CN101685435B - Multi-dimension empirical modal analysis method for analyzing image texture - Google Patents
Multi-dimension empirical modal analysis method for analyzing image texture Download PDFInfo
- Publication number
- CN101685435B CN101685435B CN200810168464.8A CN200810168464A CN101685435B CN 101685435 B CN101685435 B CN 101685435B CN 200810168464 A CN200810168464 A CN 200810168464A CN 101685435 B CN101685435 B CN 101685435B
- Authority
- CN
- China
- Prior art keywords
- group
- data
- various dimensions
- disposal route
- image data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention relates to a multi-dimension empirical modal analysis method. The technology can adaptively dissect a three-dimensional image into a plurality of featured image layers and then evidently presents a texture image in the featured image extracted. Because the technology utilizes the physical conception of field is used in the technology, an enveloping value of multi-dimensional data and a trend evaluation are broken through and reached and data mode dissection can be carried out further. The technology can also be used in the analysis of time and frequency of two dimensional data or multichannel data.
Description
Technical field
The invention relates to a kind of method of graphical analysis, and particularly relevant for a kind of multi-dimension empirical modal analysis method for analyzing image texture.
Background technology
By people such as the yellow blades of a sword (Huang N.E.), the signal decomposition that empirical mode decomposition (Empirical ModeDecomposition, EMD) method is carried out unstable state (non-stationary) and non-linear (non-linear) is proposed.The algorithm of this signal decomposition can will become the superposition of limited essential mode function (intrinsic mode function, IMF) and the remaining function of signal (monotonic function) with the signal decomposition of time correlation.At present the existing a large amount of document of one dimension empirical mode decomposition proves the signal analysis ability of its adaptability, and two-dimensional empirical mode decomposition also has be applied to that image processes for example upper: the rim detection of analyzing image texture, image, and a small amount of medical image uses.
After 2000, this technology is used in image processing, and it has been the utilization category of two-dimensional empirical mode decomposition that image is processed.From one dimension, enter into the variation of two-dimensional empirical modal on mathematical theory little.Empirical mode decomposition is to go out signature waveform from the envelope pinching in signal maximal and minmal value territory repeatedly substantially.Therefore when two-dimensional empirical modal, be only also to adopt more complicated enveloping surface to substitute the framework of simple and easy envelope.
The enveloping surface that two-dimensional empirical modal adopts be all from image lattice construction at present, or optimal values method of interpolation is assisted the construction of enveloping surface.Yet the two-dimensional empirical modal that educational circles adopts at present has three major issues to be solved.
The firstth, the very difficult definition (such as problems such as the shapes of a saddle) in the maximal and minmal value territory of two dimensional image.The secondth, current two-dimensional empirical modal is to be built in continuous data, but image is all discrete discontinuous, may have distortion.The 3rd is more than the idea of enveloping surface cannot be extended to three-dimensional.
Definition mode for extreme value (extrema), the extreme value of signal has comprised maximum value (maxima) and minimal value (minima), and extreme value definition is usingd signal intensity (intensity) extreme value except tradition and be take signal curvature (curvature) extreme value for defining as also having added definition.But prior art is not discussed when the dimension of signal is greater than 1, and how signal extreme value this define, yet being all often two dimension or three-dimensional, actual signal application even may arrive the four-dimension, as ultrasonic image, CT image and 4D ultrasonic image etc.
Empirical mode decomposition is to be completed through iteration repeatedly by screening sequence, and the construction mode that important process is exactly envelope function (envelope) in screening sequence the inside, when data dimension to be analyzed is 1, the construction mode of the envelope function being suggested is for example the Cubic Spline Functions Fitting method that the patent of United States Patent (USP) number US5983162 proposes: the fitting a straight line method that the patent of cubic spline and United States Patent (USP) number US6990436 proposes.
When data dimension to be analyzed is 2, the construction mode of the envelope function being suggested has: Mr. Huang E regards two-dimentional signal as the composition of one-dimensional signal, then use experience Mode Decomposition decomposes one-dimensional signal the essential mode function of rear recomposition two dimension, as the patent of its United States Patent (USP) number US6311130.The people such as Y.Xu adopt triangle gridding to be built in needed enveloping surface in screening sequence.The people such as Nunes J.C. adopt function interpolation mode at the bottom of radial basis to complete the construction of enveloping surface.The mode of above-mentioned employing grid has preferably result, but the method cannot be expanded to three-dimensional empirical modal analysis.Adopt method of interpolation can expand to three-dimensional in theory, but spatially differential continuity is poor for the method for interpolation.There is in addition researcher Per Cloersen to inquire into two-dimentional empirical modal analysis along with the variation of time shaft, and apply for the empirical modal method (Pub No.US2002/0186895) of three dimensionality with this.However, there is no at present the methodology of three-dimensional empirical modal analysis.
So, for the application of graphical analysis, for example, be that present medical image or other application and scientific research enter field of three dimension already, prior art is not sufficient to meet its demand.Therefore exploitation three-dimensional and more than, it is more high-dimensional that to test mode decomposition algorithm extremely important.
