CN105718877B - A kind of signal registration matching process based on dynamic time warping and wavelet character - Google Patents
A kind of signal registration matching process based on dynamic time warping and wavelet character Download PDFInfo
- Publication number
- CN105718877B CN105718877B CN201610033492.3A CN201610033492A CN105718877B CN 105718877 B CN105718877 B CN 105718877B CN 201610033492 A CN201610033492 A CN 201610033492A CN 105718877 B CN105718877 B CN 105718877B
- Authority
- CN
- China
- Prior art keywords
- signal
- wavelet
- data
- noise reduction
- time warping
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of signal registration matching process based on dynamic time warping and wavelet character, belongs to information technology field.The invention includes the following steps: carrying out wavelet decomposition to original signal first, and selected morther wavelet and decomposition level, calculating include the coefficient of wavelet decomposition of noise signal, then are filtered noise reduction to noise signal；The signal of filtering noise reduction is reconstructed, noise reduction data is obtained and is depicted as new signal；New signal is subjected to wavelet transformation, the signal data after progress wavelet transformation is converted to obtain the two-dimensional matrix component of upper different scale small echo of each time；The registration that the corresponding two-dimensional matrix component characterization data of sample carry out character pair with trace to be compared is compared.It is high-efficient that the present invention screens broken end mark information, is greatly improved criminal investigation efficiency, to the pattern-recognition of broken end trace signal with to match work particularly significant, have biggish practical value.
Description
Technical field
The present invention relates to a kind of signal registration matching process based on dynamic time warping and wavelet character, belongs to information
Technical field.
Background technique
In criminal detective's case of cable theft etc., personnel in charge of the case needs to go counter to push away crime work according to broken end mark information
Tool.The trace of broken end is because its data volume is larger and there are biggish randomnesss, if only going to screen one by one by staff
So efficiency is low, and if going auxiliary to complete by computer, it can greatly promote criminal investigation efficiency.
However it is still in initial stage of development at present for the work of the Classical correlation of cable trace laser detection signal, it studies
The people of such algorithm is less, temporarily goes to follow without mature system.Therefore for broken end trace signal pattern-recognition with match work
Just seem particularly significant there is biggish practical value.
Summary of the invention
The present invention provides a kind of signal registration matching process based on dynamic time warping and wavelet character, to be used for
The problem of existing broken end mark information screens inefficiency, temporarily goes examination without mature system.
The present invention is based on the signal registration matching process of dynamic time warping and wavelet character to be achieved in that
The present invention is to carry out Noise reducing of data and signal to real random signal of adopting according to dynamic time warping and wavelet character
The process matched.In Noise reducing of data part, wavelet decomposition is carried out to input data, corresponding noise remove is carried out after decomposition and reconstructs
To waveform signal be exactly noise reduction after data；In Signal Matching part, a Trace Data is set as template data, it is another
A is band matched data, and selected one section of data to be matched do pattern match into template, part the most matched found, by not
Data with section select after being matched, and it is maximum as final result to select similarity in these matching results
Original signal is subjected to multilayer by small echo first and decomposes the most of noise data of removal, according to accuracy of data acquisition
Difference use the small echo and different types of morther wavelet of different levels, abnormal number is then completed on the basis of wavelet de-noising again
According to amendment, finally according to the Dynamic Programming Equation of similarity carry out Signal Matching, complete registration match work.
Specific step is as follows for the signal registration matching process based on dynamic time warping and wavelet character:
Step1, wavelet decomposition, and selected morther wavelet and decomposition level are carried out to original signal, calculate and believe comprising noise
Number coefficient of wavelet decomposition, then noise reduction is filtered to noise signal；
Step2, the signal of filtering noise reduction is reconstructed, obtains noise reduction data and is depicted as new signal；
Step3, new signal is subjected to wavelet transformation, converts the signal data after progress wavelet transformation to obtain each time
The two-dimensional matrix component of upper different scale small echo；
Step4, the weight that the corresponding two-dimensional matrix component characterization data of sample are carried out to character pair with trace to be compared
Right comparison.
In the step Step1, specific steps are as follows:
Step1.1, original signal f (t) is decomposed into two parts according to following formula:
Wherein: anFor the approximation of n-th layer, diFor i-th layer of detail data, f is original signal data；
Step1.2, a threshold value r is selected to every layer of coefficient, HF noise signal is handled according to threshold value r,
Treatment process is as follows:
Wherein: ciIndicate i-th of decomposition wavelet coefficient；
Step1.3, the threshold values r by setting before, are filtered noise reduction to noise signal, the signal after noise reduction are as follows:
Wherein: anFor the approximation of n-th layer, d 'iFor i-th layer of the detail data after threshold deniosing, after f ' is noise reduction
Trace Data.
