CN102622730A - Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) - Google Patents
Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) Download PDFInfo
- Publication number
- CN102622730A CN102622730A CN2012100617229A CN201210061722A CN102622730A CN 102622730 A CN102622730 A CN 102622730A CN 2012100617229 A CN2012100617229 A CN 2012100617229A CN 201210061722 A CN201210061722 A CN 201210061722A CN 102622730 A CN102622730 A CN 102622730A
- Authority
- CN
- China
- Prior art keywords
- image
- bemd
- frequency
- laplacian pyramid
- omega
- 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.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a remote sensing image fusion processing method based on a non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) and belongs to the technical field of remote sensing image fusion processing. The method comprises the following steps of: firstly, performing decomposition on a high-resolution panchromatic image by utilizing the non-subsampled Laplacian pyramid; then, performing the BEMD on a low-frequency part and a low-resolution panchromatic image to obtain a two-dimensional intrinsic mode function, and calculating instantaneous frequencies of all layers; performing fusion on a high-frequency detail part by utilizing the instantaneous frequencies and the absolute value of a decomposition coefficient as fusion characteristics by adopting a method of weighting; and finally, performing the corresponding BEMD and the inverse transformation of the non-subsampled Laplacian pyramid to obtain a fused image. According to the method disclosed by the invention, the spectral quality and the spatial detail quality of the fused image are better improved, the spectral characteristic can be well kept, and the resolution and the definition of the image are increased.
Description
Technical field
The present invention relates to the remote sensing image fusion disposal route, refer to a kind of particularly based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD.
Background technology
Low resolution multispectral image (low resolution multispectral image; MS) and high resolving power panchromatic image (high resolution panchromatic image; PAN) fusion all has important effect at agricultural, resource detection, urban environment etc. aspect a lot.In the method for amalgamation processing of high resolving power panchromatic image and low resolution multispectral image, traditional method for amalgamation processing mainly comprises HIS transform method, principal component analysis (PCA) method etc.Though these methods can obtain the fused images of high-resolution, there is bigger spectrum distortion in fused images.
In recent years, the multiple dimensioned Fusion Model in conjunction with pyramid or wavelet transformation has obtained using widely and having obtained effect preferably.Multiple dimensioned Fusion Model reaches the purpose that improves syncretizing effect through the decomposition result on the different scale is merged, and has obvious superiority.Yet pyramid or wavelet decomposition depend on predefined wave filter or wavelet function, and under identical condition, adopt different filter or wavelet function very big to the quality influence of fused image.
Empirical modal decomposes that (empirical mode decomposition, what EMD) be that Huang etc. proposed in 1998 a kind ofly is particularly suitable for the method that nonlinear data is analyzed.Need to confirm that in advance basis function is different with method such as small echo; This method is no longer dependent on basis function; It is the adaptive analysis method that a kind of complete data drives; It can non-linear non-stationary data decomposition become to accumulate in limited, a spot of the mode function component (intrinsic mode function, imf).EMD has than better space of wavelet analysis and frequency characteristic, and the physical characteristics of the description signal that one dimension EMD is good also can be extended to the analysis to 2D signal.At present, (bidimensional empirical mode decomposition BEMD) has been applied to panchromatic image and multispectral image and has merged, and has obtained good syncretizing effect in the two-dimensional empirical modal decomposition.But when adopting BEMD that panchromatic and multispectral image are merged at present, two aspect problems below existing: the difference of resolution between panchromatic image and the multispectral image is not considered in (1), directly PAN and MS image is carried out BEMD decomposition fusion.Because PAN is different with the resolution-scale of MS image, so after carrying out the BEMD decomposition of identical layer, in the high frequency layer of correspondence; The yardstick of the detailed information that is comprised is inevitable different; Be equivalent to during fusion the details of different scale is being merged, cause the aliasing of yardstick, influence fusion mass.(2) when merging, only keep BEMD and decompose the bigger HFS of back absolute value, cause losing more detailed information.Image is after BEMD decomposes; The absolute value of coefficient can not reflect the details of image comprehensively; The high frequency of image decomposes the part details mainly by local instantaneous frequency decision; Must unite instantaneous frequency could judge details accurately, therefore when merging, only keeps the bigger coefficient of absolute value, can lose a large amount of detailed information.
