CN104463849A - Color image registration method based on color invariant and phase correlation - Google Patents

Abstract

The invention relates to a color image registration method based on color invariant and phase correlation, and belongs to the technical field of image processing. According to the color image registration method, firstly, the RGB information of a to-be-registered image is obtained, and the image color invariant is calculated according to the obtained information; secondly, the color invariant of the to-be-registered image is substituted into a phase correlated model under logarithm polar coordinates so that phase correlation conversion can be carried out, registered points can be accurately positioned to the sub pixel level through fitting on the intersection energy spectrum phase. The phase correlated model under the logarithm polar coordinates is adopted, has very high adaptability to the image translation transformation and also has adaptability to the image dimension and rotation transformation. According to the method, a high-precision sub pixel registration result can be obtained for a dimension, rotation and translation transformation color image, and high calculating efficiency and processing speed are achieved.

Description

A kind of coloured image method for registering relevant to phase place based on Color invariants

Technical field

The present invention relates to a kind of coloured image method for registering relevant to phase place based on Color invariants, belong to technical field of image processing.

Background technology

Image registration is the process different images of two width or several same targets being snapped to a coordinate system, and wherein, different images may obtain from the shooting of different time, different angles or different cameral.Image registration is an important directions in image procossing research and apply, and it is the basis of many image applications, such as, and super-resolution image reconstruction, image mosaic and image co-registration.In numerous performance index, registration accuracy is of paramount importance, and sub-pixel precision is the basic demand of many application, maybe can significantly improve the performance of image procossing application.Generally method for registering images can be divided into two large classes: based on the method for registering images in region and the method for registering images of feature based.Wherein, the image registration based on region directly processes image intensity value, and the image registration of feature based then extracts some unique points and then mates from image.

Based in the method for registering images in region, phase correlation method can have good adaptability to the image of rotation, yardstick, translation transformation, and noiseproof feature is good, but does not possess sub-pixel precision grade, and can only registration gray level image.First to carry out gray processing process for coloured image, then registration is carried out to gray level image, because the loss of colouring information can cause error hiding.

Summary of the invention

The object of this invention is to provide a kind of coloured image method for registering relevant to phase place based on Color invariants, because the method for registering that adopts gray processing process to cause colouring information to be lost to cause error hiding and phase place to be correlated with cannot the problem of registration coloured image during to solve existing coloured image registration.

The present invention is for solving the problems of the technologies described above and providing a kind of coloured image method for registering relevant to phase place based on Color invariants, and this method for registering comprises the following steps:

1) RGB information of image subject to registration is obtained, according to the information computed image Color invariants obtained;

2) Color invariants of image subject to registration is brought into the phase place correlation model under log-polar, set up the coordinate transformation model between image subject to registration;

3) cross-power spectrum of color of image invariant subject to registration is calculated;

4) cross-power spectrum is carried out Fu Ye inversion and change to spatial domain, obtain impulse function;

5) utilize Gaussian function paired pulses function to carry out curve fitting, determine Curve Maximization point, this extreme point is exactly the maximum point of cross-correlation and optimum matching coordinate;

6) bring the position coordinates of extreme point into step 2) in the coordinate transformation model set up, determine the coordinate transformation model between image subject to registration with this, according to this model by under image conversion subject to registration to same coordinate.

Described step 1) in Color invariants be:

$H=\frac{E_{\mathrm{\λ}}}{E_{\mathrm{\λ\λ}}}=\frac{\∂E/\∂\mathrm{\λ}}{{\∂}^{2}E/{\∂\mathrm{\λ}}^2}=\frac{\∂R_{\∞}(\mathrm{\λ},x)/\∂\mathrm{\λ}}{{\∂}^2{R}_{\∞}(\mathrm{\λ},x)/\∂{\mathrm{\λ}}^2}=f\left[R_{\∞}(\mathrm{\λ},x)\right]$

Wherein λ represents wavelength, and x is two-dimensional vector, represents observation position, R _∞(λ, x) represents material reflectance, the reflectance spectrum that E (λ, x) is observation place.

Described step 2) in transformation model be:

h ₁(x,y)＝h ₂(σxcosα ₀+σysinα ₀-x ₀,-σxsinα ₀+σycosα ₀-y ₀)

Wherein h ₁(x, y), h ₂(x, y) is the Color invariants of two width images subject to registration, and σ is scaling factor, α 0 is the anglec of rotation, x0, and y0 is the relative translation amount of two width figure.