Summary of the invention
In view of this, the invention relates to a kind of multi-dimension empirical modal analysis method, the method can be applicable among analyzing image texture, for example, be among medical image analysis.The method can become several characteristic functions to various dimensions data decomposition adaptability, namely carries out the decomposition of data mode, for analytical applications.The decomposition of data mode is that the physical concept of " field " in applied physics is reached, and can promote thus to reach envelope value and the trend estimation of various dimensions data.For example 3-D view is resolved into after several characteristic pattern layers, by using the equation of field of physics to extract several characteristic images, wherein can obviously present different texture images, can for further analysisly apply.Because data mode decompose to adopt the concept of the field in physics, said method, can be used in two-dimensions data or the time of multichannel data, frequency analysis, and data the mode more than three-dimensional that can expand is decomposed.
According to a first aspect of the invention, propose a kind of disposal route of various dimensions data of empirical mode decomposition, the method comprises: read first group of various dimensions data; And in the mode of iteration, screen (sifting) these first group of various dimensions data to decomposite at least one essential mode function (intrinsic mode function) by empirical mode decomposition (empiricalmode decomposition).To draw an iterative process of this essential mode function, comprise screening step: shine upon a physical quantity in each value to one physical field of this first group of various dimensions data to draw a plurality of numerical value of this physical quantity in this physical field, wherein, this physical quantity is that first group of various dimensions data have same dimension and are the function of time therewith; The numerical value of this physical quantity in this physical field of learning by the value of these first group of various dimensions data, according to the variation relation of the Physical Quantity Field of relevant this physical quantity in this physical field, determines the average envelope of the distribution of this physical quantity; And according to these first group of various dimensions data average envelope therewith, determine next group various dimensions data; For this next various dimensions data group, carry out and comprise the next iteration that mapping step starts since then.
According to a second aspect of the invention, propose a kind of disposal route of various dimensions view data of empirical mode decomposition, for analyzing image texture, the method comprises: (a) read first group of multidimensional image data; (b) by empirical mode decomposition (empirical mode decomposition), in the mode of iteration, screen (sifting) these first group of multidimensional image data to decomposite a plurality of essential mode functions (intrinsic mode function); And (c) change that to turn a plurality of essential mode functions be a plurality of special multidimensional images and export these multidimensional images, to make analyzing image texture.To draw an iterative process of one of these a little essential mode functions, comprise screening step: shine upon a physical quantity in each value to one physical field of this first group of multidimensional image data to draw a plurality of numerical value of this physical quantity in this physical field, wherein, this physical quantity is that first group of multidimensional image data have same dimension and are the function of time therewith; This little numerical value of this physical quantity in this physical field of learning by the value of these first group of multidimensional image data, according to the variation relation of the Physical Quantity Field of relevant this physical quantity in this physical field, determines the average envelope of the distribution of this physical quantity; According to these first group of multidimensional image data average envelope therewith, determine next group multidimensional image data; And for this next multidimensional image data group, carry out and comprise the next iteration that mapping step starts since then.
For foregoing of the present invention can be become apparent, a preferred embodiment cited below particularly, and coordinate appended graphicly, be described in detail below:
Accompanying drawing explanation
Figure 1 shows that the process flow diagram of the first embodiment of a multi-dimension empirical mode decomposition method of the present invention.
Fig. 2 is the dynamic range distribution figure of original image.
The upper lower envelope that Fig. 3 A signal data distribute.
The average envelope that data in Fig. 3 B schematic diagram 3A distribute.
Fig. 4 is that three-dimensional data section distributes and upper and lower envelope.
Fig. 5 A is the schematic diagram of the maximum value that searches out in data.
Fig. 5 B signal estimates out whole Temperature Distribution by thermodynamics several points of Fig. 5 A.
The envelope function that Fig. 6 A signal is changed by thermal field.
Fig. 6 B illustrates the envelope function of being changed by thermal field of Fig. 6 A through the result of smoothing processing.
Fig. 7 A to 7D be illustrate respectively an original image with and the corresponding image of high frequency empirical modal, intermediate frequency empirical modal and low frequency empirical modal.
Figure 8 shows that the process flow diagram of the second embodiment of the physical quantity of a multi-dimension empirical mode decomposition method application thermal field of the present invention.
Fig. 9 A is the ultrasonic image of normal liver, with the corresponding figure layer of its essential mode function IMF1～IMF3.
Fig. 9 B is the ultrasonic image of sclerosis liver, with the corresponding figure layer of its essential mode function IMF1～IMF3.
Figure 10 A one has tumour ultrasonic image.
Figure 10 B is the low frequency modal of Figure 10 A, and it presents tumour inhomogeneity.
Figure 11 A one has calcification tumour ultrasonic image.
Figure 11 B be Figure 11 A through removing the result of low frequency modal, it can present the clear feature of calcification.
Figure 12 is signal another embodiment of the present invention, and it is that three-dimensional mode decomposition method with the embodiment of the present invention applies to the analytical structure schematic diagram of 2-D data f (x, y) under time series.
