inhibit, delete, that is, be set to 0;
The pixel gradient to be compared is obtained using linear interpolation between two adjacent pixels across gradient direction;Its
Shown in specific mathematic(al) representation such as formula (7):
Wherein, i, j indicate the pixel of image, the maximum pixel gradient of maxgrade representative image.
Optionally, high and low frequency tongue image point is obtained using wavelet decomposition to the tongue image after segmentation in the step S3
Spirogram, specifically:
Using Daubechies-4 type small echo, i layers of wavelet decomposition are carried out to fingerprint image;
Piece image signal is square image, is divided into the two identical regions in left and right, and wherein left area is L, right side
Region is H;L is low frequency, and H is high frequency;
The picture signal is subjected to i wavelet decomposition, obtains one group of wavelet coefficient, size and shape with original image phase
Together;Wherein, upper left side region is decomposed into 4 each regions, wherein upper left side region is LLi, bottom-left quadrant LHi;Upper right side area
Domain is HLi, and lower right region is HHi;Remaining region is identical as one layer of decomposition.
Optionally, high-low frequency weight figure is merged with Wavelet Fusion technology in the step S4 specifically:
Using local variance as foundation in low frequency;Assuming that c (X) indicates the wavelet low frequency component of bifid tongue print image X
Coefficient matrix;The spatial position of p (m, n) expression wavelet coefficient;Then be designated as under c (X, p) expression low frequency component coefficient matrix (m,
N) value of element;Firstly, indicating Local Deviation conspicuousness with the weighted variance in the Q of region centered on p;U (X, p) is indicated
The mean value of bifid tongue print image X low frequency coefficient matrix, p point are the center in the region Q;If G (X, p) represents low in bifid tongue print image X
The Local Deviation conspicuousness of frequency coefficient matrix take p point as the center in the region Q, then:
G (X, p)=∑P∈Qω (p) | c (X, p)-u (X, p) |2 (8)
ω (p) indicates weight, when it is bigger closer to p point duration;The Local Deviation of the low frequency coefficient matrix of image A and B
It is expressed as G (A, p) and G (B, p);In addition, M of the Local Deviation matching degree of the low frequency coefficient matrix of image A and B by point p2
(p) it defines:
M2(p) value changes between 0~1, and value is smaller, and the matching of the low frequency coefficient matrix of two images is by lower;
If T2It is the threshold value of matching degree, common value is 0.5-1;
Work as M2(p) < T2When, select convergence strategy as follows:
Work as M2(p)≥T2When, average convergence strategy is as follows:
Wherein,
In the high frequency section of wavelet transformation, the maximum value of wavelet coefficient absolute value is selected, information between low-and high-frequency is made up
Certain part;Since the noise and defect of crackle target are all high-frequency informations, the median filtering is used to fused tongue
The high frequency coefficient of crack image is filtered, to remove the noise and defect of bifid tongue print image:
D (X, p) indicates the coefficient matrix in the small echo high fdrequency component of P point.
Compared with prior art, the present invention can be obtained including following technical effect:
1) bifid tongue print image has following feature: the reflectivity of cracks light is extremely low, and crackle gray scale is far below background gray scale,
And cracks grey scale change is strong.Therefore, the present invention selects median filtering first to show to smooth by image, reduces picture noise
Retain fracture edges information simultaneously, then reinforces slit region with frequency domain filtering.This method is divided into five parts, by median filtering
Act on original tongue image, in order to the disturbing factors such as noise in smooth original tongue image;Utilize Canny edge detection
Operator is split smoothed out tongue image;High and low frequency tongue image is obtained using wavelet decomposition to the tongue image after segmentation
Component map;High-low frequency weight figure is merged with Wavelet Fusion technology, reconstructs clearly tongue finally by wavelet inverse transformation
Crack.The present invention is directed to improve the discrimination of fissuring of tongue detection.Wavelet function is subjected to decomposition and reconstruction, and is used
Into the processing to bifid tongue print image, including tongue image denoising.By the way that the processing technique of wavelet transformation is added, fissuring of tongue figure can be made
The feature extracting and matching of picture is more accurate.
