CN101669819B - Electrocardiogram signal lossless compression method based on PT conversion and linear prediction combination - Google Patents

Abstract

The invention discloses an electrocardiogram signal lossless compression method based on PT conversion and linear prediction combination, belonging to the technical field of data processing. The method comprises the following compression processes: carrying out PT conversion on an original electrocardiogram signal; subtracting a converted original signal value and an estimated value to obtain a residual error signal of the whole electrocardiogram signal; carrying out self-adaption variable grade RICE coding on the residual error signal to output a coding bit stream and finishing the lossless compression on the electrocardiogram signal; decoding the compressed bit stream; reconstructing the signal and finishing decompression. The lossless compression method ensures the information completion and accuracy of the compressed electrocardiogram signal and can be used for transmitting and storing the electrocardiogram signal.

Description

Based on PT conversion and the bonded electrocardiogram signal lossless compression method of linear prediction

Technical field

The invention belongs to technical field of data processing, relate to data compression, in order to realize lossless compress electrocardiogram ECG signal.

Background technology

ECG signal is one of most important bio signal.Development along with modern medicine, ECG signal assosting effect to diagnostic accuracy in clinical diagnosis is also increasing, use more and more widely, the status in modern medicine is apparent more important, and ECG signal is that outstanding contribution has also been made in the development of modern medicine clinical diagnosis.In order not influence clinical diagnosis, the lossless compress of information is a best choice, because the ECG signal data volume is bigger, seems very necessary so compress for ECG signal.

In the recent period, some scholar has proposed some compression methods at ECG signal, and as based on method of wavelet, based on method of linear prediction etc., but most method is the lossy compression method method, can't guarantee fully the integrity and the correctness of ECG signal.

The RAR that existing application is wider, the ZIP method, be in field of data compression two kinds of methods of maturation and wide adaptability comparatively, RAR and ZIP method all are that original storage format of main utilization change data realizes the compression to data, the algorithm of both cores all is to realize on the basis of LZW encryption algorithm, belongs to the coding compression algorithm based on dictionary.。Though these two kinds of methods can realize the lossless compress to ECG signal, compression ratio is low excessively, causes the unnecessary storage and the waste of transmission space.

Summary of the invention

The objective of the invention is to overcome the shortcoming and defect of above-mentioned prior art, propose a kind ofly,, improve compression ratio, conserve memory and transmission space to keep the complete and correctness of information based on PT conversion and the bonded electrocardiogram signal lossless compression method of linear prediction.

The key problem in technology of realizing the object of the invention is to adopt the PT conversion, autoregression linear prediction and adaptive variable rank RICE coding, and implementation step comprises as follows:

(1) primary ECG signal is carried out the PT conversion;

(2) signal after the conversion is predicted according to the following procedure;

(2a) choose a segment signal and train, be fixed the predictive coefficient group;

(2b) to current some X _n2k before point trained, and obtains the dynamic prediction coefficient sets, and k is a prediction order;

(2c) select to adopt dynamic prediction coefficient sets or fixing predictive coefficient group according to the thresholding of setting, when the training matrix determinant is lower than threshold value, then choose fixedly the predictive coefficient group as predictive coefficient, otherwise, adopt present dynamic predictive coefficient group as predictive coefficient;

(2d), utilize the autoregression formula, try to achieve current some X according to the predictive coefficient of determining _nEstimated value;

(2e) repeat (2b)-(2d), obtain the estimated value of whole segment signal, finish prediction ECG signal after the conversion;

(3) original signal value after the conversion and estimated value are subtracted each other, obtain the residual signals of whole section ECG signal;

(4) residual signals is carried out following self-adapting changeable rank RICE coding;

(4a) establish current some e _n, according to e _nThe average of the big or small absolute value of this n point value is tried to achieve in the distribution of numerical values recited of n before point; Setting is with 2 ²Be thresholding, per 2 ²Be single order,, select e according to the thresholding at average place _nThe coding exponent number;

(4b), determine e according to as giving a definition _nPreceding sign indicating number section and exponent number sign indicating number section, wherein k is the exponent number of encoding:

Preceding sign indicating number section: e _nDivide exactly 2 ^k, the gained merchant is n _a, then with n _aIndividual 1 and one 0 as preceding sign indicating number section;

Exponent number sign indicating number section: e _nDivide exactly 2 ^k, the gained remainder is n _e, then with k n _eBinary code as exponent number sign indicating number section;

(4c) preceding sign indicating number section and exponent number sign indicating number are merged, finish e _nCoding;

(4d) coding to residual signals is finished in repetition (4a)-(4c);

(4e) output encoder bit stream is finished the lossless compress to ECG signal;

This method compared with prior art has the following advantages:

A. the present invention can recover original signal fully as a kind of lossless compression algorithm, and the complete and correctness, particularly ECG signal that have kept data are as a kind of medical signals, and the maintenance complete and correctness of its data message is very necessary.

