CN106504214B - The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion - Google Patents
The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion Download PDFInfo
- Publication number
- CN106504214B CN106504214B CN201610985086.7A CN201610985086A CN106504214B CN 106504214 B CN106504214 B CN 106504214B CN 201610985086 A CN201610985086 A CN 201610985086A CN 106504214 B CN106504214 B CN 106504214B
- Authority
- CN
- China
- Prior art keywords
- image
- banded improvement
- wavelet
- band
- high spectrum
- 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
- 230000006872 improvement Effects 0.000 title claims abstract description 95
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000001228 spectrum Methods 0.000 title claims abstract description 55
- 230000009466 transformation Effects 0.000 title claims abstract description 26
- 230000004927 fusion Effects 0.000 title claims abstract description 23
- 238000012545 processing Methods 0.000 claims description 39
- 238000000354 decomposition reaction Methods 0.000 claims description 36
- 238000012937 correction Methods 0.000 claims description 8
- 238000013178 mathematical model Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 239000000523 sample Substances 0.000 claims 1
- 230000008030 elimination Effects 0.000 abstract description 5
- 238000003379 elimination reaction Methods 0.000 abstract description 5
- 230000001788 irregular Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 16
- 238000001914 filtration Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 4
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- FPIPGXGPPPQFEQ-OVSJKPMPSA-N all-trans-retinol Chemical compound OC\C=C(/C)\C=C\C=C(/C)\C=C\C1=C(C)CCCC1(C)C FPIPGXGPPPQFEQ-OVSJKPMPSA-N 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000004321 preservation Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 230000017105 transposition Effects 0.000 description 2
- 235000019169 all-trans-retinol Nutrition 0.000 description 1
- 239000011717 all-trans-retinol Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion, first input hyper-spectral image data cube, then high spectrum image wave band data is judged, reject the high spectrum image wave band data of damage, read the hyperspectral image data of normal kth wave band, respectively with two-dimentional stationary wavelet, the Banded improvement that local interpolation carries out hyperspectral image data filters out, then fusion treatment is carried out, the present invention can not only effectively eliminate general thin Banded improvement, and it can play the role of elimination to the Banded improvement in broadband and ultra wide band and irregular distribution, the some details and edge geological information in image can also preferably be kept, it also can apply to different high spectrum images simultaneously, there is preferable universality.
Description
Technical field
The invention belongs to EO-1 hyperion Signal and Information Processing technical fields, and in particular to wavelet transformation and local interpolation fusion
High spectrum image Banded improvement removing method.
Background technique
Banded improvement is prevalent in high-spectrum remote sensing, brings great challenge to the application of high spectrum image,
Effectively removing Banded improvement is the premise and basis that high spectrum image is applied very well.Strip noise removal is related in high spectrum image
To the preservation of ground object target details and marginal information, they are the relationships of conflict.Banded improvement disappears in high spectrum image
Except usually there are two types of thinkings, i.e., hardware is eliminated and software is eliminated.If eliminated from the angle of hardware, need to change existing
The imaging mechanism of sensor grid array eliminates Banded improvement from root.But this technical difficulty is big, it can not in the short time
It completes, and process costs are very high.So a possibility that thoroughly eliminating Banded improvement from the angle of hardware in a short time is not
Greatly, the elimination of Banded improvement is concentrated mainly on software to realize.From the angles of high-spectrum remote sensing data itself into
Row processing is known as software null method come the method for eliminating Banded improvement.For the different application mesh of high-spectrum remote sensing data
, different scholars proposes many processing methods, these methods can probably be generalized into four seed types.First seed type is to pass
The statistical match method of system, typical represent is histogram matching and match by moment method, and such methods are easy to accomplish, also not multiple
Miscellaneous, operation time is short, but has higher requirement to high spectrum image, it is desirable that the distribution of ground object target is more equal in high spectrum image
Even, what is otherwise handled is ineffective, therefore, the high-spectrum remote sensing big for atural object distributional difference, both traditional sides
Method has been difficult to obtain the satisfied effect that disappears, but on the basis of them, is but suggested and applies there are many improved method；
Second class is the filtering method based on spatial domain, such method is fairly simple, easy to accomplish, but single filter in spatial domain side
Method, which filters out aspect in high spectrum image Banded improvement, has seemed helpless, because Strip noise removal effect is not only less managed
Think, and residual noise is relatively more, thus single filtering algorithm oneself through being rarely employed；Third class is the filtering side based on transform domain
Method, this kind of typical method have fourier transform method and Wavelet Transform etc., since Fourier Transform Filtering method and others are low
Pass filter is the same, and Banded improvement is often difficult to find so a suitable frequency content for item as periodic signal processing
Band noise and signal are kept completely separate, and such methods cannot not only completely remove Banded improvement, but also calculate complexity, and performance is not
Stablize, while also will cause the loss of signal, keep image detail fuzzy, wavelet transformation theory mainly has for image filtering processing
Two ways carries out, i.e. soft-threshold de-noising and hard -threshold filtering, both modes are also undesirable for Banded improvement effect, then
There is scholar to propose to handle using the thinking that coefficient of wavelet decomposition is zeroed, for thin Banded improvement, effect is also possible that
But for wide Banded improvement, filter effect cannot be satisfactory；4th seed type is based on some new theories and side
Method, such as be recently proposed Theory of Variational Principles, phase equalization, regular low-rank representation method, every kind of method is usually constructed with its needle
To property, Banded improvement can be removed in a certain range, but their universality and versatility are poor, this is because different
High spectrum image there is different Banded improvement features, such as bright wisp band, gray bars and filaments of sun band, slice band, wide band
With ultra-wide band etc., and the position occurred is also not fixed, so, if to effectively eliminate the influence of noise, it is necessary to
First the mechanism of production and distribution characteristics of Banded improvement in high spectrum image are analyzed and studied, otherwise filter effect is difficult to reach
To target.