Described step 3) in the computing formula of cross-power spectrum as follows:

A) to step 2) in transformation model carry out Fourier transform,

$H_1(u,v)=\frac{1}{{\mathrm{\σ}}^2}{e}^{-j2\mathrm{\π}({\mathrm{ux}}_0+{\mathrm{vy}}_0)}{H}_2[\frac{1}{\mathrm{\σ}}(u\cos {\mathrm{\α}}_0+v\sin {\mathrm{\α}}_0),\frac{1}{\mathrm{\σ}}(-u\sin {\mathrm{\α}}_0+v\cos {\mathrm{\α}}_0)];$

B) H1 and H2 modulus value M1 and M2 is calculated,

M ₁(r,α)＝e ^-2kM ₂(r-k,α-α ₀)

Log ρ=r, log σ=k and ρ=e ^r, σ=e ^k;

Wherein

C) Fourier transform and normalized are carried out to modulus value,

$M_1^{\′}(u,v)=M_2^{\′}(u,v)e^{-j2\mathrm{\π}(\mathrm{ur}+v{\mathrm{\α}}_0)-2k}$

$\frac{M_1^{\′}(u,v){M_2^{\′}}^{*}(u,v)}{\left|M_1^{\′}(u,v){M_2^{\′}}^{*}(u,v)\right|}=e^{-j2\mathrm{\π}(\mathrm{ur}+v{\mathrm{\α}}_0)-2k}$

Wherein M ₂ ^*for M ₂conjugation.

The invention has the beneficial effects as follows: first the present invention obtains the RGB information of image subject to registration, according to the information computed image Color invariants obtained; Then by the Color invariants of image subject to registration, the phase place correlation model brought under log-polar carries out phase place correlating transforms, by carrying out matching to its cross power spectrum phase place, registration point accurately can be positioned to sub-pixel.The present invention adopts the phase place correlation model under log-polar, not only there is good adaptability to the translation transformation of image, equally there is applicability to graphical rule and rotational transform, the present invention can obtain high precision subpixel registration result to the coloured image of yardstick, rotation and translation transformation, and has good counting yield and processing speed.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of method for registering images of the present invention;

Fig. 2-a is the original color image adopted in the embodiment of the present invention;

Fig. 2-b is the schematic diagram of original color image invariant H;

Fig. 3 is the schematic diagram of log-polar phase place related algorithm.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described.

The invention provides a kind of coloured image registration Algorithm relevant to phase place based on Color invariants, in order to registration coloured image, Color invariants is incorporated in image registration algorithm, for two images subject to registration, specific implementation process of the present invention is described below.The flow process of the method as shown in figures 1 and 3, specifically comprises the steps.

1. obtain the RGB information of image subject to registration and the invariant of computed image color

Color invariants is a kind of reflection characteristic characterizing object, with viewpoint surface orientation direction of illumination and intensity of illumination all irrelevant, it is one function about position coordinates, adopt Gauss's color model as the universal model representing spectral information and local image structure, to calculate the invariant of color space.The Kubelka-Munk theoretical description spectral characteristics of radiation of object, its model tormulation is:

E(λ,x)＝e(λ，x)[1-ρ _f(x)] ²+e(λ，x)ρ _f(x) (1)

Wherein λ represents wavelength, and x is two-dimensional vector, and represent observation position, e (λ, x) represents spectral intensity, ρ _fx () represents the Fresnel reflection coefficient at x place, R _∞(λ, x) represents material reflectance, the reflectance spectrum that E (λ, x) is observation place.

In most cases, e (λ, x) remains unchanged and relevant with position on each wavelength, and E (λ, x) is rewritten into i (x) form, then (1) formula becomes:

E(λ,x)＝i(x){[1-ρ _f(x)] ²R _∞(λ,x)+ρ _f(x)} (2)

First order derivative and second derivative are asked respectively to λ, are then divided by:

$H=\frac{E_{\mathrm{\λ}}}{E_{\mathrm{\λ\λ}}}=\frac{\∂E/\∂\mathrm{\λ}}{{\∂}^{2}E/{\∂\mathrm{\λ}}^2}=\frac{\∂R_{\∞}(\mathrm{\λ},x)/\∂\mathrm{\λ}}{{\∂}^2{R}_{\∞}(\mathrm{\λ},x)/\∂{\mathrm{\λ}}^2}=f\left[R_{\∞}(\mathrm{\λ},x)\right]---\left(3\right)$

H is the one statement of Color invariants, same to observation place, surface towards, light intensity magnitude, reflection coefficient be all irrelevant.Under the condition meeting human visual system and CIE-1964-XYZ standard, the RGB component of coloured image and E (E, E _λ, E _{λ λ}) relation be approximately:

$\left[\begin{array}{c}E\\ E_{\mathrm{\λ}}\\ E_{\mathrm{\λ\λ}}\end{array}\right]=\left[\begin{array}{ccc}0.06& 0.63& 0.27\\ 0.30& 0.04& -0.35\\ 0.34& -0.60& 0.17\end{array}\right]\×\left[\begin{array}{c}R\\ G\\ B\end{array}\right]---\left(4\right)$

The Color invariants H of coloured image can be tried to achieve by (4) formula and (3) formula.