Summary of the invention
The object of the invention is to overcome the deficiency of above-mentioned prior art and provides a kind of based on non-remote sensing image fusion disposal route (the fusion method combining Nosampled Pyramid and BEMD that falls sampling Laplacian pyramid and BEMD; FMNPB), this method can improve spectral quality and the spatial detail quality that merges image.
Realize that the technical scheme that the object of the invention adopts is: a kind of based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD, may further comprise the steps:
1) the low resolution multispectral image is carried out the HIS conversion, obtain tone H, intensity I, three components of saturation degree S; And adopt the non-sampling Laplacian pyramid that falls that the high resolving power panchromatic image is carried out the decomposition of J layer, obtain J details component d
jWith a low frequency component A
J, 1≤j≤J wherein;
2) said strength component I is carried out K layer BEMD and decompose, accumulate mode function in obtaining
And remainder function
Low frequency component A
JCarry out K layer BEMD and decompose, accumulate mode function in obtaining
And remainder function
1≤k≤K wherein;
5) will
Remainder with strength component I
Carry out the BEMD inverse transformation, obtain high frequency and merge component
6) said high frequency is merged component and combine d
jCarry out the non-sampling Laplacian pyramid inverse transformation of falling, obtain the strength component I ' of fused image, promptly
7) the strength component I ' and the saturation degree S of combination tone H, fused image carry out the HIS inverse transformation, the high-resolution multi-spectral image after obtaining merging.
In technique scheme, step 1) adopts the non-sampling Laplacian pyramid that falls the high resolving power panchromatic image is carried out the J layer to decompose and comprise: at first high resolving power panchromatic image and non-desampling fir filter group are done convolution, obtain one deck decomposed P AN of original image
HAnd PAN
L, then non-desampling fir filter is done interpolation, the non-desampling fir filter after the interpolation is again and PAN
LConvolution repeats above-mentioned steps, promptly accomplishes the non-sampling Laplacian pyramid decomposition of falling of the J layer of high resolving power panchromatic image.
Further, said decomposition number of plies J by
Confirm, wherein, R
PanBe the resolution of high resolving power panchromatic image, R
MsResolution for the low resolution multispectral image.
In technique scheme, the instantaneous characteristic in the step 3) comprises: in accumulate the instantaneous frequency ω of mode function level, vertical, 45 degree and 135 degree four directions
iWith frequency energy E
ω,
1≤i≤4 wherein.
Further, with ω
iAnd E
ωApproximate component A as the high resolving power panchromatic image
JWith the fusion standard of the strength component I of low resolution multispectral image, adopt method of weighting right
With
The HFS of image merges:
D
F(x, the pixel value after y) expression is merged, D
A(x, y) expression first width of cloth waits to merge the pixel value of image, D
B(x, y) expression second width of cloth waits to merge the pixel value of image,
The frequency energy of representing first width of cloth image,
The frequency energy of representing second width of cloth image;
As frequency energy and pixel D
A(x, y), D
B(x, when the magnitude relationship of absolute value y) is inconsistent, weight coefficient α and β choose the rule as follows:
(1) when
The time, order
β '=1-α ', when
Order
β '=1-α ';
(2) work as D
A(x, y) and D
B(x, y) during jack per line, α=α ', β=β '; Work as D
A(x, y) and D
B(x, y) during contrary sign:
If | D
A(x, y) |>=| D
B(x, y) |, α=α ' then, β=-β ';
If | D
A(x, y) |<| D
B(x, y) |, then α=-α ', β=β '.
When the inventive method merges in the high frequency details, consider pixel value and instantaneous frequency energy two aspect factors simultaneously, kept more detailed information; And add temporary at coefficient, for jack per line and contrary sign coefficient, adopt different methods of weighting, thereby made the spatial detail quality of fusion results be further improved.In addition; Because the inventive method has been carried out the non-sampling pyramid decomposition of falling to high resolving power panchromatic image PAN earlier before using BEMD to decompose fusion, the low frequency part after the decomposition has identical dimensional properties with low resolution multispectral image MS image; Thereby make in follow-up BEMD decomposes; Accumulate mode function bimf in each layer two dimension and have close dimensional properties, reduced the yardstick aliasing when merging, improved the spectrum hold facility.Therefore the present invention is based on non-integration program of falling sampling pyramid and empirical modal decomposition and can keep spectral characteristic well, improve the resolution and the sharpness of image.