By being analysed in depth and being studied to existing method, although finding that these methods can remove high-spectrum remote-sensing figure
Banded improvement as in, mainly for thin Banded improvement and relatively uniform Banded improvement, and removal effect and not bery manage
Think, the clarity of image is not very high sometimes, while some detailed information in image are also often lost.For example, small echo becomes
Use the removal in Banded improvement instead, whether traditional threshold method or decomposition coefficient return-to-zero method, all can blurred picture it is some
Geometric detail information can achieve the purpose that Banded improvement certainly.The band random for broadband and ultra wide band and distribution is made an uproar
Sound has been difficult to obtain satisfied result with traditional single filtering method.
Summary of the invention
In order to overcome the disadvantages of the above prior art, the purpose of the present invention is to provide wavelet transformations and local interpolation to merge
High spectrum image Banded improvement removing method, can not only effectively eliminate general thin Banded improvement, and to broadband and super
The Banded improvement of broadband and irregular distribution can play the role of elimination, moreover it is possible to preferably holding image in some details and
Edge geological information, while also can apply to different high spectrum images, there is preferable universality.
In order to achieve the above object, the technical scheme adopted by the invention is as follows:
The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion, comprising the following steps:
Step 1: input hyper-spectral image data cube, hyper-spectral image data cube are to pass through pretreated 2 grades
Product data, pretreatment include Noise Filter, radiant correction and geometric correction processing；
Step 2: judge high spectrum image wave band data: hyperspectral image data is in acquisition process by various
The interference of factor can make the quality of subband data be damaged, and the band image of damage is just not necessarily to continue to handle；
Step 3: rejecting the high spectrum image wave band data of damage；
Step 4: reading the hyperspectral image data of normal kth wave band: being selected in remaining hyperspectral image data
Kth wave band data carries out step 5 and step 6 operation respectively；
Step 5: the Strip noise removal of hyperspectral image data is carried out with two dimension stationary wavelet；
Step 6: being filtered out with the Banded improvement that local interpolation carries out hyperspectral image data；
Step 7: fusion treatment is carried out to the processing result of step 5 and step 6；
Step 8: whether the K-band for judging processing is that last wave band then terminates if it is last wave band；If not
Last wave band is then transferred to step 4, that is, reads the image of next wave band.
The method that high spectrum image wave band data judges in the step 2 are as follows:
Using the product δ of the quadratic sum of the gradient (GMC) of the quadratic sum and column average value of the derivative (DMC) of column average value
(k) judged；
δ (k)=D (k) × G (k) (1)
The mathematical model of the quadratic sum of the gradient (GMC) of the quadratic sum and column average value of the derivative (DMC) of column average value is such as
Under:
Wherein, k indicates that kth wave band, D (k) and G (k) respectively indicate the derivative of the derivative (DMC) of kth wave band column average value
The quadratic sum of the gradient (GMC) of quadratic sum and column average value, μi(k) andThe i-th column for respectively indicating kth band image are visited
The average value and gradient value of unit are surveyed, δ (k) indicates their product.
The method of the high spectrum image wave band data of damage is rejected in the step 3 are as follows: if δ (k) value is than scheduled
Threshold epsilon is big, i.e. δ (k) > ε, shows that the image data of the wave band has been damaged, directly deletes the data of the wave band.
