CN101266686A - An image amalgamation method based on SFIM and IHS conversion - Google Patents

Abstract

The invention discloses an image fusing method based on SFIM and IHS transformation, which carries out pretreatment, such as geometric registration, de-noising, etc., for high spatial resolution images and high spectral resolution images; carries out HIS transformation for the RGB wave bands(R<0>, G<0>,B<0>) of the original high spatial resolution images with IHS transformation formula and respectively extracts I, H, and S components; carries out SFIM calculation for I component based on (I) formula, that is substituting I for P<low</SUB>, high spatial resolution images for P<high>, and obtains I<SFIM> through calculation; obtains R<new>, G<new> and B<new> based on IHS inversion formula, and then obtains the fusing result through superimposing wave bands. The invention introduces SFIM model, and presents the fusing method of combining SFIM and HIS, which can not only conspicuously maintain spectral characteristics, but also avoid introducing complex, time-consuming frequency decomposition and reconstruction process.

Description

A kind of image interfusion method based on SFIM and IHS conversion

Technical field

The present invention relates to a kind of image processing method, especially a kind of image interfusion method.

Background technology

At characteristics complementary between remotely-sensed data magnanimityization, variation, complicated and data, the redundancy coexistence, how efficiently, handle RS data at high speed, become present remote sensing application field problem demanding prompt solution.For this reason, based on the fusion purpose of " improving its spatial resolution under the prerequisite that keeps the source images spectral information as far as possible ", multiple fusion method is suggested and is used widely: " IEEE Transactions on Geoscience andRemote Sensing " disclosed " Multiresolution-Based Image Fusion with Additive WaveletDecomposition " in 1999; 2006, " Proceeding of the 2006 IEEE International Conference onMechatronics and Automation " published " Wavelet Based Remote Sensing Image Fusionwith Color Compensation Rule and IHS Transform " literary composition.The good characteristic that said method decomposes based on multiresolution, make full use of the complementary characteristic of source images, the spatial detail information of high spatial resolution images is rationally added in the I component, effectively replenish the information that lacks in the multispectral image, thereby the distortion of fusion results image spectrum is significantly reduced.But, time, the space complexity of these class methods are higher, therefore choosing of wavelet transformation form and wavelet basis also can exert an influence to fusion results, has consumption and computational complexity when high, is difficult to satisfy the real-time requirement of the multi-source remote sensing Flame Image Process of magnanimity day by day.In addition, decompose for fear of above-mentioned multiresolution, make every effort to calculate simple, real-time is good, calendar year 2001, " Society of Photo-Optical Instrumentation Engineers " discloses the method for " EfficientIntensity-Hue-Saturation-Based Image Fusion with Saturation Compensation "; 2006, " IEEE Transactions on Geoscience and Remote Sensing " open " A NewIntenstiy-Hue-Saturation Fusion Approach to Image Fusion with a Tradeoff Parameter " literary composition, said method is based on the IHS conversion, calculate succinct, can obtain the fusion results that has high spectral resolution and high spatial resolution simultaneously, the fusion performance obtains improvement in various degree.Though the described method of above-mentioned document has been avoided the multiresolution decomposition, to calculate simply, real-time is good, and fusion results obtains improvement in various degree; But weighting coefficient wherein, adjustment coefficient and threshold size mostly are artificial appointment, have randomness, and need can draw the result who is fit to certain applications through a large amount of tests, analysis.

Summary of the invention

Can't avoid spectrum distortion in order to overcome prior art, perhaps need to introduce complexity, frequency resolution consuming time and the deficiency of process of reconstruction, the invention provides a kind of image interfusion method based on SFIM and IHS conversion, introduce the SFIM model, propose the fusion method that SFIM and IHS combine, can significantly keep spectral characteristic to avoid introducing complexity, frequency resolution and process of reconstruction consuming time again.

The technical solution adopted for the present invention to solve the technical problems may further comprise the steps:

(a) high spatial resolution images and high spectral resolution image are carried out pre-service such as geometrical registration, denoising.

(b) by IHS direct transform formula

$\left[\begin{array}{c}I\\ v_1\\ v_2\end{array}\right]=\left[\begin{array}{ccc}1/3& 1/3& 1/3\\ -\sqrt{2}/6& -\sqrt{2}/6& 2\sqrt{2}/6\\ 1/\sqrt{2}& -1/\sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{R}_0\\ G_0\\ B_0\end{array}\right]$

$H=\arctan \left(\frac{v_2}{v_1}\right),S=\sqrt{v_1^2+v_2^2}$

RGB wave band (R to former high spectral resolution image ₀, G ₀, B ₀) carry out the IHS conversion, extract I, H, S component respectively.