Description of drawings
Fig. 1 is the inventive method process flow diagram;
Fig. 2-the 1st, resolution is the original PAN image of 1m;
Fig. 2-the 2nd, resolution is the original MS image of 4m;
Fig. 2-the 3rd utilizes the non-sampling Laplacian pyramid that falls that the PAN image is carried out 2 layers of image after the decomposition;
Fig. 3-1 is the directly frequency comparison diagram of ground floor bimf after BEMD decomposes of original MS image and PAN image, and 101 is the frequency of MS image among the figure, and 102 is the frequency of PAN image;
Fig. 3-2 is the directly frequency comparison diagram of second layer bimf after BEMD decomposes of original MS image and PAN image, and 201 is the frequency of MS image among the figure, and 202 is the frequency of PAN image;
Fig. 3-3 is the directly frequency comparison diagram of the 3rd layer of bimf after BEMD decomposes of original MS image and PAN image, and 301 is the frequency of MS image among the figure, and 302 is the frequency of PAN image;
Fig. 4-1 carries out the frequency comparison diagram that BEMD decomposes back ground floor bimf for the strength component I of MS image and the PAN image low frequency part A2 after decomposing, and 401 is the frequency of MS image among the figure, and 402 is the frequency of PAN image;
Fig. 4-2 carries out the frequency comparison diagram that BEMD decomposes back second layer bimf for the strength component I of MS image and the PAN image low frequency part A2 after decomposing, and 501 is the frequency of MS image among the figure, and 502 is the frequency of PAN image.
Fig. 4-3 carries out the frequency comparison diagram that BEMD decomposes the 3rd layer of bimf in back for the strength component I of MS image and the PAN image low frequency part A2 after decomposing, and 601 is the frequency of MS image among the figure, and 602 is the frequency of PAN image;
Fig. 5-1 is original multispectral image MS
0
Fig. 5-2 is the multispectral image MS that degrades of Fig. 5-1
1
Fig. 5-3 is the full-colour image PAN that degrades of Fig. 2-1
1
Fig. 5-4 is for using HIS method fusion results figure;
Fig. 5-5 is for using DWT method fusion results figure;
Fig. 5-6 is for using SWT method fusion results figure;
Fig. 5-7 is for using BEMD method fusion results figure;
Fig. 5-8 is for using the inventive method fusion results figure.
Embodiment
Below in conjunction with accompanying drawing and specific embodiment the present invention is further described.
As shown in Figure 1, based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD, may further comprise the steps:
Step S101: MS carries out the HIS conversion to the low resolution multispectral image, obtains tone H, intensity I, three components of saturation degree S, and adopts the non-sampling Laplacian pyramid that falls that high resolving power panchromatic image PAN is carried out the decomposition of J layer, obtains J details component d
jWith a low frequency component A
J, 1≤j≤J wherein.Decompose the resolution R of number of plies J by high resolving power panchromatic image PAN
PanResolution R with the low resolution multispectral image
MsRatio confirm, promptly
Step S102: the strength component I that obtains is carried out K layer BEMD decompose, obtain accumulateing in the two dimension mode function
And remainder function
Low frequency component A
JCarry out K layer BEMD and decompose, obtain accumulateing in the two dimension mode function
And remainder function
1≤k≤K wherein.
Step S103: utilize in the two-dimentional Hilbert transform extraction and accumulate the mode function image
With
Instantaneous characteristic, the instantaneous characteristic that present embodiment extracts comprises: in accumulate the instantaneous frequency ω of mode function level, vertical, 45 degree and 135 degree four directions
iWith frequency energy E
ω,
1≤i≤4 wherein.With ω
iAnd E
ωApproximate component A as the high resolving power panchromatic image
JFusion standard with the strength component I of low resolution multispectral image.
Step S104: utilize the instantaneous characteristic that obtains; Adopt method of weighting that
and
merged, obtain
Step S105: will
Remainder with strength component I
Carry out the BEMD inverse transformation, obtain high frequency and merge component
Step S106: combine d
jCarry out the non-sampling Laplacian pyramid inverse transformation of falling, obtain the strength component I ' of fused image, promptly
Step S107: combine the strength component I ' and the saturation degree S of tone H, fused image, carry out the HIS inverse transformation, the high-resolution multi-spectral image after obtaining merging.
The effect of the inventive method is described below in conjunction with the result figure of actual image.