2) compared with prior art, wavelet coefficient absolute value described in technical solution proposed by the present invention is bigger, indicates
Original image grey scale change is bigger, and importance is higher.Irregular structural body can be converted into after wavelet transform as crackle
High-amplitude value coefficient in high-frequency sub-band calculates the quantity that those amplitudes in high-frequency sub-band are greater than the coefficient of some threshold value, as splitting
A kind of feature of print image.
3) 32 bifid tongue print images that the present invention is provided using Beijing Dongzhimen hospital, wherein 2 fissuring of tongue recognition failures,
Accuracy rate is up to 93.75%.Wavelet function feedback technical solution solves some defects present in conventional segmentation methods point
Cut the problem of imperfect, over-segmentation.The present invention can correctly detect bifid tongue print image, have certain anti-interference ability, solve
Information loses and the problem of over-segmentation.
Certainly, it implements any of the products of the present invention it is not absolutely required to while reaching all the above technical effect.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes a part of the invention, this hair
Bright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is fissuring of tongue identification process figure of the present invention;
Fig. 2 is wavelet decomposition of the present invention;Wherein, the low frequency and high fdrequency component of a representative image;The small echo one of b representative image
Grade is decomposed;The small echo second level of c representative image is decomposed.
Specific embodiment
Carry out the embodiment that the present invention will be described in detail below in conjunction with embodiment, whereby to the present invention how application technology hand
Section solves technical problem and reaches the realization process of technical effect to fully understand and implement.
The invention discloses a kind of fissuring of tongue feature extracting method based on wavelet transformation, comprising the following steps:
S1, median filtering is acted on into original tongue image, with disturbing factors such as noises in smooth original tongue image;
The median filtering method is a kind of nonlinear images enhancing technology, is the filtering method based on sequencing statistical theory.
Median filtering can significantly inhibit noise, it sets the gray value of pixel each in image in the point neighborhood window
The intermediate value of all pixels point gray value.The basic principle of median filtering is that the value of any in digital picture or Serial No. is used to be somebody's turn to do
The intermediate value of each point value replaces in one neighborhood of point, the true value for making the pixel value of surrounding close, to eliminate isolated noise
Point.Method is to use two-dimentional sleiding form, and pixel in plate is ranked up according to the size of pixel value, generates monotone increasing (under or
Drop) it is 2-D data sequence.
Two dimension median filter exports
G (x, y)=med { f (x-k, y-l), (k, l ∈ W) } (1)
Wherein f (x, y), g (x, y) are respectively original image and treated image, k, 1 be respectively pixel x, y field
Pixel.W is two dimension pattern plate, usually 3 × 3,5 × 5 region, is also possible to different shapes, such as linear, round, cross,
Circular ring shape etc..The present invention uses 3 × 3 regions.
S2, smoothed out tongue image is split using Canny edge detection operator;
The edge the Canny segmentation, edge is the boundary between object and background, moreover it is possible to indicate the side between overlapping object
Boundary.It is compared with other edge detection algorithms, canny detection can effectively inhibit picture noise and accurate determining image side
The position of edge.Canny edge detection can be divided into four steps:
S2.1, noise is filtered out with smoothed image using Gaussian filter.
In order to reduce influence of the noise to edge detection results as far as possible, so having to filter out noise to prevent from being drawn by noise
The error detection risen.For smoothed image, convolution is carried out using Gaussian filter and image, the step is by smoothed image, to subtract
Apparent influence of noise on few edge detector.Size be (2k+1) x (2k+1) Gaussian filter core growth equation formula by
It is given below:
Wherein, σ is variance, and k is the dimension of determining nuclear matrix, and i, j are the pixel values of image;
S2.2, the gradient intensity of each pixel and direction in image are calculated.
Edge in image can be directed toward all directions, therefore Canny algorithm is come in detection image using four operators
Horizontal, vertical and diagonal edge.The operator (such as Roberts, Prewitt, Sobel etc.) of edge detection returns to horizontal GxWith it is vertical
GyThus the first derivative values in direction can determine the gradient G and direction theta of pixel.Present invention selection uses Sobel
Operator, because this is one of the most common type operator.