[main element label declaration]
200: input data distribute
210,230: envelope surface
250: average envelope
400: thermal field distribution surface
410: coenvelope
420: lower envelope
1101-1190: image
Embodiment
Figure 1 shows that the first embodiment of a multi-dimension empirical mode decomposition method (multi-dimensionalEMD) 100 of the present invention, it comprise by input a plurality of data groups be considered as multi-dimensional signal f (x
1, x
2..., x
n), namely a multivariable function, for example, be that the metric table of 3-D view is shown f (x, y, z), in the mode of iteration, carries out screening process, with by input multi-dimensional signal f (x
1, x
2..., x
n) resolve into the superposition of limited essential mode function (intrinsic mode function, IMF) and the remaining functions of signal (monotonic function), be shown below:
Wherein, multi-dimension empirical mode decomposition method 100 comprises that screening step 20 (sifting) tries to achieve essential mode function with iterative manner, the concept of " field ", for example the physical quantity Q of n dimension, time t in screening step 20 applied physics:
Q＝g(x
1，x
2，...，x
n，t)，
Namely by a plurality of data group f (x
1, x
2..., x
n) corresponding or be mapped to the function g (x of the physical quantity Q in the physics of identical dimensional
1, x
2..., x
n, t), and the variation relation of the Physical Quantity Field of relevant this physical quantity Q of application carries out envelope and estimates, to determine average " envelope ".For example determine " envelope " and region minimizing " envelope " thereof of this multi-dimensional signal region maximum value, further to determine average " envelope ", further to determine average " envelope ".Afterwards, according to multi-dimensional signal and this average " envelope ", obtain component signal, to do next iteration computing.For example, multi-dimensional signal is deducted to the component signal h of this average " envelope " rear gained
1, for this component signal, in step 20, find out in the same manner average envelope and the next component signal h of this component signal
11, after iteration, the component signal of last gained or average envelope meet after a condition, for example, be that average envelope approaches in level, by the component signal h finally obtaining in this way
1jbe considered as corresponding essential mode function, as shown in step 40.Afterwards, via step 60, judge whether the essential mode function of current gained is representing monotonic quantity.If so, the multi-dimensional signal f (x that represents this input
1, x
2..., x
n) resolved into limited essential mode function and the remaining function of signal.If not, multi-dimensional signal is deducted to essential mode letter h
1jto be considered as another multi-dimensional signal, again perform step 20.In this way, in the mode of iteration, perform step 20 to 60, until the multi-dimensional signal of input is resolved into limited essential mode function and the remaining function of signal and the remaining function of signal.
Among the multi-dimension empirical mode decomposition method that the above embodiments propose, applied the principle of field of physical quantity in the hope of average envelope, thereby determined essential mode function.Because the concept of field can extend to various dimensions, can meet the requirement of multidimensional data group being carried out to essential mode function decomposition, be for example for 3 d image data, even more high-dimensional data group also can produce different application.For example, actual needs with regard to medical image analysis, application aforesaid way can resolve into several essential mode functions to 3-D view adaptability, again essential mode function is converted to several characteristic pattern layers, based on characteristic pattern layer, produces the texture image that can more original image more obviously presents feature.
In addition, the concept of the field of embodiments of the invention applied physics, decomposes the mathematics of empirical modal with physics and associates, comprising by input data and a certain physics in physical quantity produce corresponding relation.The concept of " field " in various physics, for example thermal field, electromagnetic field, also can be applicable to the decomposition of intrinsic mode function.When various dimensions data be from different physical phenomenons physical quantity, the physical quantity of the variation of wave, vibrations, air pressure, humidity, electromagnetism and heat for example, empirical mode decomposition can adopt different fields to decide interpolation and envelope.In other embodiment, can be according to their physical description, " field " of employing physics is used as envelope can more meet real mode decomposition.
Come, consider the situation of the dynamic range (dynamic range) that input data group tool is larger, the concept of the field of applied physics is appropriate.For example, Figure 1 shows that the variation of the dynamic range that pixel that the histogram of an original image represents is measured, pixel measurement value changes between 0 to 255.For example, the distribution of pixel measurement value can be regarded as the Temperature Distribution in thermal field, thereby the equation of application thermal field decides the average envelope of input data group.
The second following embodiment application thermal field explains among image modalities is decomposed.It should be noted that: the idea of " envelope " can be applicable to three-dimensional so that four-dimensional above data group in an embodiment of the present invention.Because four-dimensional and above " envelope " can not can explain so that three-dimensional is graphic, therefore following examples are conveniently to illustrate with three-dimensional data group and three-dimensional enveloping surface, therefore four-dimensional and above application also can push away thus.
Before this embodiment of explanation, first basic assumption: it is to be all built in carrier wave that all data distribute.Furtherly, any data can be disassembled into and on first carrier wave, carry a signal.And if after this signal is extracted, can be considered as again the signal that second carrier wave and top are born.So repeatedly calculate the signal being finally extracted always and also can not find after carrier wave, it is exactly the mode that will find.And the searching mode of carrier wave is found the trend that these data distribute exactly, namely to determine above-mentioned average " envelope ".Understand for convenience this idea, please first referring to shown in Fig. 2 A, data curved surface 200 is distributed in for inputting data the schematic diagram that is connected to a curved surface in three-dimensional space, the curved surface (being coenvelope) that the region maximum value that envelope surface 210 is input data forms, the curved surface (being lower envelope) that the region minimal value that envelope surface 230 is input data forms.As shown in Figure 3, the coenvelope of label 250 indications and the mean value of lower envelope, envelope surface 210, and envelope surface 230 is called average envelope 250.Average envelope 250 is exactly the distribution trend of these data or is called carrier wave.The mode that will enter to find again data is exactly former data (being data curved surface 200) deduction average envelope 250.The data mode obtaining must as above be checked again, until average envelope is almost level (carrierfree).