The resolution of low resolution multispectral image (hereinafter to be referred as the MS image) and high resolving power panchromatic image (hereinafter to be referred as the PAN image) differs bigger; Decompose if directly utilize BEMD that MS image and PAN image are carried out identical layer, then respective layer in accumulate mode function and must have different resolution.Fig. 2-1 is that a width of cloth resolution is the PAN image of 1m, and Fig. 2-2 is that a width of cloth resolution is the MS image of 4m.The MS image is carried out the HIS conversion; Strength component I and PAN image are carried out three layers of BEMD decomposition respectively; To
and
(
and
represent respectively strength component I and PAN image in accumulate mode function; 1≤k≤3) the tenth line frequency analysis of advancing; Analysis result is shown in Fig. 3-1, Fig. 3-1 and Fig. 3-3; Can find out by figure; Accumulate mode function in each layer of MS image and PAN image and all have different yardstick resolution; Therefore; If
after the direct decomposition and
are merged; With making the detailed information of different scale merge; Must cause the aliasing of yardstick, the quality of image is merged in influence to a great extent.
And the present invention adopts the non-sampling Laplacian pyramid that falls the high resolving power panchromatic image is carried out the J layer to decompose and comprise: at first high resolving power panchromatic image and non-desampling fir filter group are done convolution, obtain one deck decomposed P AN of original image
H(high frequency imaging) and PAN
L(low-frequency image) done interpolation to non-desampling fir filter then, and the non-desampling fir filter after the interpolation again and PAN
LConvolution repeats above-mentioned steps, promptly accomplishes the non-sampling Laplacian pyramid decomposition of falling of the J layer of high resolving power panchromatic image.The non-sampling Laplacian pyramid that falls has good multiresolution characteristic, makes each intersubband frequency spectrum not have aliasing.
Utilize the non-sampling Laplacian pyramid that falls the PAN image to be carried out 2 layers of decomposition, Fig. 2-the 3rd, the low frequency part A after the PAN image decomposes
2, low frequency A then
2Yardstick resolution and MS image approximately equal.Strength component I and low frequency part A to the MS image
2Carry out three layers of BEMD respectively and decompose, still to the tenth of every layer of bimf line frequency analysis of advancing, analysis result shown in Fig. 4-1, Fig. 4-2 and Fig. 4-3, can be found out low frequency part A by figure respectively
2Carry out after BEMD decomposes with the MS image, it is roughly the same to accumulate the mode function frequency content in every layer.Merge
and
this moment; Avoid the detailed information of different scale to merge each other effectively; Reduce the yardstick aliasing when merging, improve the spectral quality of fused images.
To low frequency part A
2With the MS image after BEMD decomposes, accumulate the characteristic that mode function bimf has local frequencies from high to low in each layer two dimension.The local frequencies of image and amplitude have comprised strong contrast physical message such as edge and grain details in the image, utilize two-dimentional Hilbert transform to extract the instantaneous characteristic of bimf image again.Accumulate mode component bimf in i
iTime-frequency representation do
Wherein (t)=[x, y]
T, B
i(t) be i bimf, H (B
i(t)) expression B
i(t) two-dimentional Hilbert conversion, a
i(t) be the amplitude function of bimf, θ
i(t) be the phase function of bimf.According to θ
i(t) can calculate local instantaneous frequency, because θ
iTherefore (t) be two-dimentional, can calculate the instantaneous frequency of a plurality of directions, that present embodiment is only fetched water is flat, vertical, the instantaneous frequency of 45 degree and 135 degree four directions.If
e
1=[1,0]
T, e
2=[0,1]
T, e
3=[1,1]
T, e
4=[1,1]
T, then level, instantaneous frequency vertical, that 45 degree and 135 are spent four directions are respectively ω
i=[ω
x, ω
y] e
i, 1≤i≤4.Choose the instantaneous frequency characteristic of the average of four direction instantaneous frequency quadratic sum, be frequency energy E as bimf
ω,
This frequecy characteristic can better embody the outstanding informational content of partial points.