The specific steps of Strip noise removal are carried out in the step 5 with two dimension stationary wavelet are as follows:
Step 5.1: bright wisp band first the conversion of Banded improvement type: being transformed into filaments of sun band；
Step 5.2: setting wavelet decomposition scales number: the scale parameter that two-dimentional stationary wavelet decomposes is set as 5, i.e. M=5；
Step 5.3: selecting wavelet function: selecting basic function of Daubechies (db1) function as wavelet decomposition, then
The K-band image of selection is decomposed；
Step 5.4: image is decomposed: the parameter being arranged using step 5.2 and step 5.3, to the EO-1 hyperion k of input
Band image carries out wavelet decomposition；
Step 5.5: the decomposition coefficient of scale each after wavelet decomposition being handled: first determining whether coefficient subgraph whether also
Step 5.2 is gone to if the Banded improvement for including in coefficient subgraph is more comprising a large amount of Banded improvement, to the coefficient
Continue to decompose in image, until stopping carrying out wavelet decomposition when some decomposition coefficient subgraph is almost Banded improvement；
Then the coefficient value that last decomposition layer obtains all is carried out assigning zero processing；Finally small echo inversion is carried out by scale by coefficient
It changes, finally obtains the wavelet coefficient of each decomposition scale；
Step 5.6: being handled, obtained with small with the wavelet inverse transformation that the wavelet coefficient that step 5.5 obtains carries out whole image
Wave conversion handles the image after Banded improvement.
The specific steps of local interpolation algorithm removal Banded improvement in the step 6 are as follows:
Step 6.1: calculating the column mean of high spectrum image, and obtain column mean curve graph；
Step 6.2: calculating the gradient of column mean, and obtain the gradient curve figure of column mean；
Step 6.3: by gradient curve figure, obtaining the position where wave crest and trough, here it is where Banded improvement
Position；
Step 6.4: the column where Banded improvement being removed, interpolation processing is then carried out, is as a result that of obtaining part
The image of interpolation；
Step 6.5: carry out match by moment processing: formula (4) is exactly the mathematical model of match by moment,
Wherein, k indicates kth band image in high spectrum image, Xk(i, j) and Yk(i, j) respectively indicates kth band image
Single pixel gray value adjustment front and back value, μkrAnd σkrRespectively indicate the mean value and standard deviation of entire image, μkjAnd σkjRespectively
The mean value and standard deviation of j column.
The specific steps of fusion treatment are carried out in the step 7 to the processing result of step 5 and step 6 are as follows:
Fusion rule such as following formula:
If(k)=a × Iwt(k)+b×Ili(k) (5)
K in formula indicates kth wave band, just indicates fused image, Iwt(k) after indicating that wavelet transformation removes Banded improvement
Image, Iwt(k) indicating that regional area interpolation removes the image after Banded improvement, parameter a and b are respectively respective weight coefficient, they
Value range be [0,1], and a+b=1.
The advantages of the present invention over the prior art are that:
(1) it the advantages of method designs: in processing method, devises remove Banded improvement twice first, still, this is twice
The removal of the Banded improvement not instead of cascade relationship in front and back, relationship arranged side by side.Wavelet transformation has multiple dimensioned multidirectional spy
Point, and Banded improvement has good directionality, therefore wavelet transformation can effectively extract Banded improvement, but it also eliminates simultaneously
Many detailed information.It by the local location where searching noise, is then removed, then carries out local interpolation, protect in this way
Other detailed information are protected.The result of the two is merged again, achieved the purpose that not only to remove band but also protects details.Its
Secondary, when carrying out high spectrum image decomposition with small echo, selection is two-dimentional stationary wavelet, rather than 2-d discrete wavelet.Because two
Dimension discrete wavelet is lower 2 using decomposing, and scale is every to increase by 1, and the coefficient subgraph after decomposition becomes the 1/2 of upper layer, is not easy to pair
Image is handled.It is decomposed using two-dimentional stationary wavelet, the subgraph on each scale, the subgraph after each subgraph decomposition
Picture all keeps same size, in this way convenient for processing and wavelet inverse transformation with original image.
(2) advantage in application: Banded improvement is that high light spectrum image-forming mechanism generates, and is difficult to eliminate on hardware and technique,
Therefore it can only be removed using the method for data processing.General method is for different application targets, and adaptability is non-
It is often limited.This method has stronger universality, can be used for the elimination of the high spectrum image Banded improvement of different sensors, Er Qieke
Slice band, wide band, ultra-wide band or gray bars, filaments of sun band and bright wisp band are removed simultaneously.Item can not only be effectively eliminated
Band, moreover it is possible to preferably holding detailed information.