The present embodiment adopts the original color image of Fig. 2-a to be example, and after above-mentioned calculating, the result of calculation of the colored invariant of this image is as shown in Fig. 2-b.

2. the phase place correlation model set up under log-polar carries out phase place correlating transforms

Suppose to there is rotation, translation, change of scale between two width images subject to registration, as shown in Figure 3, h ₁(x, y), h ₂(x, y) is the Color invariants of two width images, then h ₁(x, y) and h ₂the relation of (x, y) can be expressed as:

h ₁(x,y)＝h ₂(σxcosα ₀+σysinα ₀-x ₀,-σxsinα ₀+σycosα ₀-y ₀) (5)

In formula, σ is scaling factor, α ₀for the anglec of rotation, x ₀, y ₀be the relative translation amount of two width figure, (5) Fourier transform obtained:

$H_1(u,v)=\frac{1}{{\mathrm{\σ}}^2}{e}^{-j2\mathrm{\π}({\mathrm{ux}}_0+{\mathrm{vy}}_0)}{H}_2[\frac{1}{\mathrm{\σ}}(u\cos {\mathrm{\α}}_0+v\sin {\mathrm{\α}}_0),\frac{1}{\mathrm{\σ}}(-u\sin {\mathrm{\α}}_0+v\cos {\mathrm{\α}}_0)]---\left(6\right)$

If M ₁, M ₂for H ₁, H ₂modulus value, then:

$M_1(u,v)=\frac{1}{{\mathrm{\σ}}^2}{M}_2\left[\frac{1}{\mathrm{\σ}}\left(\begin{array}{cc}u\cos {\mathrm{\α}}_0+v\sin {\mathrm{\α}}_0,& -u\sin {\mathrm{\α}}_0+v\cos {\mathrm{\α}}_0\end{array}\right)\right]---\left(7\right)$

Carry out coordinate transform to formula (7) to obtain:

$M_1(\mathrm{\ρ}\cos \mathrm{\α},\mathrm{\ρ}\sin \mathrm{\α})=\frac{1}{{\mathrm{\σ}}^2}{M}_2\left[\frac{1}{\mathrm{\σ}}\left(\begin{array}{cc}\mathrm{\ρ}\cos (\mathrm{\α}-{\mathrm{\α}}_0),& \mathrm{\ρ}\sin (\mathrm{\α}-{\mathrm{\α}}_0)\end{array}\right)\right]---\left(8\right)$

Namely $M_1(\mathrm{\ρ},\mathrm{\α})=\frac{1}{{\mathrm{\α}}^2}{M}_2(\frac{\mathrm{\ρ}}{\mathrm{\σ}},\mathrm{\α}-{\mathrm{\α}}_0)---\left(9\right)$

Make log ρ=r, log σ=k and ρ=e ^r, σ=e ^kthen formula (9) can turn to:

M ₁(r,α)＝e ^-2kM ₂(r-k,α-α ₀) (10)

3. obtain the cross-power spectrum of two width color of image invariants

Carry out Fourier transform to formula (10) to obtain:

$M_1^{\′}(u,v)=M_2^{\′}(u,v)e^{-j2\mathrm{\π}(\mathrm{ur}+v{\mathrm{\α}}_0)-2k}---\left(11\right)$

Normalized power spectrum is:

$\frac{M_1^{\′}(u,v){M_2^{\′}}^{*}(u,v)}{\left|M_1^{\′}(u,v){M_2^{\′}}^{*}(u,v)\right|}=e^{-j2\mathrm{\π}(\mathrm{ur}+v{\mathrm{\α}}_0)-2k}---\left(12\right)$

Wherein M ₂ ^*for M ₂conjugation, wherein r=log ρ, k=log σ, α ₀for the anglec of rotation.