Fusion for low resolution multispectral image and high resolving power panchromatic image; Purpose is that the detailed information with full-colour image joins in the multispectral image; Make the image after the fusion both have the high resolving power characteristic of full-colour image, can keep the spectral characteristic of multispectral image again.Because detailed information often is embodied in the HFS of image, therefore the fusion to the high frequency details is a key of image co-registration.The contrast information of image is mainly by local instantaneous frequency decision, and monochrome information is mainly determined by the pixel value of image.Therefore the present invention is when merging HFS, and Combined Frequency energy and pixel absolute value are estimated as activity, and it is following to merge computing method:
D
F(x, the pixel value after y) expression is merged is
D
A(x, y) expression first width of cloth waits to merge the pixel value of image, D
B(x, y) expression second width of cloth waits to merge the pixel value of image,
The frequency energy of representing first width of cloth image,
The frequency energy of representing second width of cloth image;
As frequency energy and pixel D
A(x, y), D
B(x when the magnitude relationship of absolute value y) is inconsistent, in order to improve the order of accuarcy that the local grain details is described, adopts weighting fusion, weight coefficient α and β choose regular as follows:
(1) value of at first definite α ' and β '
When
The time, order
β '=1-α ';
When
Order
β '=1-α '.
(2) again by the value of α ' and β ' definition weight coefficient α and β
Work as D
A(x, y) and D
B(x, y) during jack per line, α=α ', β=β ';
Work as D
A(x, y) and D
B(x, y) during contrary sign:
If | D
A(x, y) |>=| D
B(x, y) |, α=α ' then, β=-β ';
If | D
A(x, y) |<| D
B(x, y) |, then α=-α ', β=β '.
Present embodiment is designated as MS with the multispectral image of original 4m resolution image as a reference
0, shown in Fig. 5-1; With MS
0Sample 16m resolution, become degraded image, be designated as MS
1, shown in Fig. 5-2; With the full-colour image of 1m resolution, sample 4m resolution, be designated as PAN
1, shown in Fig. 5-3.Present embodiment adopts following 5 kinds of methods to MS respectively
1Image and PAN
1Image merges: the inventive method (being designated as FMNPB), HIS converter technique, DWT method, SWT method, BEMD direct method.In the fusion of DWT, SWT and BEMD direct method, adopt the high frequency coefficient absolute value to get big method and merge, decompose the number of plies and all be taken as 3 layers.When adopting the FMNPB method to merge, because PAN
1Image and MS
1The resolution of image is 1: 4, so carry out non-falling when sampling pyramid decomposition, decomposes the number of plies according to step S102 and should get 2, and the decomposition number of plies of BEMD gets 3.Adopt HIS converter technique, DWT method, SWT method, BEMD direct method and FMNPB method fusion results of the present invention respectively shown in Fig. 5-4,5-5,5-6,5-7 and 5-8.
The purpose of visual fusion is to improve spatial resolution as much as possible; And keeping enough multispectral characteristics to be used for atural object identification and classification, present embodiment is analyzed comparative descriptions from spectral quality and two aspects of spatial detail quality of fused images to fusion results respectively.
In order to estimate the spatial detail quality of fused images, adopt following two evaluation criterions:
(1) spatial frequency SF: spatial frequency has reflected the overall active degree of piece image spatial domain, and spatial frequency is high more, and then spatial detail information is abundant more.
(2) what of the full-colour image detailed information that is incorporated in the fused images high frequency related coefficient HPCC:HPCC reflected; Its calculates is the related coefficient between the detail of the high frequency of panchromatic wave-band detail of the high frequency and fused images; The HPCC value is high more; Just be illustrated in the fusion process, more PAN image space information " injection " arranged to the MS image.It is 1 good more that the value of HPCC approaches more.
Table 1 is respectively the spatial detail quality comparative result that adopts after HIS converter technique, DWT method, SWT method, BEMD direct method and FMNPB method of the present invention merge image.
Table 1
Can find out from table 1; The HIS method can keep better space details quality, and the details hold facility of DWT method is relatively poor, and the details hold facility of BEMD method is better than the SWT method slightly; The spatial detail hold facility of FMNPB method is superior to the BEMD method, and is suitable with the HIS method.Because FMNPB method of the present invention when the high frequency details merges, has been considered coefficient and instantaneous frequency energy two aspect factors simultaneously, has kept more detailed information; And add temporary at coefficient, for jack per line and contrary sign coefficient, adopt different methods of weighting, thereby made the spatial detail quality of fusion results be further improved.More than analysis shows, can keep spectral characteristic well based on non-integration program of falling sampling pyramid and empirical modal decomposition, improves the resolution and the sharpness of image.