(3) the advantages of data processing: high-spectrum seems data cube, and data volume is very big, in order to improve processing speed
Degree, reduces the unnecessary processing time, processing is optimized to data in this method.First, it is determined which data be it is useful,
Which data has been damaged, and rejects to the data image of damage, and the data image then retained enters in next step
Processing, saves the time to bad data image procossing in this way.In addition, to extract Banded improvement, this method convenient for wavelet decomposition
Banded improvement type is handled, is conducive to improve the effect for removing band, such as bright wisp band becomes filaments of sun band.Third, most
Fusion treatment has been carried out to the processing result of different thinkings afterwards, has reached processing result and keeps effectively filtering out while detailed information
Banded improvement.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 (a) is the original hyperspectral image data of embodiment, and Fig. 2 (b), Fig. 2 (c), Fig. 2 (d) are respectively D (k), G
(k) and the curve graph of δ (k) value, abscissa are wave band number, and ordinate is numerical value.
Fig. 3 is the original high spectrum image of embodiment, wherein figure a is 6 wave bands, figure b is 23 wave bands, and figure c is 27 wave bands, figure
D is 52 wave bands.
Fig. 4 be using two-dimentional Stationary Wavelet Transform to high spectrum image remove band after as a result, wherein figure a be 6 wave bands,
Figure b is 23 wave bands, and figure c is 27 wave bands, and figure d is 52 wave bands.
Fig. 5 is the gradient curve figure of Fig. 3 (b).
Fig. 6 is local interpolation algorithm treatment process as a result, wherein figure a is direct interpolation, and figure b is match by moment.
Fig. 7 is local interpolation algorithm de-noising as a result, wherein figure a is 6 wave bands, and figure b is 23 wave bands, and figure c is 27 wave bands, and figure d is
52 wave bands.
Fig. 8 is match by moment de-noising as a result, wherein figure a is 6 wave bands, and figure b is 23 wave bands, and figure c is 27 wave bands, and figure d is 52 waves
Section.
Fig. 9 is the method for the present invention de-noising as a result, wherein figure a is 6 wave bands, and figure b is 23 wave bands, and figure c is 27 wave bands, and figure d is
52 wave bands.
Specific embodiment
The present invention is further elaborated with reference to the accompanying drawings and examples.
The image interception size of the present embodiment be 512 × 486, spatial resolution 20m, spectral resolution 20nm, at
As wave-length coverage is 1100-2500nm.
Referring to Fig.1, the high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion, including following step
It is rapid:
Step 1: input hyper-spectral image data cube, hyper-spectral image data cube are to pass through pretreated 2 grades
Product data, pretreatment include Noise Filter, radiant correction and geometric correction processing；All wave bands have 75, imaging
Frequency spectrum is 1100 nanometers to 2500 nanometers, i.e. short-wave infrared imaging region；
Step 2: judge high spectrum image wave band data: hyperspectral image data is in acquisition process by various
The interference of factor can make the quality of subband data be damaged, and the band image of damage is just not necessarily to continue to handle；
Using the product δ of the quadratic sum of the gradient (GMC) of the quadratic sum and column average value of the derivative (DMC) of column average value
(k) judged；
δ (k)=D (k) × G (k) (1)
The mathematical model of the quadratic sum of the gradient (GMC) of the quadratic sum and column average value of the derivative (DMC) of column average value is such as
Under:
Wherein, k indicates that kth wave band, D (k) and G (k) respectively indicate the derivative of the derivative (DMC) of kth wave band column average value
The quadratic sum of the gradient (GMC) of quadratic sum and column average value, μi(k) andThe i-th column for respectively indicating kth band image are visited
The average value and gradient value of unit are surveyed, δ (k) indicates their product, as shown in Figure 2；
Step 3: reject the high spectrum image wave band data of damage: if δ (k) value is bigger than scheduled threshold epsilon, i.e. δ (k) >
ε shows that the image data of the wave band has been damaged, it is not necessary that enters and operates in next step, directly deletes the data of the wave band；
It, can from Fig. 2 (b) and Fig. 2 (c) although D (k) and G (k) can individually judge the quality problems of high spectrum image
To know, the data of wave band 1-4,19-22,38-42,67-75 are problematic, but when between wave band 43-67, they are difficult to judge,
Because the fluctuation of curve is very big.It is problematic that wave band 27 may be judged in Fig. 2 (b), it is actually out of question, it can using δ (k)
Problematic with the accurate data for judging which wave band, fluctuation is bigger, and data are more problematic, for these problematic numbers
According to, it is not necessary that carry out the processing of next step.For the ease of comparative analysis, the wave band data of the 6th, 23,27 and 52 is selected, respectively
As shown in Fig. 3 (a), Fig. 3 (b), Fig. 3 (c), Fig. 3 (d), there are two reasons: first is that image data is not damaged；In second is that
Face contains different types of wide Banded improvement, in Fig. 3, for the elimination and visual effect detection convenient for Banded improvement, to original
Beginning image has carried out transposition preservation, i.e., lateral Banded improvement transposition at vertically to Banded improvement；
Step 4: reading the hyperspectral image data of normal kth wave band: being selected in remaining hyperspectral image data
Kth wave band data carries out step 5 and step 6 operation respectively；