In the step Step4, the path of best match is found by the way of dynamic time warping, i.e., finally in purpose
It is located in obtain maximum similarity value.
In the step Step4, specific step is as follows for registration comparison:
Step4.1, sample is set as S={ s1,s2,…,sn, if the trace of input is T={ t1,t2,…,tm, then structure
Element a (i, j) in the two-dimensional matrix A of the N*M size built, two-dimensional matrix indicates sample S={ s1,s2,…,snIn
I-th of element and T={ t1,t2,…,tmIn j-th of element distance；
Step4.2, after having obtained the similarity degree between a point, best is found by the way of dynamic time warping
The path matched, that is, maximum similarity value sims (n, m) is finally obtained at destination (n, m)；
Wherein: the Dynamic Programming Equation dynamic time warping of similarity are as follows:
The beneficial effects of the present invention are:
1, the present invention has more preferable more accurate descriptive power to fast-changing signal using wavelet transformation.
2, the present invention decompose to broken end trace using small echo can be improved effective percentage.
3, the present invention can describe that sequence length is different or X-axis can not be complete with the Time alignment function met certain condition
The signal of alignment；
4, the method examination broken end mark information that the present invention uses is high-efficient, criminal investigation efficiency is greatly improved, to broken end
The pattern-recognition of trace signal with to match work particularly significant, have biggish practical value.
Detailed description of the invention
Fig. 1 is the information exploded view of Noise reducing of data of the present invention；
Fig. 2 is each level approximate signal figure of the present invention；
Fig. 3 is each layer detailed information figure of the present invention；
Fig. 4 is that threshold values of the present invention selects A figure；
Fig. 5 is that threshold values of the present invention selects B figure；
Fig. 6 is noise reduction result figure of the present invention；
Fig. 7 is difference matching figure of the present invention；
Fig. 8 is that dynamic time warping of the present invention matches schematic diagram；
Fig. 9 is flow chart of the method for the present invention.
Specific embodiment
Embodiment 1: as shown in figs 1-9, a kind of signal registration match party based on dynamic time warping and wavelet character
Method；
Specific step is as follows for the signal registration matching process based on dynamic time warping and wavelet character:
Step1, wavelet decomposition, and selected morther wavelet and decomposition level are carried out to original signal, calculate and believe comprising noise
Number coefficient of wavelet decomposition, then noise reduction is filtered to noise signal；
Step2, the signal of filtering noise reduction is reconstructed, obtains noise reduction data and is depicted as new signal；
Step3, new signal is subjected to wavelet transformation, converts the signal data after progress wavelet transformation to obtain each time
The two-dimensional matrix component of upper different scale small echo；
Step4, the weight that the corresponding two-dimensional matrix component characterization data of sample are carried out to character pair with trace to be compared
Right comparison.
In the step Step1, specific steps are as follows:
Step1.1, original signal f (t) is decomposed into two parts according to following formula:
Wherein: anFor the approximation of n-th layer, diFor i-th layer of detail data, f is original signal data；
Step1.2, a threshold value r is selected to every layer of coefficient, HF noise signal is handled according to threshold value r,
Treatment process is as follows:
Wherein: ciIndicate i-th of decomposition wavelet coefficient；
Step1.3, the threshold values r by setting before, are filtered noise reduction to noise signal, the signal after noise reduction are as follows:
Wherein: anFor the approximation of n-th layer, d 'iFor i-th layer of the detail data after threshold deniosing, after f ' is noise reduction
Trace Data.
In the step Step4, the path of best match is found by the way of dynamic time warping, i.e., finally in purpose
It is located in obtain maximum similarity value.