Adopt following parameter to weigh the spectral quality of fused images:
(1) related coefficient CC: related coefficient is meant the related coefficient between the corresponding wave band of each wave band and reference picture of fused images.It is many more that the fusion results spectral information keeps, and the value of CC approaches 1 more, and desirable situation should be 1.
(2) torsion resistance DD: torsion resistance has directly reflected the spectrum distortion level of fused images, and torsion resistance is more little, shows that the distortion level of image is more little.More than two parameters all respectively three wave bands are estimated.
(3) averaged spectrum error extension (RASE): RASE representes that with percentage the spectral quality of fused images is high more relatively, and then RASE is low more, and ideal value should be 0.
(4) relative global error ERGAS: high more ERGAS is low more for the spectral quality of fused images, and ideal situation should be 0.
Table 2 is for adopting image and the reference picture MS after HIS converter technique, DWT method, SWT method, BEMD direct method and FMNPB method of the present invention merge respectively
0The numerical tabular of the spectral quality parameter between corresponding wave band, in order to analyze comparison better, last row in the table 2 have then provided the ideal value of each parameter.
Table 2
The spectrum retention performance that can find out the HIS method from table 2 is the poorest, and the spectral quality of SWT method and BEMD method totally is more or less the same, but all is superior to the fusion method based on DWT.And the FMNPB method that the present invention proposes is compared with the BEMD method with SWT, and each item index all has certain improvement, has better spectrum hold facility.This be since the FMNPB method before using BEMD to decompose to merge; Earlier panchromatic PAN image has been carried out the non-sampling pyramid decomposition of falling; Low frequency part after the decomposition has identical dimensional properties with multispectral MS image, thereby makes that each layer bimf has close dimensional properties in follow-up BEMD decomposes; Reduce the yardstick aliasing when merging, improved the spectrum hold facility.
Claims (5)
1. one kind based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD, it is characterized in that may further comprise the steps:
1) the low resolution multispectral image is carried out the HIS conversion, obtain tone H, intensity I, three components of saturation degree S; And adopt the non-sampling Laplacian pyramid that falls that the high resolving power panchromatic image is carried out the decomposition of J layer, obtain J details component d
jWith a low frequency component A
J, 1≤j≤J wherein;
2) said strength component I is carried out K layer BEMD and decompose, obtain accumulateing in the two dimension mode function
And remainder function
Low frequency component A
JCarry out K layer BEMD and decompose, obtain accumulateing in the two dimension mode function
And remainder function
1≤k≤K wherein;
5) will
Remainder with strength component I
Carry out the BEMD inverse transformation, obtain high frequency and merge component
6) said high frequency is merged component and combine d
jCarry out the non-sampling Laplacian pyramid inverse transformation of falling, obtain the strength component I ' of fused image, promptly
7) the strength component I ' and the saturation degree S of combination tone H, fused image carry out the HIS inverse transformation, the high-resolution multi-spectral image after obtaining merging.
2. said based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD according to claim 1; It is characterized in that in the step 1) adopting the non-sampling Laplacian pyramid that falls the high resolving power panchromatic image is carried out the J layer to decompose and comprise: at first high resolving power panchromatic image and non-desampling fir filter group are done convolution, the one deck that obtains original image decomposes high frequency imaging PAN
HWith low-frequency image PAN
L, then non-desampling fir filter is done interpolation, the non-desampling fir filter after the interpolation again with low-frequency image PAN
LConvolution repeats above-mentioned steps, promptly accomplishes the non-sampling Laplacian pyramid decomposition of falling of the J layer of high resolving power panchromatic image.
3. according to claim 1 or 2 said, it is characterized in that based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD: said decomposition number of plies J by
Confirm, wherein, R
PanBe the resolution of high resolving power panchromatic image, R
MsResolution for the low resolution multispectral image.
4. said based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD according to claim 1; It is characterized in that the instantaneous characteristic in the step 3) comprises: in accumulate the instantaneous frequency ω of mode function level, vertical, 45 degree and 135 degree four directions
iWith frequency energy E
ω,
1≤i≤4 wherein.