(c) by formula

$P_{\mathrm{SFIM}}=\frac{P_{\mathrm{low}}{P}_{\mathrm{high}}}{P_{\mathrm{mean}}}$

I component is carried out the SFIM computing, and promptly I substitutes P _Low, high spatial resolution images substitutes P _High, calculate I _SFIM

(d) by IHS inverse transformation formula

$\left[\begin{array}{c}{R}_{\mathrm{new}}\\ G_{\mathrm{new}}\\ B_{\mathrm{new}}\end{array}\right]=\left[\begin{array}{ccc}1& -1/\sqrt{2}& 1/\sqrt{2}\\ 1& -1/\sqrt{2}& -1/\sqrt{2}\\ 1& \sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{SFIM}}\\ v_1\\ v_2\end{array}\right]$

Obtain R _New, G _New, B _New, and then obtain fusion results through wave band stack.

The SFIM model is based on solar radiation and the face of land spectral reflectance principle simplified, and for high spatial resolution images and high spectral resolution image, it is defined as:

DN(λ) _sim＝ρ(λ) _lowE(λ) _high

Wherein, D (λ) _SimThe result of smothing filtering is carried out in expression to each pixel of high spatial resolution images; ρ (λ) _LowExpression high spectral resolution image reflects the spectral information of high spectral resolution image at the spectral reflectivity of wave band λ; E (λ) _HighThe expression high spatial resolution images reflects landform and texture information that high spatial resolution images is introduced in the radiant illumination of wave band λ.

Because the SPECTRAL DIVERSITY between high spatial resolution and the high spectral resolution image is not a basic factor, above-mentioned formula can further be reduced to more general SFIM model:

$P_{\mathrm{SFIM}}=\frac{P_{\mathrm{low}}{P}_{\mathrm{high}}}{P_{\mathrm{mean}}}$

Wherein, P _LowExpression and high spatial resolution images P _HighA pixel of the high spectral resolution image of registration, P _MeanBe high spatial resolution images P _HighThrough an analog pixel of neighborhood smothing filtering gained, its resolution and P _LowIdentical.P _HighWith P _MeanRatio eliminated the spectrum and the contrast information of high spatial resolution images, the boundary information that has only kept high spatial resolution images, be the spatial detail information that lacks in the high spectral resolution image, therefore, the SFIM model can accurately utilize the complementary characteristic of source images, spatial detail information is reasonable, be modulated to effectively in the high spectral resolution image of registration, and do not change its spectral characteristic and contrast, and mate high spectral resolution image this point from the spatial resolution that reduces high spatial resolution images, the SFIM model is similar to wavelet transformation, treat that SFIM has but simplified greatly than wavelet transformation under the illumination and geological information condition that fused images is similar but be based on; On the other hand, this model is not suitable for the illumination condition image co-registration different with physical characteristics, as optical imagery and radar image etc.

The invention has the beneficial effects as follows: the present invention has introduced the SFIM model, proposes the fusion method that SFIM and IHS combine, and can significantly keep spectral characteristic to avoid introducing complexity, frequency resolution and process of reconstruction consuming time again.The inventive method is not introduced the multiresolution decomposable process, and spectrum keeps superior performance, details incorporates accurate and effective simultaneously, real-time is good, objective evaluation is consistent with visual effect, have certain contradiction owing to improve spatial resolution with keeping spectral information again, so the inventive method meets the fusion purpose, the syncretizing effect optimum merges performance efficiently, at a high speed.

The present invention is further described below in conjunction with drawings and Examples.

Description of drawings

Fig. 1 (a) is original full-colour image;

Fig. 1 (b) is original multispectral image;

Fig. 1 (c) is a conventional I HS method image;

Fig. 1 (d) is the image of control methods one;

Fig. 1 (e) is the image of control methods two;

Fig. 1 (f) is the image of control methods three;

Fig. 1 (g) is traditional SFIM method image;

Fig. 1 (h) is an image of the present invention.

Fig. 2 is the fusion results objective evaluation index comparison diagram of different fusion methods.

Embodiment

Adopt SPOT panchromatic wave-band image (10m * 10m) and " No. one, resource " satellite multispectral image (20m * 20m) of 256 grades of gray scales, 300 * 300 Pixel Dimensions, and the latter gets the synthetic false color image of 2,3,4 three wave bands in its 5 wave bands, above-mentioned image is through pre-service such as precise geometrical correction, image registrations, and all methods and experiment all realize on MATLAB7.0.

Present embodiment may further comprise the steps:

(a) high spatial resolution images and high spectral resolution image are carried out pre-service such as geometrical registration, denoising.