Sobel operator is mainly used as edge detection, and technically, it is a kind of discreteness difference operator.Sobel convolution because
Son are as follows:
The operator includes two group 3 × 3 of matrix, respectively horizontal direction and vertical direction, it is made plane volume with image
Product, can obtain the brightness difference approximation of horizontal direction and vertical direction respectively.Original image, G are represented with IxAnd GyGeneration respectively
The gray value of image of table horizontal direction and vertical direction edge detection, formula are as follows:
The horizontal direction of each of image pixel and vertical gray value can be combined by following formula, to calculate this
The size of point gray scale:
Gradient direction can be calculated with following formula:
θ=arc tan (Gy/Gx) (6)
Wherein G is gradient intensity, and θ indicates that gradient direction, theta indicate gradient direction, and arctan is arctan function.
Sobel operator according to above and below pixel, left and right neck point intensity-weighted it is poor, edge reach this phenomenon of extreme value inspection
Survey edge.There is smoothing effect to noise, more accurate edge directional information is provided.
S2.3, inhibit using non-maximum (Non-Maximum Suppression), it is miscellaneous to eliminate edge detection bring
Dissipate response:
After carrying out gradient calculating to image, the edge for being based only on gradient value extraction is still very fuzzy.Rather than maximum presses down
System can then help all gradient values except local maximum being suppressed to 0.During non-maxima suppression, 3 × 3 are used
Moving window image is handled, center pixel gradient value is compared with other pixel gradient values in field, if
Center pixel value is not the maximum of field pixel, then the pixel is assigned a value of 0, on the contrary then the pixel is considered as image
Edge.The step of carrying out non-maxima suppression to pixel each in gradient image is:
It a) is approximately one (0,45,90,135,180,225,270,315) in following values (on i.e. by its gradient direction
Lower left and right and 45 ° of directions);
B) compare the gradient intensity of the pixel of the pixel and the positive negative direction of its gradient direction;
C) retain if the pixel gradient intensity maximum, otherwise inhibit (to delete, that is, be set to 0);
Generally for more accurate calculating, come between two adjacent pixels across gradient direction using linear interpolation
Obtain the pixel gradient to be compared.Its specific mathematic(al) representation such as formula
Wherein, i, j indicate the pixel of image, the maximum pixel gradient of maxgrade representative image.
Non-maxima suppression had not only effectively remained the gradient of image border, but also had achieved the purpose that image thinning.
S2.4, it detects using dual threshold (Double-Threshold) to determine true and potential edge.
Image still has many noise spots after non-maxima suppression, and canny algorithm applies a kind of technology of dual threshold,
It setting a threshold value upper bound and threshold value is next, the pixel in image then thinks necessarily boundary if it is greater than the threshold value upper bound,
Then thinking inevitable less than threshold value is not boundary.
For the edge canny segmented image, the actual situation marginal information of target defect is contained in high frequency imaging, while also wrapping
Some noises in bifid tongue print image are contained.Low-frequency image includes the profile information of target defect.
S3, high and low frequency tongue image component map is obtained using wavelet decomposition to the tongue image after segmentation;
The wavelet transformation, be it is a kind of by picture breakdown be different frequency domains.Then in different frequency domains using different
Fusion rule, finally, image is reconstructed using wavelet inverse transformation.Fissuring of tongue picture breakdown can be average by wavelet transformation
The combination of image and detail pictures, this respectively represents the different structure of image.Therefore it is easy to extract the structure letter of original image
Breath and detailed information.Wavelet transformation is built upon on the basis of Fourier analysis, utilizes the multiresolution point of wavelet transformation
The characteristics of analysis, can characterize the local features of signal in time-domain and frequency-domain, the variable spy of and shape constant according to window size
Point uses the higher resolution ratio of frequency in the low frequency part of picture signal, and higher same using temporal resolution in high frequency section
When the lower method of frequency resolution, be used in the pretreatment stage of fissuring of tongue identification, bifid tongue that can be irregular to signal
Line is handled.
For two-dimensional wavelet transformation, it obtain after handling as two continuous one-dimensional wavelet transforms
's.The processing that image is carried out by two-dimensional wavelet transformation, can be broken down into a series of low frequency subgraph pictures, and result depends on small
The type of wave base is decided by the type of filter, the present invention uses widely used Daubechies-4 type small echo, to fingerprint
Image carries out 2 layers of wavelet decomposition.
Piece image signal is carried out wavelet decomposition by the present invention, can obtain one group of wavelet coefficient, size and shape with
Original image is identical.The image of one pair 300 × 300 pass through two layers of wavelet decomposition, obtain 7 pieces of wavelet decompositions as shown in Figure 2 as a result,
One shares 90000 coefficients.