5. said based on non-remote sensing image fusion disposal route of falling sampling Laplacian pyramid and BEMD according to claim 4, it is characterized in that: with ω
iAnd E
ωApproximate component A as the high resolving power panchromatic image
JWith the fusion standard of the strength component I of low resolution multispectral image, adopt method of weighting right
With
The HFS of image merges:
D
F(x, the pixel value after y) expression is merged, D
A(x, y) expression first width of cloth waits to merge the pixel value of image, D
B(x, y) expression second width of cloth waits to merge the pixel value of image,
The frequency energy of representing first width of cloth image,
The frequency energy of representing second width of cloth image;
As frequency energy and pixel D
A(x, y), D
B(x, when the magnitude relationship of absolute value y) is inconsistent, weight coefficient α and β choose the rule as follows:
(1) when
The time, order
β '=1-α ', when
Order
β '=1-α ';
(2) work as D
A(x, y) and D
B(x, y) during jack per line, α=α ', β=β '; Work as D
A(x, y) and D
B(x, y) during contrary sign:
If | D
A(x, y) |>=| D
B(x, y) |, α=α ' then, β=-β ';
If | D
A(x, y) |<| D
B(x, y) |, then α=-α ', β=β '.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012100617229A CN102622730A (en) | 2012-03-09 | 2012-03-09 | Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012100617229A CN102622730A (en) | 2012-03-09 | 2012-03-09 | Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102622730A true CN102622730A (en) | 2012-08-01 |
Family
ID=46562635
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012100617229A Pending CN102622730A (en) | 2012-03-09 | 2012-03-09 | Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102622730A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103617604A (en) * | 2013-08-28 | 2014-03-05 | 内蒙古科技大学 | Image fusion method based on characteristic extraction of two dimension empirical mode decomposition method |
CN107027018A (en) * | 2017-04-19 | 2017-08-08 | 西安诺瓦电子科技有限公司 | Image processing method and device and sending card |
CN108171679A (en) * | 2017-12-27 | 2018-06-15 | 合肥君正科技有限公司 | A kind of image interfusion method, system and equipment |
CN108765359A (en) * | 2018-05-31 | 2018-11-06 | 安徽大学 | Fusion method of hyperspectral remote sensing image and full-color image based on JSK model and NSCT technology |
CN110021002A (en) * | 2018-01-10 | 2019-07-16 | 青柠优视科技(北京)有限公司 | A kind of image interfusion method and device |
CN111626934A (en) * | 2020-05-14 | 2020-09-04 | 山东师范大学 | Method, system, medium and apparatus for reconstruction of high spatial resolution multispectral images |
CN113591754A (en) * | 2018-11-16 | 2021-11-02 | 北京市商汤科技开发有限公司 | Key point detection method and device, electronic equipment and storage medium |
CN114140359A (en) * | 2021-12-08 | 2022-03-04 | 中国科学技术大学 | Remote sensing image fusion sharpening method based on progressive cross-scale neural network |
CN117408902A (en) * | 2023-10-23 | 2024-01-16 | 山东锋士信息技术有限公司 | Ka frequency band SAR image colorization method based on combination of BEMD and IHS conversion |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6816627B2 (en) * | 2001-04-12 | 2004-11-09 | Lockheed Martin Corporation | System for morphological image fusion and change detection |
CN101216557A (en) * | 2007-12-27 | 2008-07-09 | 复旦大学 | Residual hypercomplex number dual decomposition multi-light spectrum and full-color image fusion method |
CN101226635A (en) * | 2007-12-18 | 2008-07-23 | 西安电子科技大学 | Multisource image anastomosing method based on comb wave and Laplace tower-shaped decomposition |
-
2012
- 2012-03-09 CN CN2012100617229A patent/CN102622730A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6816627B2 (en) * | 2001-04-12 | 2004-11-09 | Lockheed Martin Corporation | System for morphological image fusion and change detection |
CN101226635A (en) * | 2007-12-18 | 2008-07-23 | 西安电子科技大学 | Multisource image anastomosing method based on comb wave and Laplace tower-shaped decomposition |
CN101216557A (en) * | 2007-12-27 | 2008-07-09 | 复旦大学 | Residual hypercomplex number dual decomposition multi-light spectrum and full-color image fusion method |
Non-Patent Citations (2)
Title |
---|
ZHANG XIAODONG ET AL: "A fusion algorithm for remote sensing images based on nonsubsample pyramids and bidimensional empirical decomposition", 《SCIENCE CHINA TECHNOLOGICAL SCIENCES》 * |