Please refer to Fig. 8, it is for application thermal field is in the process flow diagram of an embodiment of image modalities decomposition method.Step 810 to 870 for iterative manner, screen (sifting process) with a plurality of data groups for given (or input) determining essential mode function, wherein, a plurality of data groups are for example with 3 d image data Q=f (x, y, z). Step 810 and 820 screens with iterative manner the setting of doing for preparing.In computing first, first set C
1for Q, h
j-1(t)=C
i, wherein j=1, i=1, the value of i and j will increase progressively along with the process of iteration thereafter, as shown in step 815 and 825; Wherein, the step of these settings be allow iteration calculation carrying out and use, for the common knowledge in this field, in fact can other different mode or carry out with the order being different from Fig. 8.In addition, for trying to achieve essential mode function, the data for one dimension in prior art can find average packet winding thread, and 2-D data can find average envelope face.But three-dimensional above data just cannot be with mathematical way construction.The step 830 of the present embodiment, 840 and 850 adopt the ideas of the field in physics described above, at this, are to take thermal field to try to achieve this corresponding envelope as example.
Ut=α (Uxx+Uyy+Uzz) (equation)
As for equation, calculate the method for heat distribution, for example: (1), first by the maximum value searching out, puts into math matrix computing.Schematic diagram if Fig. 5 A is the maximum value that searches out in data, wherein only has several known points and temperature (they are large values of value of data), and the graphic of script is to represent different temperatures with different colours.Thermodynamics is exactly the only information of Fig. 5 A, calculates the field that there is Temperature Distribution each position, as Fig. 5 B, shows.Used herein is thermodynamics numerical algorithm-method of finite difference (finitedifference method) in textbook, iteration to temperature stabilization convergence only.In addition, can, by the thermodynamical equilibrium equation formula of stable state directly by Matrix Solving, accelerate to try to achieve the distribution of thermal field.
(2) then through the analysis of numerical method, can obtain the temperature of all positions, distribution 400 as shown in Figure 4.Therefore the temperature of each point can be reduced into image values, for example 50 ℃ convert brightness of image 50 to.Therefore all image values after converting are exactly the coenvelope being comprised of image maximum value, in like manner also can image minimal value calculate lower envelope.Finally, as Fig. 4 section, very big codomain max[x, y, z, f (x, y, z)] to be distributed as Cmax be the coenvelope 410 on thermal field distribution surface 400 for the thermal field that forms, minimum codomain min[x, y, z,, f (x, y, z)] to be distributed as Cmin be lower envelope 430 for the thermal field that forms.
Afterwards, step 850, calculating mean value envelope is exactly the mean value of upper lower envelope.Mean value is Cmean=(Cmax+Cmin)/2.Try to achieve after the distribution of thermal field universe, more all temperature are changed into the numerical value of data.Pie graph 6A, through smoothing techniques, allows this envelope function meet differentiable character figure, as shown in Figure 6B person.
In order to verify that this empirical modal is for the ability of picture breakdown, we illustrate with a simple grid image: original image comprises thick and careful grid chart 7A, through after empirical mode decomposition, the empirical modal of a low high-frequency domain resolves into a separation graph that only comprises careful grid by original image, as Fig. 7 B.The separate picture that last empirical modal presents infra-low frequency and lowest frequency is Fig. 7 C and Fig. 7 D.
The extraction of empirical modal is progressively to complete one by one Fig. 8.One group of original three dimensionality view data is resolved into i empirical modal (wherein comprise signal remaining function), and the computation process of each empirical modal whether must have j iteration to take inspection carrier wave be level.In this process, the component signal h that step 870 judgement is current
j(t) whether be essential mode function.Its judgment mode can have different modes to implement.For example, with current component signal h
j(t) or average envelope Cmean whether meet a condition with judgement, as whether approached level with average envelope.Or, utilize the change of average envelope whether enough little, that is whether lower than threshold value, judgement component signal h
j(t) whether can be considered essential mode function.Can whether equate in fact to judge with continuous several component signal next again.
In addition, relevant for inputting the corresponded manner of the amount in data and physical field, in an example, through the discrete point after extreme value search by image actual numerical value, with linear transformation, correspond to the physics scale (for example intensity, density, temperature) of " physical field ", and carry out the calculating of whole.Scale after the calculating again antilinear transformation correspondence image values of going back completes various dimensions envelope.In another example, be by the discrete point after the search of image actual numerical value process extreme value, correspond to the physical vector of " physical field ", and carry out the calculating of whole equipotential surface, gradient.Equipotential surface after calculating is carried out numerical interpolation to complete various dimensions envelope.
In addition, above-mentioned application " physical field " in the hope of enveloping method in, take thermal field as embodiment, can and three-dimensionally with identical mathematical method, identical equation, substitute the problem that adopts enveloping surface in known technology above by one dimension, two dimension.And a boundary condition calculating adopts numerical interpolation to process.
Cited embodiment has illustrated above, adopt the concept of field and extend to various dimensions, can meet the requirement of multidimensional data group being carried out to essential mode function decomposition, for example for 3 d image data, even more high-dimensional data group also can produce different application, for example with the angle of empirical modal image, come the place of difference and the feature of analysis image, to judge accordingly analysis.
Just take Medical Image Processing below as example, actual needs with explanation with regard to medical image analysis, application aforesaid way embodiment can resolve into several essential mode functions to 3-D view adaptability, again essential mode function is converted to several characteristic pattern layers, based on characteristic pattern layer, produces the texture image that can more original image more obviously presents feature.
Tumour can manifest the clear border of focus when iconology checks, but when tumour is infiltrative growth sometimes, can cause obscurity boundary unclear, so claim " wellability " tumour, prognosis comes poorly than general tumour.