4. inverse fourier transform, and carry out curve fitting

Normalized cross-power spectrum is an exponential function, through inverse fourier transform to spatial domain, obtains impulse function δ (x-r, y-α ₀), utilize Gaussian function paired pulses function to carry out curve fitting, find Curve Maximization point, the point that cross-correlation is maximum i.e. optimum matching coordinate, the position coordinates of extreme point is exactly (r, α ₀) value, and scale factor σ can be obtained further.

5. locate the extreme point of sub-pix, obtain converting parameter.

The value of anglec of rotation α 0 and scale factor σ is substituted into formula (6), and be just converted into the form of simple translation relationship consistency, adopting uses the same method can calculate translation parameters.According to the anglec of rotation, scale factor and translation parameters, the transformation model between two width images can be determined, just can by under two width image conversions to same coordinate.

Claims (4)

1. based on the coloured image method for registering that Color invariants is relevant to phase place, it is characterized in that, this method for registering comprises the following steps:

1) RGB information of image subject to registration is obtained, according to the information computed image Color invariants obtained;

2) Color invariants of image subject to registration is brought into the phase place correlation model under log-polar, set up the coordinate transformation model between image subject to registration;

3) cross-power spectrum of color of image invariant subject to registration is calculated;

4) cross-power spectrum is carried out Fu Ye inversion and change to spatial domain, obtain impulse function;

5) utilize Gaussian function paired pulses function to carry out curve fitting, determine Curve Maximization point, this extreme point is exactly the maximum point of cross-correlation and optimum matching coordinate;

6) bring the position coordinates of extreme point into step 2) in the coordinate transformation model set up, determine the coordinate transformation model between image subject to registration with this, according to this model by under image conversion subject to registration to same coordinate.

2., according to claim 1 based on the coloured image method for registering that Color invariants is relevant to phase place, it is characterized in that, described step 1) in Color invariants be:

$H=\frac{E_{\mathrm{\λ}}}{E_{\mathrm{\λ\λ}}}=\frac{\∂E/\∂\mathrm{\λ}}{{\∂}^{2}E/{\∂\mathrm{\λ}}^2}=\frac{{\∂R}_{\∞}(\mathrm{\λ},x)/\∂\mathrm{\λ}}{{\∂}^2{R}_{\∞}(\mathrm{\λ},x)/{\∂\mathrm{\λ}}^2}=f\left[R_{\∞}(\mathrm{\λ},x)\right]$

Wherein λ represents wavelength, and x is two-dimensional vector, represents observation position, R _∞(λ, x) represents material reflectance, the reflectance spectrum that E (λ, x) is observation place.

3. the coloured image method for registering relevant to phase place based on Color invariants according to claim 2, is characterized in that, described step 2) in transformation model be:

h ₁(x,y)＝h ₂(σxcosα ₀+σysinα ₀-x ₀,-σxsinα ₀+σycosα ₀-y ₀)

Wherein h ₁(x, y), h ₂(x, y) is the Color invariants of two width images subject to registration, and σ is scaling factor, α 0 is the anglec of rotation, x0, and y0 is the relative translation amount of two width figure.

4. the coloured image method for registering relevant to phase place based on Color invariants according to claim 3, is characterized in that, described step 3) in the computing formula of cross-power spectrum as follows:

A) to step 2) in transformation model carry out Fourier transform,

$H_1(u,v)=\frac{1}{{\mathrm{\σ}}^2}{e}^{-j2\mathrm{\π}({\mathrm{ux}}_0+{\mathrm{vy}}_0)}{H}_2[\frac{1}{\mathrm{\σ}}(u\cos {\mathrm{\α}}_0+v\sin {\mathrm{\α}}_0),\frac{1}{\mathrm{\σ}}({-u\sin \mathrm{\α}}_0+v\cos {\mathrm{\α}}_0)];$

B) H1 and H2 modulus value M1 and M2 is calculated,

M ₁(r,α)＝e ^-2kM ₂(r-k,α-α ₀)

Log ρ=r, log σ=k and ρ=e ^r, σ=e ^k;

Wherein

C) Fourier transform and normalized are carried out to modulus value,

$M_1^{\′}(u,v)=M_2^{\′}(u,v)e^{-j2\mathrm{\π}(\mathrm{ur}+{\mathrm{v\α}}_0)-2k}$

$\frac{M_1^{\′}(u,v)M_2^{{\′}^{*}}(u,v)}{\left|M_1^{\′}(u,v)M_2^{{\′}^{*}}(u,v)\right|}=e^{-j2\mathrm{\π}(\mathrm{ur}+{\mathrm{v\α}}_0)-2k}$

Wherein M ₂ ^*for M ₂conjugation.