(b) by IHS direct transform formula

$\left[\begin{array}{c}I\\ v_1\\ v_2\end{array}\right]=\left[\begin{array}{ccc}1/3& 1/3& 1/3\\ -\sqrt{2}/6& -\sqrt{2}/6& 2\sqrt{2}/6\\ 1/\sqrt{2}& -1/\sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{R}_0\\ G_0\\ B_0\end{array}\right]$

$H=\arctan \left(\frac{v_2}{v_1}\right),S=\sqrt{v_1^2+v_2^2}$

RGB wave band (R to former high spectral resolution image ₀, G ₀, B ₀) carry out the IHS conversion, extract I, H, S component respectively.

(c) by formula

$P_{\mathrm{SFIM}}=\frac{P_{\mathrm{low}}{P}_{\mathrm{high}}}{P_{\mathrm{mean}}}$

I component is carried out the SFIM computing, and promptly I substitutes P _Low, high spatial resolution images substitutes P _High, calculate I _SFIM

(d) by IHS inverse transformation formula

$\left[\begin{array}{c}{R}_{\mathrm{new}}\\ G_{\mathrm{new}}\\ B_{\mathrm{new}}\end{array}\right]=\left[\begin{array}{ccc}1& -1/\sqrt{2}& 1/\sqrt{2}\\ 1& -1/\sqrt{2}& -1/\sqrt{2}\\ 1& \sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{SFIM}}\\ v_1\\ v_2\end{array}\right]$

Obtain R _New, G _New, B _New, and then obtain fusion results through wave band stack.

In the present embodiment, method (c) expression conventional I HS method;

Method (d) expression control methods one, promptly 2006, " IEEE Transactions on Geoscience and RemoteSensing " disclosed " A New Intenstiy-Hue-Saturation Fusion Approach to Image Fusionwith a Tradeoff Parameter " method (t=2);

Method (e) expression control methods two, promptly 2003, " 2nd GRSS/ISPRS Joint Workshop on ' DataFusion and Remote Sensing over Urban Areas " disclosed " High Resolution Image FusionBased on Wavelet and IHS Transformations " method;

Method (f) expression control methods three, promptly 2006, " Wavelet Based RemoteSensing Image Fusion with Color Compensation Rule and IHS Transform " method that " Proceeding of the 2006 IEEE InternationalConference on Mechatronics and Automation " publish;

The traditional SFIM method of method (g) expression;

Method (h) expression the inventive method.

From visual effect (Fig. 1), the c method has been lost the information in the I component fully because of " directly replace ", therefore, though kept the spatial detail information of full-colour image, has improved resolution greatly, serious distortion the spectral information of original multispectral image.The d method had both incorporated certain spatial detail information, kept certain spectral information again, but spectrum distortion still exists, even the value that changes balance factor t in order to adjust the contradiction between spatial resolution and the spectral resolution, still fails to reach better balance between the two.The e method is used the mallat wavelet transformation, has both incorporated the important information of full-colour image, reduced spectrum distortion again greatly, but had the details blooming, and blocking artifact is obvious.The f method is at the fusion rule of basic enterprising each frequency component of step refining wavelet decomposition of e method, blocking artifact and details blooming have been eliminated, image is more clear, but be not difficult to find out from result images, the more original multispectral image of picture contrast changes, and obviously occur deceptive information in the lake, thereby influenced the further maintenance of spectral information.The fusion results image of g method has shown the significant advantage of SFIM model, be that spectrum keeps superior performance, spatial detail obviously adds, the inventive method is then avoided the interpolation that repeats of information on this basis, further improved spatial resolution and spectral resolution, it is more accurate and effective to incorporate information.