Although many researchists are constantly made great efforts to find out effective characteristics of image or are carried out figure image intensifying by ultrasonic image and CT Scan, make experienced doctor can from image, assess infiltration degree and the classification of tumour, yet still cannot judge accurately for comparatively serious situation.In addition also there are many key characters in the inside of tumour, such as high echo region (hyperechoic area), low echo region (Hypoechoic area), organize the features such as homogenieity (Heterogeneity), microcalcifications (Microcalcification), fiberization (Fibrosis).These are all relevant to the grade malignancy of tumour.
The image of two dimension comes from the technician of hospital acquisition, and the artificial otherness that exists tangent plane to select cannot be as future in a large amount of screenings or the standard program in strong inspection field.Therefore must develop and capture 3-D view as auxiliary diagnosis.Meanwhile, the technology in image place also must be followed this standard and carry out more high-dimensional calculating.The rim detection of tumor image is a complicated calculation program often, adopts in the past statistics, frequency spectrum and taxonomic mode to still have its restriction.Because fresh rare algorithm is to carry out graphical analysis self adaptively, and dummy's imposing a condition of being not completely.
In order to process more high-dimensional tumor image analysis, the ultrasonic image that especially comes from image background and be easily disturbed.Must develop the adaptability image processing techniques of more high-dimensional (being at least three-dimensional).We utilize empirical mode decomposition method, and this Technique Popularizing is above to meet new medical image standard from generation to generation at present to more high-dimensional three-dimensional.
Due to the processing mode of technology fine setting adaptive of the present invention, be applicable to the inevasible image of ground unrest and external interference and process, for example ultrasonic medical image.
Therefore the image modalities decomposition method that proposes a kind of various dimensions image of the 3rd embodiment according to the present invention, at least comprises the following step: first, provide image pre-treatment standardisation process; When decomposing, mode must to have enveloping methods more than various dimensions (at least two dimension); Finally, after mode is decomposed, show that each figure layer carries out image to do texture analysis.In the middle of, each figure layer can be optionally in addition computing and make other image and process, to help the use of texture analysis.
If for the analysis of tumor image, can be for comprising the analysis of tissue odds's matter (Heterogeneity), microcalcifications (Micro calcification), fiberization (Fibrosis) feature.In the texture analysis of tumor image, utilize the high frequency figure layer (detail section) after the model analysis of the first or second embodiment of the invention described above to carry out the modification (superposition is the simplest embodiment) of former figure, strengthening borderline tumor auxiliary diagnosis.
High frequency figure layer can carry out the texture unrest degree (entropy) of tumour and analyze with assessment liver fibrosis.The ultrasonic liver fibrosis image of take is example, and the obvious ultrasonic image of fiberization presents compared with graininess, inhomogeneous texture.These phenomenons of past are very easily subject to signal along with the impact that scans the degree of depth and penetrate different tissues decay, cause texture analysis not objective.Take this ultrasonic medical image is embodiment, original ultrasonic liver fibrosis sectioning image is carried out to empirical mode decomposition, can be in order to compare the difference of the tissue of normal condition (normal) and abnormal conditions, respectively as Fig. 9 A and 9B those shown, the difference of the second layer (IMF2) and the 3rd layer of corresponding image of (IMF3) empirical modal in Fig. 9 A and 9B especially.
The random degree (Entropy) of IMF3 has good ability (p<0.005 on the textural characteristics of describing cirrhosis, CV<10%), therefore be effective with Empirical mode decomposition in cirrhosis ultrasonic analyzing image texture.Past adopts doctor's thing professional according to naked eyes, and ultrasonic only can be diagnosed for the sufferer of falling ill or liver cancer is very serious.If can take the part empirical mode decomposition of original image, carry out the standard mode of the random metrization of image, really can be at the liver cancer initial stage, or cirrhosis initial stage more early and carry out screening and postoperative long-term tracking etc.
Low frequency figure layer (background) can carry out tumour inhomogeneity (Heterogeneity) assessment, as shown in Figure 10 A and Figure 10 B.Figure 10 A has been depicted as tumour ultrasonic image, and Figure 10 B is its low frequency modal (being background), has presented tumour inhomogeneity.
In addition, from original image, remove low frequency figure layer (background), can clearlyer present calcified tissue.Figure 11 A is for there being calcification tumour ultrasonic image, and Figure 11 B is for removing the clear feature that low frequency modal (background) presents calcification afterwards.
At front the 3rd described embodiment, be applied in the pretreatment process of medical image, especially ultrasonic medical image is very easily subject to depth compensation (depth compensation), signal attenuation (attenuation) causes image disruption background uneven.Image is discrete data, if image when poor physical quantity resolution (as intensity, density, temperature etc.) is poor, can cause mode to decompose mistake.Therefore we propose the check of the essential process of medical image pixels-brightness histogram.This histogram can be used as the reference of capture, and in dynamic range (Dynamic Range), number of pixels is greater than 1 number n and must exceeds a critical value (for example critical value is about 128) and can carry out empirical mode decomposition.
In addition, application embodiments of the invention decompose and also can apply to the time frequency analysis of 2-D data f (x, y) under time series t in three-dimensional mode, as shown in figure 12, as image 1101 to 1190, be respectively 2-D data, be for example a certain object situation of activity in time.Time t can be expanded to third dimension entity coordinate, Using such method is decomposition space and time mode feature decomposition simultaneously.Be applied on medical image analysis, be for example when the image of observing as the image of heart or liver be on the application changing in time.So, on can also the analysis for the upper time dependent view data of three-dimensional.