B. the present invention makes signal have better predictability after conversion owing to adopted the PT alternative approach.

C. the present invention has adopted autoregressive linear predictor, and has designed a kind of at the adaptive RICE coded method of signal characteristic.Make it all can reach more excellent coding compression performance to different signals.

D. simulation result shows, the present invention and RAR, and the ZIP compression method is compared, and the lossless compress performance is more excellent

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention;

Fig. 2 carries out result after the PT conversion with the present invention to one section ECG signal.

The specific embodiment

With reference to Fig. 1, specific implementation step of the present invention is as follows:

Step 1 is carried out PT (peak transform) conversion to primary ECG signal;

The PT conversion is a kind of nonlinear geometric transformation, and it can be converted into low frequency with high-frequency signal, and it is defined as follows:

Curved section: be defined as the continuous function f (x) in the finite interval [a, b].

Curved section stack: given two curved section f ₁(x), f ₂(x) be defined in finite interval [a respectively ₁, b ₁], [a ₂, b ₂] on, and b ₁≤ a ₂, the new curved section f (x) that stack is generated is defined in [a ₁, b ₁+ b ₂-a ₂].Then:

$f\left(x\right)=\left\{\begin{array}{cc}{f}_1\left(x\right),& x\∈[a_1,b_1]\\ f_2\left(x\right)-f_2\left(a\right)+f_1\left(b\right),& x\∈(b_1,b_1+b_2-a_2]\end{array}\right.$

Wherein, f ₂(a) be f ₂(x) functional value of ordering at a, f ₁(b) be f ₁(x) functional value of ordering at b.New curved section f (x) is exactly by f ₁(x), f ₂(x) obtain by connection and displacement.

The PT conversion has only changed the order of curved section, and is reversible.Reciprocal transformation can come superimposed curves section again according to their original order.

By as above definition, we carry out a PT conversion to primary ECG signal, obtain the signal after the conversion, as shown in Figure 2, wherein Fig. 2 a is an original electrocardiographicdigital figure signal, and Fig. 2 b is the ECG signal after the PT conversion, can see, signal after the conversion, more level and smooth than primary signal, high fdrequency component obviously reduces.

Step 2, the signal after using the autoregression linear predictor to conversion carries out the linear prediction of k rank, tries to achieve estimated signal, further tries to achieve residual signals by estimated signal.

(2a) choose one section signal after the conversion, train, be fixed the predictive coefficient group according to following formula:

$\stackrel{\→}{a}={\left(C^{T}C\right)}^{-1}{C}^T\stackrel{\→}{X}---\left(1\right)$

Wherein

Be the vector of predictive coefficient group, C is an estimated matrix,

Vector for selected signal segment;

(2b) basis (1) formula is to current some X _n2k before point trained, and obtains the dynamic prediction coefficient sets, and k is a prediction order;

(2c) select to adopt dynamic prediction coefficient sets or fixing predictive coefficient group, the matrix (C in (1) formula according to the thresholding of setting ^TC) ^-1The value of determinant during less than this thresholding, then choose fixedly the predictive coefficient group as predictive coefficient, otherwise, adopt present dynamic predictive coefficient group as predictive coefficient;

(2d), carry out autoregression according to following formula according to the predictive coefficient of determining:

$\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|\right|X_{\mathrm{en}}-\underset{t=1}{\overset{k}{\mathrm{\Σ}}}{a}_t{X}_{n-t}\left|\right|+\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\left|\right|X_n-\underset{t=1}{\overset{k}{\mathrm{\Σ}}}{a}_t{X}_{e(n-t)}\left|\right|---\left(2\right)$

Wherein, X _EnBe current some X _nEstimated value, k is a prediction order, the predictive coefficient of a for determining, X _N-tBe X _nThe before t point, X _{E (n-t)}Be X _EnThe before t point.