The picture signal is subjected to i wavelet decomposition, obtains one group of wavelet coefficient, size and shape with original image phase
Together;Wherein, upper left side region is decomposed into 4 each regions, wherein upper left side region is LLi, bottom-left quadrant LHi;Upper right side area
Domain is HLi, and lower right region is HHi;Remaining region is identical as one layer of decomposition;L is low frequency, and H is high frequency;Subscript 1 and 2 indicates one
Secondary or second decomposition.On each decomposition layer, image is broken down into tetra- wave bands of LL, LH, HH, HL.Only decompose next layer low
Frequency component.Each of four subgraphs are generated by the inner product of original image and a wavelet basis function.Then in X
It is sampled 2 times with Y-direction.Inverse transformation, i.e. image reconstruction are sample frequency and convolution by increasing image to realize.From place
The data managed can exceed that 255 or negative occur, it needs to normalize to 0-255 and shows image.
S4, high-low frequency weight figure is merged with Wavelet Fusion technology:
The Wavelet Fusion, the wavelet coefficient of low frequency include the profile information of image, it is therefore desirable to select suitable small echo
Coefficient carries out mixing operation.The present invention is in low frequency using local variance as foundation.Assuming that c (X) indicates bifid tongue print image X's
The coefficient matrix of wavelet low frequency component.The spatial position of p (m, n) expression wavelet coefficient.Then c (X, p) indicates low frequency component system
The value of the element of (m, n) is designated as under matrix number.Firstly, indicating that Local Deviation is aobvious with the weighted variance in the Q of region centered on p
Work property.U (X, p) indicates the mean value of bifid tongue print image X low frequency coefficient matrix, and p point is the center in the region Q.If G (X, p) is represented
The Local Deviation conspicuousness of low frequency coefficient matrix in bifid tongue print image X take p point as the center in the region Q, then:
G (X, p)=∑P∈Qω (p) | c (X, p)-u (X, p) |2 (8)
ω (p) indicates weight, when it is bigger closer to p point duration.The Local Deviation of the low frequency coefficient matrix of image A and B
It is expressed as G (A, p) and G (B, p).In addition, M of the Local Deviation matching degree of the low frequency coefficient matrix of image A and B by point p2
(p) it defines:
M2(p) value changes between 0~1, and value is smaller, and the matching degree of the low frequency coefficient matrix of two images is lower.
If T2It is the threshold value of matching degree, common value is (0.5-1).
Work as M2(p) < T2When, select (optimal) convergence strategy as follows:
Work as M2(p)≥T2When, average convergence strategy is as follows:
Wherein
Above strategy is that this correlation, which can effectively retain, to be based on based on the correlation between the pixel of field
The details and edge of canny detection.
The wavelet coefficient absolute value is bigger, indicates that original image grey scale change is bigger, importance is higher.As crackle this
The irregular structural body of sample can be converted into the high-amplitude value coefficient in high-frequency sub-band after wavelet transform, calculate those in high-frequency sub-band
Amplitude is greater than the quantity of the coefficient of some threshold value, a kind of feature as crack image.
After the segmentation of the accurate edge canny, the contrast of image is stronger.Radio-frequency head of the present invention in wavelet transformation
Point, the maximum value of wavelet coefficient absolute value is selected, can making up information between low-and high-frequency, real part divides really.Due to crackle target
Noise and defect are all high-frequency informations, thus the median filtering be used to the high frequency coefficient of fused bifid tongue print image into
Row filtering, to remove the noise and defect of bifid tongue print image.
D (X, p) indicates the coefficient matrix in the small echo high fdrequency component of P point.
S5, clearly fissuring of tongue feature is reconstructed by wavelet inverse transformation.
The wavelet inverse transformation reconstructs clearly fissuring of tongue figure by the high and low frequency component that acts on that treated
Picture.
32 bifid tongue print images that the present invention uses Beijing Dongzhimen hospital to provide, wherein 2 fissuring of tongue recognition failures, quasi-
True rate is up to 93.75%.Wavelet function feedback technical solution solves some defect Segmentations present in conventional segmentation methods
The problem of imperfect, over-segmentation.The present invention can correctly detect bifid tongue print image, have certain anti-interference ability, solve
The problem of information loss and over-segmentation.