In the step Step4, specific step is as follows for registration comparison:
Step4.1, sample is set as S={ s1,s2,…,sn, if the trace of input is T={ t1,t2,…,tm, then structure
Element a (i, j) in the two-dimensional matrix A of the N*M size built, two-dimensional matrix indicates sample S={ s1,s2,…,snIn
I-th of element and T={ t1,t2,…,tmIn j-th of element distance；
Step4.2, after having obtained the similarity degree between a point, best is found by the way of dynamic time warping
The path matched, that is, maximum similarity value sims (n, m) is finally obtained at destination (n, m)；
Wherein: the Dynamic Programming Equation dynamic time warping of similarity are as follows:
Embodiment 2: as shown in figs 1-9, a kind of signal registration match party based on dynamic time warping and wavelet character
Method；Specific step is as follows for the method:
Step1, wavelet decomposition, and selected morther wavelet and decomposition level are carried out to original signal f (t), calculate comprising making an uproar
The coefficient of wavelet decomposition of acoustical signal, then noise reduction is filtered to noise signal；
The number of plies is decomposed more, and to the more of detail data processing, the noise that can be eliminated is more, but may smooth out
Details is also more.Therefore the Decomposition order of a balance is found；It is general to be decomposed into two using by original signal in decomposable process
Part, it is assumed that the decomposition of n-layer is carried out, then the composition of original signal can be described below:
Wherein: anFor the approximation of n-th layer, diFor i-th layer of detail data, f is initial data；
As shown in Figs. 1-2, (I, 0) is expressed as in figure；diFor i-th layer of detail data, being expressed as in figure (i,
1)；F is initial data, is expressed as (0,0) in figure；
It is illustrated in figure 3 three layers of details expanded view, according to the approximate data obtained in upper figure to example signal decomposition
With detail data it has been discovered that for current tool scans, the approximate data that the expansion of level 1 obtains still has
More burr, the approximate signal of level 2 is unfolded that great improvement has been obtained for the expansion compared to level 1, big absolutely
Partial burr is improved, and the approximate signal of level 3 then further improves this phenomenon, anti-from obtained figure
From the point of view of feedback, in close proximity to desired pattern.Therefore, under normal circumstances, also it is to proceed to the expansion of level 3.That
According to the formula of noise reduction
Wherein: ciIndicate i-th of decomposition wavelet coefficient
A threshold value is set, the data under threshold value are all removed.Because general noise is all irregular with high frequency
Noise exist, it is all lesser for being converted into the value of detail section.Therefore a threshold value is set to each details, be filtered
Noise reduction:
Wherein: d 'iFor the detail data after threshold deniosing；F ' is the Trace Data after noise reduction.
Step2, the signal of filtering noise reduction is reconstructed, obtains noise reduction data and is depicted as new signal；
As illustrated in figures 4-5, it is the threshold value exemplary diagram of a selection noise reduction details, has carried out the Wavelet Expansions of three-level in the middle,
The details that centre is mingled with just removes, as noise remove.Threshold value selects bigger, then noise reduction effect is more obvious, and wherein Fig. 4 is one
A standard selection, Fig. 5 are the result of a raising threshold value.
As shown in fig. 6, above for needing the data of noise reduction to carry out different configurations in each layer details, for configuration
The threshold range of A selection is smaller, and the scale handled is also low, should theoretically retain in greater detail.And configure B's
Threshold value selection is then bigger some, should be theoretically the noise for eliminating the overwhelming majority.It has carried out as we can see from the figure not
With configuration, the effect of noise reduction be it is differentiated, configure A and configure B in all had been removed the overwhelming majority shake and
Noise, but A is configured because noise reduction configuration dynamics is smaller, at 600 positions, still remain many noise datas, hence it is evident that be not so good as
Configure the excellent noise reduction effect of B.
Either configuration A still configures B, and the signal relative to non-noise reduction has improvement.It is best in actual application
An equilibrium and selection are exactly done according to service condition and trace signal quality so that noise remove is maximized simultaneously can
Guarantee not losing for details.
Step3, wavelet transformation is carried out to noise reduction data, converts the signal data after progress wavelet transformation to obtain each time
The two-dimensional matrix component of upper different scale small echo, as shown in Figure 8；
After carrying out dynamic time warping calculating for a wavelet character similar matrix using two trace signals, one is found
The schematic diagram of a Optimum Matching.In the diagram, multiple matching is then represented if it is vertical bar or horizontal line, and if
It is that oblique line then represents is completely newly to match.
Wherein the Dynamic Programming Equation of similarity is:
Its core concept is that dynamic local is optimal (greedy method), and is completely newly matched higher weight.In dynamic time
In regular, some position in some position either in sample, or input trace, as long as multiple with other side
Position matches, then give 1/2 be both that matched weight occurs for the first time, this ensure that can permit two
In the case that a signal length is inconsistent, accomplish optimal matching.