Obtain estimated value X by (2) formula _En

(2e) repeat (2b)-(2d), obtain the estimated value X of whole segment signal _e, finish prediction to ECG signal after the conversion;

Step 3 is with original signal value X after the conversion and estimated value X _eSubtract each other, further encoded by following formula and compress required predicted residual signal value e _x

e _x＝X-X _e (3)

Step 4 is carried out following self-adapting changeable rank RICE coding to residual signals;

(4a) establish current some e of residual signals _n, according to e _nThe average of the big or small absolute value of this n point value is tried to achieve in the distribution of numerical values recited of n before point; Setting is with 2 ²Be thresholding, per 2 ²Be single order,, select e according to the thresholding at average place _nThe coding exponent number;

(4b), determine e according to as giving a definition _nPreceding sign indicating number section and exponent number sign indicating number section, wherein k is the exponent number of encoding:

Preceding sign indicating number section: e _nDivide exactly 2 ^k, the gained merchant is n _a, then with n _aIndividual 1 and one 0 as preceding sign indicating number section;

Exponent number sign indicating number section: e _nDivide exactly 2 ^k, the gained remainder is n _e, then with k n _eBinary code as exponent number sign indicating number section;

(4c) preceding sign indicating number section and exponent number sign indicating number are merged, finish e _nCoding;

(4d) coding to residual signals is finished in repetition (4a)-(4c);

(4e) output encoder bit stream is finished the lossless compress to ECG signal;

With reference to Fig. 1, decompression process of the present invention is as follows:

Steps A, according to the bit stream that decoding end is received, decode according to the following procedure:

(1A) 0,1 number of bits of receiving according to decoding end is determined current some e of residual signals _nPreceding sign indicating number section decoding n _aDecoding n with exponent number sign indicating number section _e

(2A) determine exponent number k according to the number of bits of the exponent number sign indicating number section that obtains;

(3A) by preceding sign indicating number section decoding n _a, exponent number sign indicating number section decoding n _eWith exponent number k, obtain current some e _n:

e _n＝n _a×2 ^k+n _e；

(4A) repeat (1A)-(3A), obtain whole section residual signals value e _X

Step B is with the residual signals value e that obtains _XWith the ECG signal estimated value X after the conversion _eAddition obtains the original signal value X after the conversion, that is:

X＝e _X+X _e；

Step C carries out the PT inverse transformation to the ECG signal after the conversion, obtains original electrocardiographicdigital figure signal, realizes reconstruct, and decompression is finished.

Effect of the present invention can further specify by following concrete experimental data.

1, experimental condition and content

Experiment of the present invention is 8 file size unanimities choosing among the MIT-BIH ECG signal compression verification data base, the identical ECG signal file of signal length is as test sample book, to the test sample book chosen respectively according to compression method of the present invention, the RAR compression method, the ZIP compression method carries out lossless compress test, and the compression ratio of three kinds of methods is compared.

2, result of the test

The compression ratio that three kinds of distinct methods obtain is as shown in table 1.

The compression ratio that three kinds of methods of table 1 obtain

Test sample book	The inventive method compression ratio	RAR compression method compression ratio	ZIP compression method compression ratio
				11442_01	2.03	?1.91	?1.27
08730_01	2.25	?2.02	?1.23
				08730_02	2.15	?1.92	?1.18
11950_01	2.12	?1.96	?1.27
				08730_04	2.06	?1.86	?1.13
12531_02	2.05	?1.95	?1.32
				12621_02	2.16	?1.94	?1.18
12490_01	2.22	?2.06	?1.35

As seen from Table 1, compression method of the present invention and RAR compression method, the ZIP compression method is compared, the compression ratio of all test files all be higher than the back both.Contrast three kinds of methods, the compression ratio of RAR compression method is a little less than the inventive method, and the ZIP compression method then all is lower than the inventive method and RAR compression method on compression ratio.