At present, image syncretizing effect also need carry out objective evaluation except above-mentioned subjective assessment.Consider that from fusion purpose and fusion performance angle efficient, at a high speed this method selects for use 6 indexs such as spectral correlation coefficient, spatial detail related coefficient, standard deviation, average gradient, entropy and operation time to analyze based on " under the prerequisite that keeps the source images spectral information as far as possible, improving its spatial resolution ".Wherein, the spectral correlation coefficient is represented the similarity degree of fusion results and former multispectral image, and reflection spectrum keeps performance; The spatial detail related coefficient is represented fusion results and the similarity degree of former full-colour image aspect high-frequency information, and average gradient is contrast and the texture variations feature and the sharpness of minor detail in the presentation video then, and said two devices reflection spatial detail incorporates ability; What of information standard deviation presentation video contrast size, entropy then presentation video carry, the degree of enriching of said two devices reflection fusion results quantity of information; Operation time is based on 1.8GHzPentium4 PC, WindowsXP, and Matlab7.0 running environment directly reflects the method real-time.Outside aforementioned each index division operation time, be all big more excellent more type index.Table 2 has been enumerated the comparative result of objective evaluation index, and wherein, c method, d method, g method and the inventive method are all non-fusion method of decomposing based on multiresolution, and e method and f method are the fusion method of decomposing based on multiresolution.Carry out analysis-by-synthesis according to the statistics of table 2 below: the spectral correlation coefficient of the inventive method is obviously maximum, and the g method is taken second place, and the c method is the poorest, the fusion method that shows the spectrum maintenance performance of the inventive method even be better than decomposing based on multiresolution.For spatial detail related coefficient index, c method＞f method＞the inventive method＞e method＞g method＞d method, for the average gradient index, f method＞e method ≈ the inventive method＞g method＞c method＞d method, here it may be noted that, this two classes index can not be as the absolute measurement index of fusion results, and should weigh with other index such as spectral correlation coefficient etc. are common, even this two classes index is more excellent in the fusion results, if there is the spectrum distortion in the result, phenomenons such as deceptive information (described in subjective assessment) so also can not accurately be passed judgment on the quality of fusion results because of coming from non-former figure information.Therefore, according to the These parameters data relatively, the whole related coefficient maximum of spatial detail of the inventive method and spectrum, average gradient obviously improves, the inventive method is described in the spatial detail that keeps effectively having added on the basis of spectrum property former multispectral image disappearance, sharpness is better.From the standard deviation angle, the inventive method slightly is worse than f method and g method, entropy then slightly is worse than f method and e method, wherein, the e method is because blocking artifact causes deceptive information, the f method is also because of mallat small echo form and the undue refinement introducing of fusion rule deceptive information, the g method is repeating to add and bring bigger image contrast because of detailed information then, therefore all has higher information content, and the information that incorporates of the inventive method accurately, rationally, and do not have the deceptive information of being introduced of choosing by wavelet decomposition form and wavelet basis.And then the operation time of the inventive method with decompose based on multiresolution respectively improve one's methods to compare greatly and shorten, with non-respectively improving one's methods quite based on the multiresolution decomposition, wherein, increase the operation time than the g method to some extent, this is that comprehensive SFIM model and IHS conversion cause that computational complexity becomes greatly because the present invention improves one's methods, but compare with respectively improving one's methods, still satisfy the requirement of real-time, and spectrum keeps, a plurality of indexs such as details incorporates, information content all are better than g method and IHS method.In sum, the inventive method is not introduced the multiresolution decomposable process, and spectrum keeps superior performance, the while details incorporates accurate and effective, and real-time is good, and objective evaluation is consistent with visual effect, there is certain contradiction owing to improve spatial resolution with keeping spectral information again, therefore the inventive method meets the fusion purpose, and the syncretizing effect optimum merges performance efficiently, at a high speed.

Claims (1)

1, a kind of image interfusion method based on SFIM and IHS conversion is characterized in that comprising the steps:

(a) high spatial resolution images and high spectral resolution image are carried out pre-service such as geometrical registration, denoising;

(b) by IHS direct transform formula

$\left[\begin{array}{c}I\\ v_1\\ v_2\end{array}\right]=\left[\begin{array}{ccc}1/3& 1/3& 1/3\\ -\sqrt{2}/6& -\sqrt{2}/6& 2\sqrt{2}/6\\ 1/\sqrt{2}& -1/\sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{R}_0\\ G_0\\ B_0\end{array}\right]$

$H=\arctan \left(\frac{v_2}{v_1}\right),S=\sqrt{v_1^2+v_2^2}$

RGB wave band (R to former high spectral resolution image ₀, G ₀, B ₀) carry out the IHS conversion, extract I, H, S component respectively;

(c) by formula $P_{\mathrm{SFIM}}=\frac{P_{\mathrm{low}}{P}_{\mathrm{high}}}{P_{\mathrm{mean}}}$ I component is carried out the SFIM computing, and promptly I substitutes P _Low, high spatial resolution images substitutes P _High, calculate I _SFIM

(d) by IHS inverse transformation formula $\left[\begin{array}{c}{R}_{\mathrm{new}}\\ G_{\mathrm{new}}\\ B_{\mathrm{new}}\end{array}\right]=\left[\begin{array}{ccc}1& -1/\sqrt{2}& 1/\sqrt{2}\\ 1& -1/\sqrt{2}& -1/\sqrt{2}\\ 1& \sqrt{2}& 0\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{SFIM}}\\ v_1\\ v_2\end{array}\right]$ Obtain R _New, G _New, B _New, and then obtain fusion results through wave band stack.

Images