Step 5: the Strip noise removal of hyperspectral image data is carried out with two dimension stationary wavelet:
Step 5.1: the conversion of Banded improvement type: since the image of processing is 2 grades of products, having been subjected to a series of processing,
Such as Noise Filter, radiant correction and geometric correction are certainly also handled slice band, but Banded improvement is still
In the presence of especially some wide/thick Banded improvements and ultra-wide band.Here what is selected is Stationary Wavelet Transform, is conducive to band
The removal of noise is handled, and in wavelet decomposition, bright Banded improvement and geometric detail information are difficult to distinguish extraction, they usually exist
In the sub- coefficient image of high frequency.In order to eliminate as much as Banded improvement, bright wisp band is first transformed into filaments of sun band, in this way in small wavelength-division
Xie Hou, Banded improvement is primarily present in high yardstick decomposition layer, and geometric detail information is primarily present in low Scale Decomposition coefficient；
Step 5.2: setting wavelet decomposition scales number: the scale parameter that two-dimentional stationary wavelet decomposes is set as 5, i.e. M=5；
Step 5.3: selecting wavelet function: selecting basic function of Daubechies (db1) function as wavelet decomposition, then
The K-band image of selection is decomposed；
Step 5.4: image is decomposed: the parameter being arranged using step 5.2 and step 5.3, to the EO-1 hyperion k of input
Band image carries out wavelet decomposition；
Step 5.5: the decomposition coefficient of scale each after wavelet decomposition being handled: first determining whether coefficient subgraph whether also
Step 5.2 is gone to if the Banded improvement for including in coefficient subgraph is more comprising a large amount of Banded improvement, to the coefficient
Continue to decompose in image, until stopping carrying out wavelet decomposition when some decomposition coefficient subgraph is almost Banded improvement；
Then the coefficient value that last decomposition layer obtains all is carried out assigning zero processing；Finally small echo inversion is carried out by scale by coefficient
It changes, finally obtains the wavelet coefficient of each decomposition scale；
Step 5.6: being handled, obtained with small with the wavelet inverse transformation that the wavelet coefficient that step 5.5 obtains carries out whole image
Wave conversion handles the image after Banded improvement；
As shown in figure 4, (a)-(d) corresponding diagram 3 (a)-(d) in Fig. 4, figure 4, it is seen that for bright wisp band and gray scale
Banded improvement, wavelet transformation can effectively remove, but for filaments of sun band, wavelet transformation is not easy to remove.Because of filaments of sun band and image
Background information it is similar, eliminate filaments of sun band and have also been removed background；
Step 6: being filtered out with the Banded improvement that local interpolation carries out hyperspectral image data；
Step 6.1: calculating the column mean of high spectrum image, and obtain column mean curve graph；
Step 6.2: calculating the gradient of column mean, and obtain the gradient curve figure of column mean；
Step 6.3: by gradient curve figure, obtaining the position where wave crest and trough, here it is where Banded improvement
Position；
Step 6.4: the column where Banded improvement being filtered out, interpolation processing is then carried out, is as a result that of obtaining part
The image of interpolation remains the geometric detail Information invariability of the overwhelming majority because only localized region is handled；
Step 6.5: carry out match by moment processing: although moment-matching method method is a kind of Strip noise removal algorithm of classics,
But when being single use, it is desirable that type of ground objects difference is uniform, therefore when atural object distribution is non-homogeneous, filter effect is not satisfactory, such as
The filter effect that fruit will obtain increases the application range of moment-based operator, and different methods is needed to be combined, and formula (4) is just
It is the mathematical model of match by moment,
Wherein, k indicates kth band image in high spectrum image, Xk(i, j) and Yk(i, j) respectively indicates kth band image
Single pixel gray value adjustment front and back value, μkrAnd σkrRespectively indicate the mean value and standard deviation of entire image, μkjAnd σkjRespectively
The mean value and standard deviation of j column；
It is illustrated by taking Fig. 3 (b) as an example, as a result as shown in figure 5, there is 5 apparent bands, and the width of band is all super
A pixel is crossed, from Fig. 5 it will be clear that position where Banded improvement.After finding Banded improvement position, followed by
Banded improvement and interpolation processing are removed, as a result as shown in Fig. 6 (a), effect is not fine.For narrow strips, local interpolation may be used also
With, the band of special single pixel has better effect, and with the increase of strip width, the noise removal capability of local interpolation gradually weakens,
As shown in Fig. 6 (a), match by moment processing is carried out again on the basis of interpolation, effect is significantly improved, can be ultra wide band
Noise remove, as shown in Fig. 6 (b).Although effect be not also it is very ideal, can receive.As shown in fig. 7, Fig. 7 (a)-
(d) Fig. 3 (a)-(d) processing result is respectively corresponded.It will be seen in fig. 7 that local interpolation algorithm can substantially remove wider item
Band noise, what it is due to progress is local interpolation, thus detailed information also protect it is relatively good.