Above description has shown and described several preferred embodiments of invention, but as previously described, it should be understood that invention is not
It is confined to form disclosed herein, should not be regarded as an exclusion of other examples, and can be used for various other combinations, modification
And environment, and can be carried out within that scope of the inventive concept describe herein by the above teachings or related fields of technology or knowledge
Change.And changes and modifications made by those skilled in the art do not depart from the spirit and scope of invention, then it all should be in the appended power of invention
In the protection scope that benefit requires.
Claims (8)
1. a kind of fissuring of tongue feature extracting method based on wavelet transformation, which comprises the following steps:
S1, median filtering is acted on into original tongue image, with the noise jamming factor in smooth original tongue image;
S2, smoothed out tongue image is split using Canny edge detection operator;
S3, high and low frequency tongue image component map is obtained using wavelet decomposition to the tongue image after segmentation;
S4, high-low frequency weight figure is merged with Wavelet Fusion technology;
S5, clearly fissuring of tongue feature is reconstructed by wavelet inverse transformation.
2. the method according to claim 1, wherein median filtering is acted on original tongue figure in the step S1
Picture specifically:
With two-dimentional sleiding form, pixel in plate is ranked up according to the size of pixel value, generates monotone increasing or decline is
2-D data sequence;
Two dimension median filter exports
G (x, y)=med { f (x-k, y-l), (k, l ∈ W) } (1)
Wherein f (x, y), g (x, y) are respectively original image and treated image, and k, l are respectively the field pixel of pixel x, y;
W is two dimension pattern plate, is 3 × 3 regions.
3. the method according to claim 1, wherein utilizing Canny edge detection operator pair in the step S2
Smoothed out tongue image is split specifically:
S2.1, noise is filtered out with smoothed image using Gaussian filter;
S2.2, the gradient intensity of each pixel and direction in image are calculated;
S2.3, using non-maxima suppression, to eliminate edge detection bring spurious response;
S2.4, true and potential edge is determined using dual threshold detection.
4. according to the method described in claim 3, it is characterized in that, Gaussian filter is used in the step S2.1, with smooth
Image filters out noise specifically: for smoothed image, carries out convolution using Gaussian filter and image, size is (2k+1) x
The growth equation formula of the Gaussian filter core of (2k+1) is given by:
Wherein, σ is variance, and k is the dimension of determining nuclear matrix, and i, j are the pixel values of image.
5. according to the method described in claim 3, nomogram each pixel as in it is characterized in that, the step S2.2 falls into a trap
Gradient intensity and direction specifically:
Edge in image can be directed toward all directions, therefore Canny algorithm carrys out the water in detection image using Sobel operator
Flat, vertical and diagonal edge;The operator of edge detection returns to horizontal GxWith vertical GyThe first derivative values in direction, thus can
Determine the gradient G and direction theta of pixel;
Sobel warp factor are as follows:
The operator includes two group 3 × 3 of matrix, respectively horizontal direction and vertical direction, it is made planar convolution with image, point
The brightness difference approximation of horizontal direction and vertical direction is not obtained;Original image, G are represented with IxAnd GyRespectively represent level side
To and vertical direction edge detection gray value of image, formula is as follows:
The horizontal direction of each of image pixel and vertical gray value are combined by following formula, to calculate the gray scale
Size:
Gradient direction is calculated with following formula:
θ=arc tan (Gy/Gx) (6)
Wherein, G is gradient intensity, and θ indicates that gradient direction, theta indicate gradient direction, and arctan is arctan function;
Sobel operator according to above and below pixel, left and right neck point intensity-weighted it is poor, edge reach this phenomenon of extreme value detection side
Edge.
6. according to the method described in claim 3, it is characterized in that, non-maxima suppression is applied in the step S2.3, to disappear
Except edge detection bring spurious response specifically:
During non-maxima suppression, image is handled using 3 × 3 moving window, center pixel gradient value and neck
Other pixel gradient values in domain are compared, if center pixel value is not the maximum of field pixel, the pixel
It is assigned a value of 0, edge that the is on the contrary then pixel being considered as image;Non-maxima suppression is carried out to pixel each in gradient image
Step is:
A) by one (0,45,90,135,180,225,270,315) that its gradient direction is approximately in following values, i.e., left up and down
Right and 45 ° of directions;
B) compare the gradient intensity of the pixel of the pixel and the positive negative direction of its gradient direction;
C) retain if the pixel gradient intensity maximum, otherwise inhibit, delete, that is, be set to 0;
The pixel gradient to be compared is obtained using linear interpolation between two adjacent pixels across gradient direction;It is specific
Mathematic(al) representation such as formula (7) shown in:
Wherein, i, j indicate the pixel of image, the maximum pixel gradient of maxgrade representative image.