The embodiment of the present invention also discloses a kind of computer-readable information storage media, stores program on it, and this program can be used for carrying out the multi-dimension empirical mode decomposition method of the embodiment of the present invention.The computer-readable information storage media of the present embodiment such as but be not limited to optical information Storage Media, magnetic-type information storage media.Optical information Storage Media is such as comprising CD, DVD, HD-DVD, blue-ray DVD etc.Magnetic-type information storage media are such as comprising floppy drive, Winchester disk drive, magnetic tape station, magneto-optical drive (MagneticOptical) etc.In addition, also comprise can be at the upper program code transmitting of network/transmission medium (as air etc.) etc. for computer-readable information storage media.
The embodiment of the present invention also discloses a kind of computer program.When having the electronic installation of memory buffer, load after this computer program, this electronic installation is carried out a plurality of programmed instruction, and those programmed instruction are for carrying out the multi-dimension empirical mode decomposition method of the embodiment of the present invention.
The embodiment of the present invention also discloses a kind of electronic installation, for example personal computer or notebook computer or hand-held arithmetic unit, to such an extent as to there is image capturing device or medical image device and the analytical equipment thereof of data processor, when thering is the electronic installation of memory buffer and data processor, load after above-mentioned computer program, this electronic installation is carried out a plurality of programmed instruction, and those programmed instruction are for carrying out the multi-dimension empirical mode decomposition method of the embodiment of the present invention.In addition, this electronic installation also can have display with the figure layer of display analysis.In other example, this electronic installation also has input media with acquisition or from the external world, reads multidimensional data group.In other example, this electronic installation also has User's Interface, optionally the figure layer of analyzing is done to arithmetic operation or image processing, with exhibit textural feature with tool.
Cited embodiment has illustrated above, adopt the concept of field and extend to various dimensions, can meet the requirement of multidimensional data group being carried out to essential mode function decomposition, for example for 3 d image data, even more high-dimensional data group also can produce different application, for example with the angle of empirical modal image, come the place of difference and the feature of analysis image, to judge accordingly analysis.As the intrinsic mode function of above-mentioned application image, also there are many important textural characteristics in the inside that can effectively describe tumour, carries out objective quantitative test with early detection liver fibrosis.
In sum, although the present invention discloses as above with preferred embodiment, so it is not in order to limit the present invention.Persond having ordinary knowledge in the technical field of the present invention, without departing from the spirit and scope of the present invention, when being used for a variety of modifications and variations.Therefore, protection scope of the present invention is when being as the criterion depending on the appended claim scope person of defining.
Claims (27)
1. a disposal route for the various dimensions data of empirical mode decomposition, the method comprises:
Read first group of various dimensions data;
By empirical mode decomposition, in the mode of iteration, screen these first group of various dimensions data to decomposite at least one essential mode function, wherein, to draw an iterative process of this essence mode function, comprise screening step:
Shine upon a physical quantity in each value to one physical field of this first group of various dimensions data to draw a plurality of numerical value of this physical quantity in this physical field, wherein, this physical quantity is to have same dimension with these first group of various dimensions data and is the function of time;
Those numerical value of this physical quantity in this physical field of learning by the value of these first group of various dimensions data, according to the variation relation of the physical field of relevant this physical quantity in this physical field, determine the average envelope of the distribution of this physical quantity; And
According to this first group of various dimensions data and this average envelope, determine next group various dimensions data;
For this, next group various dimensions data, carries out and comprises the next iteration starting from this mapping step,
Wherein, this step that determines this average envelope comprises:
According to the variation relation of this physical field of relevant this physical quantity in this physical field, determine to correspond to coenvelope and the lower envelope of these first group of various dimensions data;
Determine this average envelope of this coenvelope and this lower envelope.
2. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, this mapping step is, in linear relationship mode, each value of these first group of various dimensions data is mapped to this physical quantity in this physical field to draw a plurality of numerical value of this physical quantity in this physical field.
3. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, this physical field is thermal field, and this physical quantity is the temperature value in this thermal field.
4. the disposal route of the various dimensions data of empirical mode decomposition according to claim 3, wherein, the variation relation of this physical field is thermal field equation, represents that in thermal field, the Temperature Distribution with space changes.
5. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, also comprises in an iterative process in this screening step:
Utilize the change of average envelope whether enough little, judge whether these next group various dimensions data can be considered essential mode function.
6. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, these next group various dimensions data are the component signal of these first group of various dimensions data, and those steps in this screening step are to carry out until next those component signals are equal in fact with iterative manner.
7. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, first of this physical field that the very big codomain that this coenvelope is these first group of various dimensions data forms distributes, and second of this physical field that the minimum codomain that this lower envelope is these first group of various dimensions data forms distributes.
8. the disposal route of the various dimensions data of empirical mode decomposition according to claim 7, wherein, this average envelope is the mean value of this coenvelope and this lower envelope.
9. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, with this average envelope, this step that determines these next group various dimensions data comprises:
Reduce this shine upon this this average physical quantity for and this first group of numerical value that various dimensions data are corresponding;
These next group various dimensions data are the difference of this average envelope after these first group of various dimensions data and reduction.
10. the disposal route of the various dimensions data of empirical mode decomposition according to claim 1, wherein, these first group of various dimensions data is 3 d image data.