Experimental results are shown in figure 8 for moment-matching method, and as can be known from Fig. 8, direct match by moment goes Banded improvement to be not thorough,
The trace left is relatively more, since the supposed premise of match by moment is that atural object distribution pattern is relatively uniform, for unevenly distributed
Atural object, denoising effect is not so good, and therefore, match by moment is seldom used alone；
Step 7: fusion treatment is carried out to the processing result of step 5 and step 6；Fusion rule such as following formula:
If(k)=a × Iwt(k)+b×Ili(k) (5)
K in formula indicates kth wave band, just indicates fused image, Iwt(k) after indicating that wavelet transformation removes Banded improvement
Image, Iwt(k) indicating that regional area interpolation removes the image after Banded improvement, parameter a and b are respectively respective weight coefficient, they
Value range be [0,1], fusion factor a=b=0.5 is arranged in and a+b=1；
Step 8: whether the K-band for judging processing is that last wave band then terminates if it is last wave band；If not
Last wave band is then transferred to step 4, that is, reads the image of next wave band.
The result of the method for the present invention as shown in figure 9, from Fig. 4, Fig. 7, Fig. 8 and Fig. 9 it is found that the effect of the method for the present invention most
It is good, reach expected purpose.It can not only effectively remove Banded improvement, moreover it is possible to protect more detailed information.
Claims (6)
1. the high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion, which is characterized in that including following
Step:
Step 1: input hyper-spectral image data cube, hyper-spectral image data cube are to pass through pretreated 2 grades of products
Data, pretreatment include Noise Filter, radiant correction and geometric correction processing；
Step 2: judge high spectrum image wave band data: hyperspectral image data is in acquisition process by various factors
Interference, the quality of subband data can be made to be damaged, the band image of damage is just not necessarily to continue to handle；
Step 3: rejecting the high spectrum image wave band data of damage；
Step 4: reading the hyperspectral image data of normal kth wave band: selecting kth wave in remaining hyperspectral image data
Segment data carries out step 5 and step 6 operation respectively；
Step 5: the Strip noise removal of hyperspectral image data is carried out with two dimension stationary wavelet；
Step 6: being filtered out with the Banded improvement that local interpolation carries out hyperspectral image data；
Step 7: fusion treatment is carried out to the processing result of step 5 and step 6；
Step 8: whether the K-band for judging processing is that last wave band then terminates if it is last wave band；If not last
Wave band is then transferred to step 4, that is, reads the image of next wave band.
2. the high spectrum image Banded improvement removing method of wavelet transformation according to claim 1 and local interpolation fusion,
It is characterized in that, the method that high spectrum image wave band data judges in the step 2 are as follows:
Judged using the product δ (k) of the quadratic sum of the gradient of the quadratic sum and column average value of the derivative of column average value；
δ (k)=D (k) × G (k) (1)
The mathematical model of the quadratic sum of the gradient of the quadratic sum and column average value of the derivative of column average value is as follows:
Wherein, k indicates that kth wave band, D (k) and G (k) respectively indicate the quadratic sum and column average of the derivative of kth wave band column average value
The quadratic sum of the gradient of value, μi(k) andRespectively indicate the average value and ladder of the i-th column probe unit of kth band image
Angle value, δ (k) indicate the product of the quadratic sum of the quadratic sum of the derivative of column average value and the gradient of column average value.