7. the method according to claim 1, wherein utilizing small echo to the tongue image after segmentation in the step S3
Decomposition obtains high and low frequency tongue image component map, specifically:
Using Daubechies-4 type small echo, i layers of wavelet decomposition are carried out to fingerprint image;
Piece image signal is square image, is divided into the two identical regions in left and right, and wherein left area is L, right area
For H;L is low frequency, and H is high frequency;
The picture signal is subjected to i wavelet decomposition, obtains one group of wavelet coefficient, size and shape is identical as original image;
Wherein, upper left side region is decomposed into 4 each regions, wherein upper left side region is LLi, bottom-left quadrant LHi;Upper right side region
For HLi, lower right region is HHi;Remaining region is identical as one layer of decomposition.
8. the method according to claim 1, wherein by high-low frequency weight figure Wavelet Fusion in the step S4
Technology is merged specifically:
Using local variance as foundation in low frequency;Assuming that c (X) indicates the coefficient of the wavelet low frequency component of bifid tongue print image X
Matrix;The spatial position of p (m, n) expression wavelet coefficient;Then c (X, p) indicates to be designated as (m, n) under low frequency component coefficient matrix
The value of element;Firstly, indicating Local Deviation conspicuousness with the weighted variance in the Q of region centered on p;U (X, p) indicates bifid tongue
The mean value of print image X low frequency coefficient matrix, p point are the center in the region Q;If G (X, p) represents low frequency system in bifid tongue print image X
The Local Deviation conspicuousness of matrix number take p point as the center in the region Q, then:
G (X, p)=∑p∈Qω(p)]c(X,p)-u(X,p)]2 (8)
ω (p) indicates weight, when it is bigger closer to p point duration;The Local Deviation of the low frequency coefficient matrix of image A and B indicates
For G (A, p) and G (B, p);In addition, M of the Local Deviation matching degree of the low frequency coefficient matrix of image A and B by point p2(p) fixed
Justice:
M2(p) value changes between 0~1, and value is smaller, and the matching degree of the low frequency coefficient matrix of two images is lower;
If T2It is the threshold value of matching degree, common value is 0.5-1;
Work as M2(p)<T2When, select convergence strategy as follows:
Work as M2(p)≥T2When, average convergence strategy is as follows:
Wherein,Wmax=1-Wmin;
In the high frequency section of wavelet transformation, the maximum value of wavelet coefficient absolute value is selected, makes up the certain of information between low-and high-frequency
Part;Since the noise and defect of crackle target are all high-frequency informations, the median filtering is used to fused fissuring of tongue
The high frequency coefficient of image is filtered, to remove the noise and defect of bifid tongue print image:
D (X, p) indicates the coefficient matrix in the small echo high fdrequency component of P point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910169550.9A CN109919929B (en) | 2019-03-06 | 2019-03-06 | Tongue crack feature extraction method based on wavelet transformation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910169550.9A CN109919929B (en) | 2019-03-06 | 2019-03-06 | Tongue crack feature extraction method based on wavelet transformation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109919929A true CN109919929A (en) | 2019-06-21 |
CN109919929B CN109919929B (en) | 2023-03-24 |
Family
ID=66963572
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910169550.9A Active CN109919929B (en) | 2019-03-06 | 2019-03-06 | Tongue crack feature extraction method based on wavelet transformation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109919929B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110599410A (en) * | 2019-08-07 | 2019-12-20 | 北京达佳互联信息技术有限公司 | Image processing method, device, terminal and storage medium |