郑有志等: "基于二维经验模态分解的医学图像融合算法", 《软件学报》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103617604A (en) * | 2013-08-28 | 2014-03-05 | 内蒙古科技大学 | Image fusion method based on characteristic extraction of two dimension empirical mode decomposition method |
CN103617604B (en) * | 2013-08-28 | 2016-06-15 | 内蒙古科技大学 | Fusion method based on the image that two-dimensional empirical mode decomposition method feature extracts |
CN107027018A (en) * | 2017-04-19 | 2017-08-08 | 西安诺瓦电子科技有限公司 | Image processing method and device and sending card |
CN108171679A (en) * | 2017-12-27 | 2018-06-15 | 合肥君正科技有限公司 | A kind of image interfusion method, system and equipment |
CN110021002A (en) * | 2018-01-10 | 2019-07-16 | 青柠优视科技(北京)有限公司 | A kind of image interfusion method and device |
CN108765359A (en) * | 2018-05-31 | 2018-11-06 | 安徽大学 | Fusion method of hyperspectral remote sensing image and full-color image based on JSK model and NSCT technology |
CN108765359B (en) * | 2018-05-31 | 2021-08-24 | 安徽大学 | Fusion method of hyperspectral remote sensing image and full-color image based on JSK model and NSCT technology |
CN113591754A (en) * | 2018-11-16 | 2021-11-02 | 北京市商汤科技开发有限公司 | Key point detection method and device, electronic equipment and storage medium |
CN111626934A (en) * | 2020-05-14 | 2020-09-04 | 山东师范大学 | Method, system, medium and apparatus for reconstruction of high spatial resolution multispectral images |
CN114140359A (en) * | 2021-12-08 | 2022-03-04 | 中国科学技术大学 | Remote sensing image fusion sharpening method based on progressive cross-scale neural network |
CN114140359B (en) * | 2021-12-08 | 2023-03-10 | 中国科学技术大学 | Remote sensing image fusion sharpening method based on progressive cross-scale neural network |
CN117408902A (en) * | 2023-10-23 | 2024-01-16 | 山东锋士信息技术有限公司 | Ka frequency band SAR image colorization method based on combination of BEMD and IHS conversion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102622730A (en) | Remote sensing image fusion processing method based on non-subsampled Laplacian pyramid and bi-dimensional empirical mode decomposition (BEMD) | |
CN101504766B (en) | Image amalgamation method based on mixed multi-resolution decomposition | |
CN101847257B (en) | Image denoising method based on non-local means and multi-level directional images | |
CN110415199B (en) | Multispectral remote sensing image fusion method and device based on residual learning | |
Muresan et al. | Demosaicing using optimal recovery | |
CN102890820B (en) | Based on shearlet conversion and the image de-noising method of Wiener filtering | |
CN101482617A (en) | Synthetic aperture radar image denoising method based on non-down sampling profile wave | |
CN101266686A (en) | An image amalgamation method based on SFIM and IHS conversion | |
CN104123705B (en) | A kind of super-resolution rebuilding picture quality Contourlet territory evaluation methodology | |
Hogrefe et al. | Derivation and integration of shallow-water bathymetry: Implications for coastal terrain modeling and subsequent analyses | |
CN110532615B (en) | Progressive decomposition method for shallow sea complex terrain | |
CN102509263A (en) | K-SVD (K-means singular value decomposition) speckle inhibiting method based on SAR (synthetic aperture radar) image local statistic characteristic | |
CN101866479A (en) | Edge self-adaptive image amplification method based on non-downsampling Contourlet conversion | |
CN104182941A (en) | Hyperspectral image band noise removing method | |
CN103679661A (en) | Significance analysis based self-adaptive remote sensing image fusion method | |
CN101996321A (en) | Fingerprint recognition pretreatment method and device | |
CN104268833A (en) | New image fusion method based on shift invariance shearlet transformation | |
CN104635188A (en) | K-space reconstruction method and magnetic resonance imaging method | |
CN102129694A (en) | Method for detecting salient region of image | |
CN103106660A (en) | Synthetic aperture radar (SAR) image quality evaluation method based on contrast sensitivity characteristics | |
Zou et al. | Joint wavelet sub-bands guided network for single image super-resolution | |
CN105469358A (en) | Image processing method | |
CN109064394B (en) | Image super-resolution reconstruction method based on convolutional neural network | |
CN102800047B (en) | Method for reconstructing super resolution of single-frame image | |
CN104463822A (en) | Multi-focus image fusing method and device based on multi-scale overall filtering |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120801 |