The disposal route of the various dimensions data of 11. empirical mode decompositions according to claim 1, wherein, these first group of various dimensions data is the two-dimensional image data under time series.
The disposal route of the various dimensions view data of 12. 1 kinds of empirical mode decompositions, for analyzing image texture, the method comprises:
(a) read first group of multidimensional image data;
(b) by empirical mode decomposition, in the mode of iteration, screen these first group of multidimensional image data to decomposite a plurality of essential mode functions, wherein, to draw an iterative process of one of those essential mode functions, comprise screening step:
Shine upon a physical quantity in each value to one physical field of this first group of multidimensional image data to draw a plurality of numerical value of this physical quantity in this physical field, wherein, this physical quantity is to have same dimension with these first group of multidimensional image data and is the function of a time;
Those numerical value of this physical quantity in this physical field of learning by the value of these first group of multidimensional image data, according to the variation relation of the physical field of relevant this physical quantity in this physical field, determine the average envelope of the distribution of this physical quantity;
According to this first group of multidimensional image data and this average envelope, determine next group multidimensional image data; And
For this, next group multidimensional image data, carries out and comprises the next iteration starting from this mapping step; And
(c) change that to turn those essential mode functions be a plurality of special multidimensional images and export those multidimensional images, to make analyzing image texture,
This step that determines this average envelope comprises:
According to the variation relation of this physical field of relevant this physical quantity in this physical field, determine to correspond to coenvelope and the lower envelope of these first group of various dimensions data;
Determine this average envelope of this coenvelope and this lower envelope.
The disposal route of the multidimensional image data of 13. empirical mode decompositions according to claim 12, wherein, this mapping step be in linear relationship mode by each value of these first group of multidimensional image data to this physical quantity in this physical field to draw a plurality of numerical value of this physical quantity in this physical field.
The disposal route of the multidimensional image data of 14. empirical mode decompositions according to claim 12, wherein, this physical field is thermal field, and this physical quantity is the temperature value in this thermal field.
The disposal route of the multidimensional image data of 15. empirical mode decompositions according to claim 14, wherein, the variation relation of this physical field is thermal field equation, represents Temperature Distribution variation in time in thermal field.
The disposal route of the multidimensional image data of 16. empirical mode decompositions according to claim 12, wherein, also comprises in an iterative process in this screening step:
Judge whether this next group multidimensional data can be considered essential mode function.
The disposal route of the multidimensional image data of 17. empirical mode decompositions according to claim 12, wherein, these next group various dimensions data are the component signal of these first group of multidimensional image data, and those steps in this screening step are to carry out until next those component signals are equal in fact with iterative manner.
The disposal route of the multidimensional image data of 18. empirical mode decompositions according to claim 12, wherein, first of this thing amount field that the very big codomain that this coenvelope is these first group of various dimensions data forms distributes, and second of this thing amount field that the minimum codomain that this lower envelope is these first group of various dimensions data forms distributes.
The disposal route of the multidimensional image data of 19. empirical mode decompositions according to claim 18, wherein, this average envelope is the mean value of this coenvelope and this lower envelope.
The disposal route of the multidimensional image data of 20. empirical mode decompositions according to claim 12, wherein, with this average envelope, this step that determines these next group multidimensional image data comprises:
Reduce this shine upon this this average physical quantity for and this first group of numerical value that multidimensional image data are corresponding;
These next group multidimensional image data are the difference of this average envelope after these first group of multidimensional image data and reduction.
The disposal route of the multidimensional image data of 21. empirical mode decompositions according to claim 12, wherein, those multidimensional images also pass through smoothing techniques.
The disposal route of the multidimensional image data of 22. empirical mode decompositions according to claim 12 also comprises:
According to those multidimensional images at least one, optionally carry out image processing.
The disposal route of the multidimensional image data of 23. empirical mode decompositions according to claim 12 also comprises:
According to this first group of multidimensional image, produce corresponding histogram, according to this histogram, determine whether the dynamic range of this first group of multidimensional image meets a critical value, thereby determine whether to be applicable to carrying out this screening step.
The disposal route of the various dimensions data of 24. empirical mode decompositions according to claim 12, wherein, these first group of various dimensions data is ultrasonic image, CT image or 4D ultrasonic image.
The disposal route of the various dimensions data of 25. empirical mode decompositions according to claim 12, wherein, these first group of various dimensions data is the two-dimensional image data under time series.
The disposal route of the various dimensions data of 26. empirical mode decompositions according to claim 12, wherein, these first group of multidimensional image data is tumor image, the difference figure layer that those multidimensional images are this tumor image, for tumor image analysis.