Step4, the weight that the corresponding two-dimensional matrix component characterization data of sample are carried out to character pair with trace to be compared
Right comparison, as shown in Figure 7；
It is the matching result between two trace signals, there is biggish similarity degree, warp between trace 16 and trace 17
Signal Matching algorithm is crossed, the two has obtained effective alignment matching.
As shown in table 1 below, the data result tested using wavelet character method, wherein success rate is 80%, failure rate
It is 16.67%, obscures rate 6.367%, wave characteristic method has preferable recognition effect, obtains effective result and infers.
1 test result of table
| Input data | 1# | Similarity | 2# | Similarity | 3# | Similarity | 4# | Similarity | 5# | Similarity | As a result |
| 16 | 17 | 0.90 | 18 | 0.74 | 23 | 0.68 | 22 | 0.66 | 24 | 0.65 | Correctly |
| 17 | 16 | 0.90 | 18 | 0.76 | 23 | 0.68 | 24 | 0.66 | 22 | 0.66 | Correctly |
| 18 | 20 | 0.78 | 19 | 0.78 | 17 | 0.76 | 16 | 0.74 | 24 | 0.65 | Correctly |
| 19 | 20 | 0.79 | 18 | 0.78 | 21 | 0.64 | 17 | 0.60 | 23 | 0.60 | Correctly |
| 20 | 21 | 0.79 | 19 | 0.79 | 18 | 0.78 | 23 | 0.66 | 44 | 0.66 | Correctly |
| 21 | 20 | 0.79 | 42 | 0.68 | 43 | 0.68 | 40 | 0.66 | 39 | 0.66 | Failure |
| 22 | 24 | 0.67 | 17 | 0.66 | 16 | 0.66 | 23 | 0.66 | 40 | 0.65 | Correctly |
| 23 | 17 | 0.68 | 16 | 0.68 | 20 | 0.66 | 22 | 0.66 | 24 | 0.65 | Correctly |
| 24 | 40 | 0.69 | 39 | 0.68 | 44 | 0.68 | 42 | 0.67 | 45 | 0.67 | Failure |
| 25 | 19 | 0.35 | 39 | 0.31 | 43 | 0.31 | 40 | 0.31 | 44 | 0.31 | Failure |
| 26 | 27 | 0.99 | 28 | 0.83 | 33 | 0.57 | 34 | 0.56 | 29 | 0.55 | Correctly |
| 27 | 26 | 0.99 | 28 | 0.83 | 33 | 0.57 | 34 | 0.56 | 29 | 0.55 | Correctly |
| 28 | 27 | 0.83 | 26 | 0.83 | 33 | 0.55 | 34 | 0.54 | 29 | 0.54 | Correctly |
| 29 | 30 | 0.89 | 26 | 0.55 | 27 | 0.55 | 33 | 0.54 | 28 | 0.54 | Correctly |
| 30 | 29 | 0.89 | 33 | 0.55 | 27 | 0.54 | 26 | 0.54 | 28 | 0.54 | Correctly |
| 31 | 32 | 0.83 | 35 | 0.55 | 36 | 0.55 | 34 | 0.53 | 33 | 0.53 | Correctly |
| 32 | 31 | 0.83 | 35 | 0.52 | 36 | 0.52 | 34 | 0.51 | 33 | 0.51 | Correctly |
| 33 | 34 | 0.93 | 37 | 0.62 | 38 | 0.62 | 16 | 0.60 | 17 | 0.60 | Correctly |
| 34 | 33 | 0.93 | 37 | 0.63 | 38 | 0.62 | 16 | 0.61 | 17 | 0.60 | Correctly |
| 35 | 36 | 0.89 | 21 | 0.59 | 33 | 0.57 | 18 | 0.57 | 34 | 0.57 | It is fuzzy |
| 36 | 35 | 0.89 | 21 | 0.58 | 34 | 0.57 | 33 | 0.57 | 37 | 0.57 | Correctly |
| 37 | 38 | 0.94 | 16 | 0.64 | 21 | 0.63 | 34 | 0.63 | 17 | 0.62 | It is fuzzy |
| 38 | 37 | 0.94 | 16 | 0.64 | 17 | 0.62 | 21 | 0.62 | 18 | 0.62 | Failure |
| 39 | 40 | 0.94 | 43 | 0.75 | 42 | 0.74 | 44 | 0.69 | 24 | 0.68 | Correctly |
| 40 | 39 | 0.94 | 42 | 0.76 | 43 | 0.75 | 24 | 0.69 | 44 | 0.69 | Correctly |
| 41 | 45 | 0.64 | 44 | 0.64 | 40 | 0.63 | 24 | 0.63 | 39 | 0.63 | Correctly |
| 42 | 43 | 0.93 | 40 | 0.76 | 39 | 0.74 | 21 | 0.68 | 24 | 0.67 | Correctly |
| 43 | 42 | 0.93 | 40 | 0.75 | 39 | 0.75 | 21 | 0.68 | 24 | 0.67 | Correctly |
| 44 | 45 | 0.96 | 40 | 0.69 | 39 | 0.69 | 24 | 0.68 | 42 | 0.67 | Correctly |
| 45 | 44 | 0.96 | 40 | 0.68 | 39 | 0.68 | 24 | 0.67 | 42 | 0.67 | Correctly |
Above in conjunction with attached drawing, the embodiment of the present invention is explained in detail, but the present invention is not limited to above-mentioned
Embodiment within the knowledge of a person skilled in the art can also be before not departing from present inventive concept
Put that various changes can be made.