Claims (3)

1. one kind based on PT conversion and the bonded electrocardiogram signal lossless compression method of linear prediction, comprises the steps:

(1) primary ECG signal is carried out the PT conversion;

(2) signal after the conversion is predicted according to the following procedure;

(2a) choose a segment signal and train, be fixed the predictive coefficient group according to following formula:

$\stackrel{\→}{a}={\left(C^{T}C\right)}^{-1}{C}^T\stackrel{\→}{x}---1)$

Wherein

Be the vector of predictive coefficient group, C is an estimated matrix,

Vector for selected signal segment;

(2b) according to 1) formula is to current some X _n2k before point trained, and obtains the dynamic prediction coefficient sets, and k is a prediction order;

(2c) select to adopt dynamic prediction coefficient sets or fixing predictive coefficient group, when 1 according to the thresholding of setting) matrix (C in the formula ^TC) ^-1Determinant when being lower than threshold value, then choose fixedly the predictive coefficient group as predictive coefficient, otherwise, adopt present dynamic predictive coefficient group as predictive coefficient;

(2d) according to the predictive coefficient of determining, utilize As followsThe autoregression formula is tried to achieve current some X _nEstimated value:

$\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|\right|X_{\mathrm{en}}-\underset{t=1}{\overset{k}{\mathrm{\Σ}}}{a}_t{X}_{n-t}\left|\right|+\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\left|\right|X_n-\underset{t=1}{\overset{k}{\mathrm{\Σ}}}{a}_t{X}_{e(n-t)}\left|\right|---\left(2\right)$

Wherein, X _EnBe current some X _nEstimated value, k is a prediction order, the predictive coefficient of a for determining, X _N-tBe X _nThe before t point, X _{E (n-t)}Be X _EnThe before t point;

(2e) repeat (2b)-(2d), obtain the estimated value of whole segment signal, finish prediction ECG signal after the conversion;

(3) original signal value after the conversion and estimated value are subtracted each other, obtain the residual signals of whole section ECG signal;

(4) residual signals is carried out following self-adapting changeable rank RICE coding;

(4a) establish current some e _n, according to e _nThe average of the big or small absolute value of this n point value is tried to achieve in the distribution of numerical values recited of n before point; Setting is with 2 ²Be thresholding, per 2 ²Be single order,, select e according to the thresholding at average place _nThe coding exponent number;

(4b), determine e according to as giving a definition _nPreceding sign indicating number section and exponent number sign indicating number section, wherein k is the exponent number of encoding:

Preceding sign indicating number section: e _nDivide exactly 2 ^k, the gained merchant is n _a, then with n _aIndividual 1 and one 0 as preceding sign indicating number section;

Exponent number sign indicating number section: e _nDivide exactly 2 ^k, the gained remainder is n _e, then with k n _eBinary code as exponent number sign indicating number section;

(4c) preceding sign indicating number section and exponent number sign indicating number are merged, finish e _nCoding;

(4d) coding to residual signals is finished in repetition (4a)-(4c);

(4e) output encoder bit stream is finished the lossless compress to ECG signal.

2. electrocardiogram signal lossless compression method according to claim 1, wherein described preceding sign indicating number section is merged with exponent number sign indicating number section of step (4c) is that exponent number sign indicating number Duan Yuqian sign indicating number section is connected, preceding sign indicating number section pro-, exponent number sign indicating number Duan Zaihou is connected to become e _nWhole sign indicating number section.

3. decompression method based on PT conversion and the bonded ECG signal of linear prediction comprises:

Steps A, according to the bit stream that decoding end is received, decode according to the following procedure:

(1A) 0,1 number of bits of receiving according to decoding end is determined current some e of residual signals _nPreceding sign indicating number section decoding n _aDecoding n with exponent number sign indicating number section _e: i.e. pro-sign indicating number section: use e _nDivide exactly 2 ^k, the gained merchant is n _a, then with n _aIndividual 1 and one 0 as preceding sign indicating number section; In exponent number sign indicating number section: use e _nDivide exactly 2 ^k, the gained remainder is n _e, then with k n _eBinary code as exponent number sign indicating number section;

(2A) determine exponent number k according to the number of bits of the exponent number sign indicating number section that obtains;

(3A) by preceding sign indicating number section decoding n _a, exponent number sign indicating number section decoding n _eWith exponent number k, obtain current some e _n:

e _n＝n _a×2 ^k+n _e；

(4A) repeat (1A)-(3A), obtain whole section residual signals value e _X

Step B is with the residual signals value e that obtains _XEstimated value X with ECG signal after the conversion _eAddition obtains the original signal value X after the PT conversion, that is:

X＝e _X+X _e；

Step C carries out the PT inverse transformation to the ECG signal after the conversion, obtains original electrocardiographicdigital figure signal.

Images