The disposal route of the various dimensions data of 27. empirical mode decompositions according to claim 26 is can be for analysis and the auxiliary diagnosis of tissue odds's matter, microcalcifications or fiberization feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200810168464.8A CN101685435B (en) | 2008-09-26 | 2008-09-26 | Multi-dimension empirical modal analysis method for analyzing image texture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200810168464.8A CN101685435B (en) | 2008-09-26 | 2008-09-26 | Multi-dimension empirical modal analysis method for analyzing image texture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101685435A CN101685435A (en) | 2010-03-31 |
| CN101685435B true CN101685435B (en) | 2014-03-12 |
Family
ID=42048599
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200810168464.8A Expired - Fee Related CN101685435B (en) | 2008-09-26 | 2008-09-26 | Multi-dimension empirical modal analysis method for analyzing image texture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101685435B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853401B (en) * | 2010-06-25 | 2012-05-30 | 哈尔滨工业大学 | Multi-packet image classification method based on two-dimensional empirical mode decomposition |
| CN101916439B (en) * | 2010-07-27 | 2012-08-08 | 哈尔滨工业大学 | Space debris hyperspectral sequence detection method based on Hilbert-Huang transform |
| CN101872425B (en) * | 2010-07-29 | 2012-06-27 | 哈尔滨工业大学 | Empirical mode decomposition based method for acquiring image characteristics and measuring corresponding physical parameters |
| CN102184529B (en) * | 2011-05-12 | 2012-07-25 | 西安电子科技大学 | Empirical-mode-decomposition-based edge detecting method |
| CN103455977B (en) * | 2012-06-04 | 2016-12-14 | 中国科学院深圳先进技术研究院 | A kind of method and system eliminating endoscopic images uneven illumination |
| CN105651778B (en) * | 2016-01-20 | 2018-11-20 | 成都理工大学 | Mineral surfaces roughness value calculation method based on confocal microscope observation data |
| TWI619475B (en) * | 2017-02-21 | 2018-04-01 | Ultrasonic data analysis system and method for liver fibrosis | |
| CN107220665B (en) * | 2017-05-18 | 2019-12-20 | 清华大学深圳研究生院 | Image interpolation method and two-dimensional empirical mode decomposition method based on same |
| CN108062768A (en) * | 2017-12-12 | 2018-05-22 | 中国矿业大学 | It is a kind of based on the wirerope axis of surface texture feature to mobile status recognition methods |
| CN110398363B (en) * | 2018-04-19 | 2021-03-16 | 中国科学院沈阳自动化研究所 | RV reducer performance state monitoring method based on graphical transformation |
| CN111275022B (en) * | 2020-03-19 | 2023-05-16 | 山东宜佳成新材料有限责任公司 | Stain detection analysis method based on forgetting factor type empirical mode decomposition and application |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101046497A (en) * | 2006-03-29 | 2007-10-03 | 宝山钢铁股份有限公司 | Method of detecting local discharge signal in electric power equipment |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100583656C (en) * | 2006-11-02 | 2010-01-20 | 中山大学 | Audio frequency noise removing method based on communication equipment users' end |
-
2008
- 2008-09-26 CN CN200810168464.8A patent/CN101685435B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101046497A (en) * | 2006-03-29 | 2007-10-03 | 宝山钢铁股份有限公司 | Method of detecting local discharge signal in electric power equipment |
Non-Patent Citations (2)
| Title |
|---|
| 史坤鹏.基于信号能量分析的电力系统低频振荡模式识别.《华北电力大学工学硕士学位论文》.2008,9－53. |
| 基于信号能量分析的电力系统低频振荡模式识别;史坤鹏;《华北电力大学工学硕士学位论文》;2008062;9－53 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101685435A (en) | 2010-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101685435B (en) | Multi-dimension empirical modal analysis method for analyzing image texture | |
| US8428389B2 (en) | Multi-dimensional empirical mode decomposition (EMD) method for image texture analysis | |
| McGarigal et al. | The gradient concept of landscape structure | |
| Dhruv et al. | Study of Haralick’s and GLCM texture analysis on 3D medical images | |
| Guo et al. | Edge guided reconstruction for compressive imaging | |
| Sneha et al. | Cervical cancer detection and classification using texture analysis | |
| US20160225146A1 (en) | Method and system for analysis of volumetric data | |
| CN102760236B (en) | Priori shape modeling method based on combined sparse model | |
| CN103337074A (en) | Active contour model based method for segmenting mammary gland DCE-MRI focus | |
| JP2008515466A (en) | Method and system for identifying an image representation of a class of objects | |
| CN104392444A (en) | Method of extracting characteristics of medical MR (magnetic resonance) images based on ensemble empirical mode decomposition | |
| Kollem et al. | Modified transform‐based gamma correction for MRI tumor image denoising and segmentation by optimized histon‐based elephant herding algorithm | |
| Srivastava et al. | A combined approach for the enhancement and segmentation of mammograms using modified fuzzy C-means method in wavelet domain | |
| CN107392863A (en) | SAR image change detection based on affine matrix fusion Spectral Clustering | |
| CN101782895B (en) | Method and device thereof for performing hierarchical empirical mode decomposition on signal | |
| Doraiswamy et al. | Topological saliency | |
| Moussa et al. | Structure analysis and surface simulation of woven fabrics using fast Fourier transform techniques | |
| Yu et al. | iVAR: Interactive visual analytics of radiomics features from large-scale medical images | |
| Jaffar et al. | Tumor detection from enhanced magnetic resonance imaging using fuzzy curvelet | |
| Xiong et al. | Singularity-exponent-domain image feature transform | |
| Wang et al. | Using chaos world cup optimization algorithm for medical images contrast enhancement | |
| Krishnammal et al. | Automated brain image classification using neural network approach and abnormality analysis | |
| Wei et al. | An improved image segmentation algorithm ct superpixel grid using active contour | |
| Zhu et al. | Learning classification of big medical imaging data based on partial differential equation | |
| Kahali et al. | Convolution of 3D Gaussian surfaces for volumetric intensity inhomogeneity estimation and correction in 3D brain MR image data |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140312 Termination date: 20190926 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |