CN102982523B - Multisource and multi-focus color image fusion method - Google Patents

Abstract

The invention discloses a multisource and multi-focus color image fusion method and relates to a computation method of multisource and multi-focus color image fusion. The multisource and multi-focus color image fusion method solves the problems of color distortion and a blurring effect in the available multisource and multi-focus color image fusion technology. The fusion method comprises the steps of modeling a plurality of original color images to be fused in a two-dimensional exchangeable Clifford algebra manner, achieving the integrated processing of the color images, conducting Clifford shearlet transformation on the color images after the integrated processing respectively, obtaining thick scale coefficients and thin scale coefficients of the color images, selecting the coefficient with a minimal norm value as the thick scale coefficient of fused color image, selecting the coefficient with a maximal norm value as the thin scale coefficient of the fused color image, conducting inverse Clifford shearlet transformation, and obtaining the fused color image. The fusion method can remove the image blurring farthest.

Description

The fusion method of multi-source multi-focus color image

Technical field

The present invention relates to the computing method that a kind of multi-source multi-focus color image merges, be specifically related to the transform domain computing method of coloured image during Multi-sensor Image Fusion.

Background technology

Image co-registration has important application in medical science, remote sensing and military target identification etc.In the application of reality, need the image to multiple sensor obtains to carry out information fusion, realize the message complementary sense between different images, to strengthen the transparency of information in image, to target carry out clear, complete, describe accurately, for follow-up decision-making provides reliable Informational support.

For ensureing the quality of multi-source Color Image Fusion, require that the color rendition degree in image co-registration computation process is high, image blur is little.In the computation process of image co-registration, usually adopt the method for spatial domain or transform domain.The method in spatial domain needs to be divided into several subimages to original image, selects suitable fuzzy mearue and fusion rule to obtain fused images.Although the advantage that this method has computation complexity low, its maximum shortcoming is exactly that fused images easily occurs mosaic effect, there is certain fuzzy distortion simultaneously.For the method for frequency domain, typically want base in the fusion method of wavelet transformation.Although the method does not exist mosaic effect, there is cross-color and blurring effect in fused images.This is main because the method based on small echo is that subchannel carries out, and cannot realize the bulk treatment of coloured image.It is high that current existing fusion calculation method all cannot reach color rendition degree, the requirement that image blur is little.

Summary of the invention

The present invention solves cross-color existing in existing multi-source multi-focus color image integration technology and blurring effect problem, provides a kind of fusion method of multi-source multi-focus color image.

The fusion method of multi-source multi-focus color image, to the bulk treatment of coloured image and the application of employing Clifford shearlet conversion, the method is realized by following steps:

Step one, original several coloured images to be fused are carried out modeling with the form of the commutative Clifford algebra of two dimension, realize the bulk treatment to coloured image;

Detailed process is: source images is carried out modeling with the form of the commutative Clifford algebra of two dimension, two-dimentional commutative Clifford algebra in element f there is following expression-form: f=a ₁ε ₁+ a ₂ε ₂+ a ₁₂ε ₁₂, wherein a ₁, a ₂, a ₁₂∈ R, ε _imultiplying meet law of commutation, its operation rule is provided by formula one:

Formula one: ε _iε _j=ε _jε _i, i ≠ j; i is more than or equal to the positive integer that 1 is less than or equal to n;

Adopt commutative Clifford algebra to achieve the modeling of coloured image, be expressed as with formula two:

Formula two: f (x, y)=f _r(x, y) ε ₁+ f _g(x, y) ε ₂+ f _b(x, y) ε ₁₂

In formula, f _r(x, y), f _g(x, y), f _b(x, y) is R, G and B color component of coloured image respectively;

Step 2, traditional shearlet conversion is generalized to Clifford algebra aspect, give definition and discretize algorithm thereof that Clifford shearlet converts, described Clifford shearlet conversion is defined as follows:

Step 2 one, a series of dilation transformation, shear transformation and translation transformation are carried out to construct vector value shearlet: φ to the basic function in commutative Clifford algebra aspect _{j, k, l}(x), j, k ∈ Z, l ∈ Z ², j and k is direction factor in dilation transformation and shear transformation process and scale factor; L is the translational component in translation transformation process, and Z represents integer set;

Step 2 two, provide Clifford shearlet convert and inverse transformation, provided by formula three and formula four:

Formula three, $<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>{\&Integral;}_{R^2}f\left(x\right)\stackrel{\&OverBar;}{{\mathrm{\&phi;}}_{j,k,l}\left(x\right)}\mathrm{dx}$

Formula four, $f\left(x\right)=\underset{j,k,l}{\mathrm{\&Sigma;}}<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>{\mathrm{\&phi;}}_{j,k,l}\left(x\right)$

Formula three is vector value shearlet: φ _{j, k, l}x () and image function carry out inner product operation, obtain coloured image at different directions, the projection coefficient on different scale and diverse location; Formula four is inverse transformations of formula three, is the Clifford shearlet series expansion of image function; In above-mentioned two formula, x=(x ₁, x ₂) ∈ R ², R represents real number set;

Step 3, utilize formula three to carry out Clifford shearlet conversion to original several coloured images to be fused, provide the multiresolution analysis in Clifford algebra aspect, be i.e. thick scale coefficient τ _cwith thin scale coefficient τ _f, τ _c, described thick scale coefficient τ _cwith thin scale coefficient τ _fvalue comprises the R of coloured image corresponding pixel points simultaneously, the colouring information of G and channel B;

Step 4, on the pixel of correspondence, to the thick scale coefficient of each coloured image to be fused, select the minimum coefficient of norm value as the thick scale coefficient of fused images; To the thin scale coefficient of every piece image, the coefficient selecting norm value maximum is as the thin scale coefficient of fused images; Obtain the multiresolution analysis of fused images; Utilize formula four, inverse Clifford shearlet is carried out to the multiresolution analysis of fused images and converts, obtain the coloured image after merging.

Beneficial effect of the present invention: the invention provides the computing method that a kind of transform domain image merges, namely based on the fusion method that Clifford shearlet converts, breach during conventional color image merges the processing mode of subchannel adopted, but coloured image is regarded as an entirety process.The present invention can solve the cross-color problem existing for fused images effectively, can remove image blur phenomena to greatest extent simultaneously.

Embodiment

Embodiment one, multi-source multi-focus color image fusion method, based on the fusion calculation that Clifford shearlet converts, specific implementation process is:

Step one, source images is carried out modeling with the form of the commutative Clifford of two dimension (Clifford) algebraically, realize the bulk treatment of coloured image.The commutative Clifford algebraically of two dimension in element f there is following expression-form: f=a ₁ε ₁+ a ₂ε ₂+ a ₁₂ε ₁₂, wherein a ₁, a ₂, a ₁₂∈ R.ε _imultiplying meet law of commutation, its operation rule is provided by formula one:

Formula one: ε _iε _j=ε _jε _i, i ≠ j; , i=1,2 ..., n.

Commutative Clifford algebraically is adopted to achieve the modeling of coloured image, i.e. formula two:

Formula two: f (x, y)=f _r(x, y) ε ₁+ f _g(x, y) ε ₂+ f _b(x, y) ε ₁₂

Wherein, f _r(x, y), f _g(x, y), f _b(x, y) is R, G and B color component of coloured image respectively.This step achieves the bulk treatment of coloured image, effectively overcomes the cross-color problem of fused images, and this is also different from traditional fusion method a essence.

Step 2, traditional shearlet conversion be generalized to Clifford algebra aspect, give Clifford shearlet convert definition and discretize algorithm.Described Clifford shearlet conversion is defined as follows:

Step a: a series of dilation transformation, shear transformation and translation transformation are carried out to construct vector value shearlet: φ to the basic function in commutative Clifford algebra aspect _{j, k, l}x (), makes it possess good directivity and multiple dimensioned character.Here, j, k ∈ Z, l ∈ Z ².J and k is direction factor in dilation transformation and shear transformation process and scale factor; L is the translational component in translation transformation process, and Z represents integer set.

Step b: provide Clifford shearlet and convert and inverse transformation, provided by formula three and formula four:

Formula three: $<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>{\&Integral;}_{R^2}f\left(x\right)\stackrel{\&OverBar;}{{\mathrm{\&phi;}}_{j,k,l}\left(x\right)}\mathrm{dx}$

Formula four: $f\left(x\right)=\underset{j,k,l}{\mathrm{\&Sigma;}}<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>{\mathrm{\&phi;}}_{j,k,l}\left(x\right)$

Formula three is vector value shearlet: φ in essence _{j, k, l}x () and image function carry out inner product operation, obtain coloured image at different directions, the projection coefficient on different scale and diverse location.Formula four is inverse transformations of formula three, is the Clifford shearlet series expansion of image function.In above-mentioned two formula, x=(x ₁, x ₂) ∈ R ², R represents real number set.

Step 3, utilize formula three to carry out Clifford shearlet conversion to original several coloured images to be fused, provide the multiresolution analysis in Clifford algebra aspect, be i.e. thick scale coefficient τ _cwith thin scale coefficient τ _f.Here, τ _c, that is, coefficient value comprises the R of coloured image corresponding pixel points simultaneously, the colouring information of G and channel B, this is different from Traditional Wavelet or after other, small echo can only carry out the single channel colouring information embodied in subchannel processing procedure on process coloured image.

Step 4, this step mainly realize the fusion of multi-source coloured image.On the pixel of correspondence, to the thick scale coefficient of each image to be fused, the coefficient selecting norm value minimum is as the thick scale coefficient of fused images; For the thin scale coefficient of every piece image, the coefficient selecting norm value maximum is as the thin scale coefficient of fused images.Like this, the multiresolution analysis of fused images is just obtained.Utilize formula four, namely inverse Clifford shearlet conversion is carried out to the multiresolution analysis of fused images and just obtain the coloured image after merging.

The discretize algorithm concrete steps that Clifford shearlet described in present embodiment step 2 converts are as follows:

First, with the form of commutative Clifford algebra, modeling is carried out to coloured image, obtain matrix I, pseudo-polar coordinates Clifford Fourier is carried out to I and converts, obtain the matrix I after converting _p;

Secondly, to I _pbe multiplied by corresponding weight by coordinate points, obtain matrix I _pw, the computing method of weight convert the same with traditional shearlet.

Finally, to I _pwbe multiplied by corresponding shearlet window function, then carry out inverse pseudo-polar coordinates Clifford Fourier and convert, just obtain Ford shearlet coefficient in a series of gram, i.e. transformation results.

The thick scale coefficient to each image to be fused in present embodiment described in step 4, the coefficient selecting norm value minimum is as the thick scale coefficient of fused images; For the thin scale coefficient of every piece image, the coefficient selecting norm value maximum is as the thin scale coefficient of fused images.Like this, just obtain the multiresolution analysis of fused images, specifically refer to:

The thick scale coefficient supposing the coloured image that original n web merges is τ _c1, τ _c1..., τ _cn, thin scale coefficient is τ _f1, τ _f1..., τ _fn, the thick scale coefficient τ of the coloured image after described fusion _cbe expressed as by formula five:

Formula five, τ _c=min{ τ _c1, τ _c1..., τ _cn,

The thin scale coefficient τ of the coloured image after described fusion _fbe expressed as by formula six:

Formula six, τ _f=max{ τ _f1, τ _f1..., τ _fn.

Claims (1)

1. the fusion method of multi-source multi-focus color image, is characterized in that, to the bulk treatment of coloured image and the application of employing Clifford shearlet conversion, the method is realized by following steps:

Step one, original several coloured images to be fused are carried out modeling with the form of the commutative Clifford algebra of two dimension, realize the bulk treatment to coloured image;

Detailed process is: source images is carried out modeling with the form of the commutative Clifford algebra of two dimension, two-dimentional commutative Clifford algebra in element f there is following expression-form: f=a ₁ε ₁+ a ₂ε ₂+ a ₁₂ε ₁₂, wherein a ₁, a ₂, a ₁₂∈ R, ε _imultiplying meet law of commutation, its operation rule is provided by formula one:

Formula one: ε _iε _j=ε _jε _i, i ≠ j; ε _i ²=ε _i, i is more than or equal to the positive integer that 1 is less than or equal to n;

Adopt commutative Clifford algebra to achieve the modeling of coloured image, be expressed as with formula two:

Formula two: f (x, y)=f _r(x, y) ε ₁+ f _g(x, y) ε ₂+ f _b(x, y) ε ₁₂

In formula, f _r(x, y), f _g(x, y), f _b(x, y) is R, G and B color component of coloured image respectively;

Step 2, traditional shearlet conversion is generalized to Clifford algebra aspect, give definition and discretize algorithm thereof that Clifford shearlet converts, described Clifford shearlet conversion is defined as follows:

Step 2 one, a series of dilation transformation, shear transformation and translation transformation are carried out to construct vector value shearlet: φ to the basic function in commutative Clifford algebra aspect _{j, k, l}(x), j, k ∈ Z, l ∈ Z ², j and k is direction factor in dilation transformation and shear transformation process and scale factor; L is the translational component in translation transformation process, and Z represents integer set;

Step 2 two, provide Clifford shearlet convert and inverse transformation, provided by formula three and formula four:

Formula three, $<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>={\&Integral;}_{R^2}f\left(x\right)\stackrel{\&OverBar;}{{\mathrm{\&phi;}}_{j,k,l}\left(x\right)}\mathrm{dx}$

Formula four, $f\left(x\right)=\underset{j,k,l}{\mathrm{\&Sigma;}}<f\left(x\right),{\mathrm{\&phi;}}_{j,k,l}\left(x\right)>{\mathrm{\&phi;}}_{j,k,l}\left(x\right)$

Formula three is vector value shearlet: φ _{j, k, l}x () and image function carry out inner product operation, obtain coloured image at different directions, the projection coefficient on different scale and diverse location; Formula four is inverse transformations of formula three, is the Clifford shearlet series expansion of image function; In above-mentioned formula three and formula four, x=(x ₁, x ₂) ∈ R ², R represents real number set;

Step 3, utilize formula three to carry out Clifford shearlet conversion to original several coloured images to be fused, provide the multiresolution analysis in Clifford algebra aspect, be i.e. thick scale coefficient τ _cwith thin scale coefficient τ _f, described thick scale coefficient τ _cwith thin scale coefficient τ _fvalue comprises the R of coloured image corresponding pixel points simultaneously, the colouring information of G and channel B;

Step 4, on the pixel of correspondence, to the thick scale coefficient of each coloured image to be fused, select the minimum coefficient of norm value as the thick scale coefficient of fused images; To the thin scale coefficient of every piece image, the coefficient selecting norm value maximum is as the thin scale coefficient of fused images; Obtain the multiresolution analysis of fused images; Utilize formula four, inverse Clifford shearlet is carried out to the multiresolution analysis of fused images and converts, obtain the coloured image after merging.