3. the high spectrum image Banded improvement removing method of wavelet transformation according to claim 2 and local interpolation fusion,
It is characterized in that, the method for rejecting the high spectrum image wave band data of damage in the step 3 are as follows: if δ (k) value is than predetermined
Threshold epsilon it is big, i.e. δ (k) > ε shows that the image data of the wave band has been damaged, directly deletes the data of the wave band.
4. the high spectrum image Banded improvement removing method of wavelet transformation according to claim 1 and local interpolation fusion,
It is characterized in that, the specific steps of Strip noise removal are carried out in the step 5 with two dimension stationary wavelet are as follows:
Step 5.1: bright wisp band first the conversion of Banded improvement type: being transformed into filaments of sun band；
Step 5.2: setting wavelet decomposition scales number: the scale parameter that two-dimentional stationary wavelet decomposes is set as 5, i.e. M=5；
Step 5.3: it selects wavelet function: selecting Daubechies function as the basic function of wavelet decomposition, the k wave then selected
Section image is decomposed；
Step 5.4: image is decomposed: the parameter being arranged using step 5.2 and step 5.3, to the EO-1 hyperion K-band of input
Image carries out wavelet decomposition；
Step 5.5: the decomposition coefficient of scale each after wavelet decomposition is handled: first determine whether coefficient subgraph whether also include
A large amount of Banded improvement goes to step 5.2 if the Banded improvement for including in coefficient subgraph is more, is scheming to the coefficient
As continuing to decompose, until stopping carrying out wavelet decomposition when some decomposition coefficient subgraph is almost Banded improvement；Then
The coefficient value that last decomposition layer obtains all is carried out assigning zero processing；Wavelet inverse transformation finally is carried out by scale by coefficient,
Finally obtain the wavelet coefficient of each decomposition scale；
Step 5.6: being handled with the wavelet inverse transformation that the wavelet coefficient that step 5.5 obtains carries out whole image, acquisition is become with small echo
Image after changing processing Banded improvement.
5. the high spectrum image Banded improvement removing method of wavelet transformation according to claim 1 and local interpolation fusion,
It is characterized in that, local interpolation algorithm removes the specific steps of Banded improvement in the step 6 are as follows:
Step 6.1: calculating the column mean of high spectrum image, and obtain column mean curve graph；
Step 6.2: calculating the gradient of column mean, and obtain the gradient curve figure of column mean；
Step 6.3: by gradient curve figure, obtaining the position where wave crest and trough, here it is the positions where Banded improvement；
Step 6.4: the column where Banded improvement being filtered out, interpolation processing is then carried out, is as a result that of obtaining local interpolation
Image；
Step 6.5: carry out match by moment processing: formula (4) is exactly the mathematical model of match by moment,
Wherein, k indicates kth band image in high spectrum image, Xk(i, j) and YkIt is single that (i, j) respectively indicates kth band image
Grey scale pixel value adjustment front and back value, μkrAnd σkrRespectively indicate the mean value and standard deviation of entire image, μkjAnd σkjRespectively jth arranges
Mean value and standard deviation.
6. the high spectrum image Banded improvement removing method of wavelet transformation according to claim 1 and local interpolation fusion,
It is characterized in that, the specific steps of fusion treatment are carried out in the step 7 to the processing result of step 5 and step 6 are as follows:
Fusion rule such as following formula:
If(k)=a × Iwt(k)+b×Ili(k) (5)
K in formula indicates kth wave band, just indicates fused image, Iwt(k) indicate that wavelet transformation removes the figure after Banded improvement
Picture, Iwt(k) indicating that regional area interpolation removes the image after Banded improvement, parameter a and b are respectively respective weight coefficient, they
Value range is [0,1], and a+b=1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610985086.7A CN106504214B (en) | 2016-10-31 | 2016-10-31 | The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610985086.7A CN106504214B (en) | 2016-10-31 | 2016-10-31 | The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106504214A CN106504214A (en) | 2017-03-15 |
| CN106504214B true CN106504214B (en) | 2019-03-05 |
Family
ID=58324060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610985086.7A Expired - Fee Related CN106504214B (en) | 2016-10-31 | 2016-10-31 | The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106504214B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108233987A (en) * | 2018-01-10 | 2018-06-29 | 福建师范大学 | Eliminate the system and method for ultra-broadband signal intersymbol interference under multipath noise channel |