Claims (3)
1. a kind of signal registration matching process based on dynamic time warping and wavelet character, it is characterised in that: described to be based on
Specific step is as follows for the signal registration matching process of dynamic time warping and wavelet character:
Step1, wavelet decomposition, and selected morther wavelet and decomposition level are carried out to original signal, calculate comprising noise signal
Coefficient of wavelet decomposition, then noise reduction is filtered to noise signal；
Step2, the signal of filtering noise reduction is reconstructed, obtains noise reduction data and is depicted as new signal；
Step3, new signal is subjected to wavelet transformation, the signal data after progress wavelet transformation is converted to obtain on each time not
With the two-dimensional matrix component of multi-scale wavelet；
Step4, the registration that the corresponding two-dimensional matrix component characterization data of sample are carried out to character pair with trace to be compared
It compares；
In the step Step4, specific step is as follows for registration comparison:
Step4.1, sample is set as S={ s1,s2,...,sn, if the trace of input is T={ t1,t2,...,tm, then constructing
The two-dimensional matrix A of a N*M size, the element a (i, j) in two-dimensional matrix indicate sample S={ s1,s2,...,snIn the
I element and T={ t1,t2,...,tmIn j-th of element distance；
Step4.2, after having obtained the similarity degree between each point, best match is found by the way of dynamic time warping
Path, that is, maximum similarity value sims (n, m) is finally obtained at destination (n, m)；
Wherein: the Dynamic Programming Equation dynamic time warping of similarity are as follows:
2. the signal registration matching process according to claim 1 based on dynamic time warping and wavelet character, special
Sign is: in the step Step1, specific steps are as follows:
Step1.1, original signal f (t) is decomposed into two parts according to following formula:
Wherein: anFor the approximation of n-th layer, diFor i-th layer of detail data, f is original signal data；
Step1.2, a threshold value r is selected to every layer of coefficient, HF noise signal is handled according to threshold value r, handled
Process is as follows:
Wherein: ciIndicate i-th of coefficient of wavelet decomposition；
Step1.3, the threshold values r by setting before, are filtered noise reduction to noise signal, the signal after noise reduction are as follows:
Wherein: anFor the approximation of n-th layer, di' for i-th layer of detail data after threshold deniosing, f' is the trace after noise reduction
Mark data.