CN113239790A (en) * | 2021-05-11 | 2021-08-10 | 上海道生医疗科技有限公司 | Tongue crack feature identification and length measurement method and system |
CN114187209A (en) * | 2022-02-16 | 2022-03-15 | 深圳清瑞博源智能科技有限公司 | Image defogging method, device, equipment and storage medium |
CN114693682A (en) * | 2022-06-01 | 2022-07-01 | 青岛大学附属医院 | Spine feature identification method based on image processing |
CN115641455A (en) * | 2022-09-16 | 2023-01-24 | 杭州视图智航科技有限公司 | Image matching method based on multi-feature fusion |
CN116399379A (en) * | 2023-06-07 | 2023-07-07 | 山东省科学院激光研究所 | Distributed optical fiber acoustic wave sensing system and measuring method thereof |
CN117252794A (en) * | 2023-09-25 | 2023-12-19 | 徐州医科大学 | Multi-wavelength transmission image fusion device in frequency domain |
CN117670877A (en) * | 2024-01-31 | 2024-03-08 | 陕西雷诺贝尔铝业有限公司 | Surface treatment method and system applied to aluminum veneer processing |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006017233A1 (en) * | 2004-07-12 | 2006-02-16 | Lehigh University | Image fusion methods and apparatus |
US20100225789A1 (en) * | 2008-11-14 | 2010-09-09 | ATI Technologies LLC | Image noise filter and method |
CN104504740A (en) * | 2015-01-23 | 2015-04-08 | 天津大学 | Image fusion method of compressed sensing framework |
CN106446905A (en) * | 2016-07-25 | 2017-02-22 | 广东省自动化研究所 | Surface crack texture extraction method based on fusion of seepage algorithm and adaptive Canny algorithm |
CN106910179A (en) * | 2017-01-22 | 2017-06-30 | 南京觅踪电子科技有限公司 | Multimode medical image fusion method based on wavelet transformation |
CN107133952A (en) * | 2017-05-26 | 2017-09-05 | 浙江工业大学 | A kind of alligatoring recognition methods for merging time-frequency characteristics |
CN107169979A (en) * | 2017-05-11 | 2017-09-15 | 南宁市正祥科技有限公司 | A kind of method for detecting image edge of improvement Canny operators |
-
2019
- 2019-03-06 CN CN201910169550.9A patent/CN109919929B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006017233A1 (en) * | 2004-07-12 | 2006-02-16 | Lehigh University | Image fusion methods and apparatus |
US20100225789A1 (en) * | 2008-11-14 | 2010-09-09 | ATI Technologies LLC | Image noise filter and method |
CN104504740A (en) * | 2015-01-23 | 2015-04-08 | 天津大学 | Image fusion method of compressed sensing framework |
CN106446905A (en) * | 2016-07-25 | 2017-02-22 | 广东省自动化研究所 | Surface crack texture extraction method based on fusion of seepage algorithm and adaptive Canny algorithm |
CN106910179A (en) * | 2017-01-22 | 2017-06-30 | 南京觅踪电子科技有限公司 | Multimode medical image fusion method based on wavelet transformation |
CN107169979A (en) * | 2017-05-11 | 2017-09-15 | 南宁市正祥科技有限公司 | A kind of method for detecting image edge of improvement Canny operators |
CN107133952A (en) * | 2017-05-26 | 2017-09-05 | 浙江工业大学 | A kind of alligatoring recognition methods for merging time-frequency characteristics |
Non-Patent Citations (4)
Title |
---|
张军: "基于形态学和小波变换的瓷砖缺陷边缘检测", 《计算机仿真》 * |
张军: "基于形态学和小波变换的瓷砖缺陷边缘检测", 《计算机仿真》, 5 January 2019 (2019-01-05) * |
朱石争: "小波域多聚焦图像融合算法的研究", 《中国优秀硕士论文全文电子数据库》 * |
朱石争: "小波域多聚焦图像融合算法的研究", 《中国优秀硕士论文全文电子数据库》, 15 July 2009 (2009-07-15), pages 30 - 33 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110599410A (en) * | 2019-08-07 | 2019-12-20 | 北京达佳互联信息技术有限公司 | Image processing method, device, terminal and storage medium |
CN113239790A (en) * | 2021-05-11 | 2021-08-10 | 上海道生医疗科技有限公司 | Tongue crack feature identification and length measurement method and system |
CN114187209A (en) * | 2022-02-16 | 2022-03-15 | 深圳清瑞博源智能科技有限公司 | Image defogging method, device, equipment and storage medium |
CN114187209B (en) * | 2022-02-16 | 2022-05-24 | 深圳清瑞博源智能科技有限公司 | Image defogging method, device, equipment and storage medium |