| CN108234367A (en) * | 2018-01-10 | 2018-06-29 | 福建师范大学 | Eliminate the system and method for multipath noisiless channel ultra-broadband signal intersymbol interference |
| CN109949252B (en) * | 2019-04-15 | 2020-12-25 | 北京理工大学 | Infrared image light spot removing method based on compensation coefficient fitting |
| CN112053302B (en) * | 2020-09-03 | 2023-10-17 | 五邑大学 | Denoising method and device for hyperspectral image and storage medium |
| CN113850802B (en) * | 2021-10-19 | 2024-06-14 | 内蒙古工业大学 | MATLAB-based automatic extraction method for spectral data of welding arc characteristic spectral line image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745487A (en) * | 2013-12-20 | 2014-04-23 | 西北工业大学 | Bayes high-spectral unmixing compressive sensing method based on structured sparsity prior |
| CN104732566A (en) * | 2015-03-16 | 2015-06-24 | 西北工业大学 | Hyperspectral image compressive sensing method based on nonseparable sparse prior |
| CN105844592A (en) * | 2016-01-14 | 2016-08-10 | 辽宁师范大学 | Wavelet domain total variation mixed denoising method for hyperspectral images |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2617119A1 (en) * | 2008-01-08 | 2009-07-08 | Pci Geomatics Enterprises Inc. | Service oriented architecture for earth observation image processing |
-
2016
- 2016-10-31 CN CN201610985086.7A patent/CN106504214B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745487A (en) * | 2013-12-20 | 2014-04-23 | 西北工业大学 | Bayes high-spectral unmixing compressive sensing method based on structured sparsity prior |
| CN104732566A (en) * | 2015-03-16 | 2015-06-24 | 西北工业大学 | Hyperspectral image compressive sensing method based on nonseparable sparse prior |
| CN105844592A (en) * | 2016-01-14 | 2016-08-10 | 辽宁师范大学 | Wavelet domain total variation mixed denoising method for hyperspectral images |
Non-Patent Citations (2)
| Title |
|---|
| Noise-Resistant Wavelet-Based Bayesian Fusion of Multispectral and Hyperspectral Images;Yifan Zhang;《IEEE Transactions on Geoscience and Remote Sensing》;20090526;第47卷(第11期);全文 |
| 高光谱遥感数据自适应小波滤噪;路威 等;《信息工程大学学报》;20050630;第6卷(第2期);全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106504214A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106504214B (en) | The high spectrum image Banded improvement removing method of wavelet transformation and local interpolation fusion | |
| CN108389188B (en) | Sparse hyperspectral abnormal target detection method | |
| Rasti et al. | Image restoration for remote sensing: Overview and toolbox | |
| Tom et al. | Morphology-based algorithm for point target detection in infrared backgrounds | |
| Mäkitalo et al. | Denoising of single-look SAR images based on variance stabilization and nonlocal filters | |
| Tania et al. | A comparative study of various image filtering techniques for removing various noisy pixels in aerial image | |
| CN109919870B (en) | SAR image speckle suppression method based on BM3D | |
| Li et al. | Hyperspectral image recovery employing a multidimensional nonlocal total variation model | |
| Somvanshi et al. | Comparative statistical analysis of the quality of image enhancement techniques | |
| CN110646851B (en) | Adaptive threshold seismic random noise suppression method based on Shearlet transformation | |
| Chen et al. | A novel infrared small target detection method based on BEMD and local inverse entropy | |
| CN112347992B (en) | Remote sensing estimation method for time sequence AGB in desert area | |
| Zhang et al. | Random noise attenuation of non-uniformly sampled 3D seismic data along two spatial coordinates using non-equispaced curvelet transform | |
| CN106504208B (en) | High-spectrum image width destriping method based on orderly minimum value and wavelet filtering | |
| CN113421198B (en) | Hyperspectral image denoising method based on subspace non-local low-rank tensor decomposition | |
| Yang et al. | Application and evaluation of wavelet-based denoising method in hyperspectral imagery data | |
| Cremer et al. | An image transform based on temporal decomposition | |
| Shao et al. | Seismic random noise attenuation using nonlocal-means via smooth patch ordering | |
| Lenti et al. | 2D TSVD to enhance the resolution of radiometer data | |
| Yasar et al. | Comparison of real and complex-valued versions of wavelet transform, curvelet transform and ridgelet transform for medical image denoising | |
| Huang et al. | Methods of removal wide-stripe noise in short-wave infrared hyperspectral remote sensing image | |
| Su-xia et al. | Image fusion method based on NSCT and robustness analysis | |
| Schmitt et al. | Adaptive multilooking of airborne Ka-band multi-baseline InSAR data of urban areas | |
| Yang et al. | Destriping for MODIS data via wavelet shrinkage | |
| Misra et al. | Wavelet based SAR data denoising and analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190305 Termination date: 20211031 |