3. the signal registration matching process according to claim 1 based on dynamic time warping and wavelet character, special
Sign is: in the step Step4, the path of best match is found by the way of dynamic time warping, i.e., finally in purpose
It is located in obtain maximum similarity value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610033492.3A CN105718877B (en) | 2016-01-19 | 2016-01-19 | A kind of signal registration matching process based on dynamic time warping and wavelet character |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610033492.3A CN105718877B (en) | 2016-01-19 | 2016-01-19 | A kind of signal registration matching process based on dynamic time warping and wavelet character |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105718877A CN105718877A (en) | 2016-06-29 |
| CN105718877B true CN105718877B (en) | 2019-02-05 |
Family
ID=56147653
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610033492.3A Active CN105718877B (en) | 2016-01-19 | 2016-01-19 | A kind of signal registration matching process based on dynamic time warping and wavelet character |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105718877B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447017A (en) * | 2018-11-07 | 2019-03-08 | 中国科学院大气物理研究所 | A kind of method of automatic discrimination lightning electromagnetic pulse type |
| CN112054806B (en) * | 2020-09-15 | 2023-07-25 | 西南交通大学 | Subway sliding plug door data compression method based on two-dimensional self-adaptive threshold wavelet |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188107A (en) * | 2007-09-28 | 2008-05-28 | 中国民航大学 | A voice recognition method based on wavelet decomposition and mixed Gauss model estimation |
| CN101832471A (en) * | 2010-04-19 | 2010-09-15 | 哈尔滨工程大学 | Signal identification and classification method |
| CN102542262A (en) * | 2012-01-04 | 2012-07-04 | 东南大学 | Waveform identification method based on operating-characteristic working condition waveform library of high-speed rail |
-
2016
- 2016-01-19 CN CN201610033492.3A patent/CN105718877B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188107A (en) * | 2007-09-28 | 2008-05-28 | 中国民航大学 | A voice recognition method based on wavelet decomposition and mixed Gauss model estimation |
| CN101832471A (en) * | 2010-04-19 | 2010-09-15 | 哈尔滨工程大学 | Signal identification and classification method |
| CN102542262A (en) * | 2012-01-04 | 2012-07-04 | 东南大学 | Waveform identification method based on operating-characteristic working condition waveform library of high-speed rail |
Non-Patent Citations (4)
| Title |
|---|
| 基于小波变换和DTW算法的ECG身份识别系统研究;郭豹;《中国优秀硕士学位论文全文数据库》;20110615(第2011年06期);全文 |
| 基于局部小波能量工具痕迹识别的研究;杨敏, 李冬云, 王纬东;《中国人民公安大学学报(自然科学版)》;20080515(第2008年02期);全文 |
| 基于能量特征匹配的在线签名验证算法研究;徐陶祎;《中国优秀硕士学位论文全文数据库》;20070415(第2007年04期);全文 |
| 基于跳变点能量的DTW匹配在线手写签名验证算法;徐陶祎, 郑建彬;《计算机应用研究》;20070801(第2007年08期);正文第1页右栏倒数第2段至第3页左栏第5段 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105718877A (en) | 2016-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020100027A4 (en) | Electroencephalogram-based negative emotion recognition method and system for aggressive behavior prediction | |
| US9345413B2 (en) | Heart rate extraction using neural wavelet adaptive gain control and neural pattern processing | |
| DE102004046237A1 (en) | Method and system for mixed rigid registration of medical 2D / 3D images | |
| CN106446952A (en) | Method and apparatus for recognizing score image | |
| Euán et al. | The hierarchical spectral merger algorithm: A new time series clustering procedure | |
| CN101930598A (en) | Natural image denoising method based on non-local mean value of shearlet region | |
| CN102722892A (en) | SAR (synthetic aperture radar) image change detection method based on low-rank matrix factorization | |
| Al-Salman et al. | Detection of EEG K-complexes using fractal dimension of time frequency images technique coupled with undirected graph features | |
| CN115100421B (en) | Antagonistic sample generation method based on image frequency domain decomposition reconstruction | |
| CN104008537A (en) | Novel noise image fusion method based on CS-CT-CHMM | |
| CN105718877B (en) | A kind of signal registration matching process based on dynamic time warping and wavelet character | |
| CN111222442A (en) | Electromagnetic signal classification method and device | |
| Abdolmaleki et al. | Selecting optimum base wavelet for extracting spectral alteration features associated with porphyry copper mineralization using hyperspectral images | |
| CN103886553A (en) | Method and system for non-local average value denoising of image | |
| CN104580829A (en) | Terahertz image enhancing method and system | |
| CN106709997A (en) | Three-dimensional key point detection method based on deep neural network and sparse auto-encoder | |
| Al-Khuzaay et al. | New Proposed Mixed Transforms: CAW and FAW and Their Application in Medical Image Classification | |
| CN106971377A (en) | A kind of removing rain based on single image method decomposed based on sparse and low-rank matrix | |
| Roald et al. | Tracing network evolution using the PARAFAC2 model | |
| Truchetet et al. | Review of industrial applications of wavelet and multiresolution-based signal and image processing | |
| CN104680176B (en) | A kind of brain wave (EEG) Modulation recognition method based on the selection of non-gaussian neutrality vector characteristics | |
| Long et al. | Contourlet spectral histogram for texture classification | |
| CN105674887B (en) | A kind of incomplete linear trace feature registration comparison method in scene | |
| Wang et al. | An improved Iris recognition method based on wavelet packet transform | |
| Hong‐jun et al. | Image denoising method based on grey relational threshold |
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 |