CN114693682A (en) * | 2022-06-01 | 2022-07-01 | 青岛大学附属医院 | Spine feature identification method based on image processing |
CN114693682B (en) * | 2022-06-01 | 2022-08-26 | 青岛大学附属医院 | Spine feature identification method based on image processing |
CN115641455A (en) * | 2022-09-16 | 2023-01-24 | 杭州视图智航科技有限公司 | Image matching method based on multi-feature fusion |
CN115641455B (en) * | 2022-09-16 | 2024-01-09 | 杭州视图智航科技有限公司 | Image matching method based on multi-feature fusion |
CN116399379A (en) * | 2023-06-07 | 2023-07-07 | 山东省科学院激光研究所 | Distributed optical fiber acoustic wave sensing system and measuring method thereof |
CN116399379B (en) * | 2023-06-07 | 2023-11-03 | 山东省科学院激光研究所 | Distributed optical fiber acoustic wave sensing system and measuring method thereof |
CN117252794A (en) * | 2023-09-25 | 2023-12-19 | 徐州医科大学 | Multi-wavelength transmission image fusion device in frequency domain |
CN117252794B (en) * | 2023-09-25 | 2024-04-16 | 徐州医科大学 | Multi-wavelength transmission image fusion device in frequency domain |
CN117670877A (en) * | 2024-01-31 | 2024-03-08 | 陕西雷诺贝尔铝业有限公司 | Surface treatment method and system applied to aluminum veneer processing |
Also Published As
Publication number | Publication date |
---|---|
CN109919929B (en) | 2023-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109919929A (en) | A kind of fissuring of tongue feature extracting method based on wavelet transformation | |
Dhage et al. | Watershed segmentation brain tumor detection | |
Flores et al. | Breast ultrasound despeckling using anisotropic diffusion guided by texture descriptors | |
Chan | Quantitative characterization of electron micrograph image using fractal feature | |
Jeena et al. | A comparative analysis of MRI and CT brain images for stroke diagnosis | |
Jeevakala | Sharpening enhancement technique for MR images to enhance the segmentation | |
Ganvir et al. | Filtering method for pre-processing mammogram images for breast cancer detection | |
Dash et al. | Curvelet Transform Based on Edge Preserving Filter for Retinal Blood Vessel Segmentation. | |
Bodzioch et al. | New approach to gallbladder ultrasonic images analysis and lesions recognition | |
Magud et al. | Medical ultrasound image speckle noise reduction by adaptive median filter | |
Kaur et al. | Review on medical image denoising techniques | |
Huang et al. | Wavelet k-means clustering and fuzzy-based method for segmenting MRI images depicting Parkinson’s disease | |
Rahmani et al. | Image Fusion of Noisy Images Based on Simultaneous Empirical Wavelet Transform. | |
Malathi et al. | Statistical Measurement of Ultrasound Placenta Images Complicated by Gestational Diabetes Mellitus Using Segmentation Approach. | |
Singh et al. | Feature enhancement in medical ultrasound videos using multifractal and contrast adaptive histogram equalization techniques | |
Edwin et al. | Liver and tumour segmentation from abdominal CT images using adaptive threshold method | |
Nageswara Reddy et al. | BRAIN MR IMAGE SEGMENTATION BY MODIFIED ACTIVE CONTOURS AND CONTOURLET TRANSFORM. | |
Patil et al. | Machine Learning Approach for Comparative Analysis of De-Noising Techniques in Ultrasound Images of Ovarian Tumors | |
Jabason et al. | Multimodal neuroimaging fusion in nonsubsampled shearlet domain using location-scale distribution by maximizing the high frequency subband energy | |
Shanmugavadivu et al. | Wavelet transformation-based detection of masses in digital mammograms | |
Tamilselvi et al. | Improved Gabor filter for extracting texture edge features in ultrasound kidney images | |
Jalab et al. | Liver CT enhancement using fractional differentiation and integration | |
Muthu et al. | Morphological operations in medical image pre-processing | |
Venkatachalam et al. | Processing of abdominal ultrasound images using seed based region growing method | |
Jeena et al. | Analysis of stroke using texture features |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |