CN103268596A - Method for reducing image noise and enabling colors to be close to standard - Google Patents

Abstract

The invention provides a method for reducing image noise and enabling colors to be close to a standard. Under the working scene of a precise electronic assembling device, images are obtained through multiple camera lenses, the image noise is reduced, and the colors are made to be close to the standard. The method carries out corresponding point matching on two shot images by the utilization of homography matrixes, then one image is used as a manuscript to carry out fusion on images of a public area, and noise of the images is reduced. Meanwhile, a method that colors correct the matrixes is used for carrying out global color correction on the whole fused image, and the color of the image is made to be more close to a standard color.

Description

A kind of method that reduces picture noise and color is near the mark

Technical field

The present invention relates to the image processing field of precise electronic mounting equipment, be specifically related to a kind of under precise electronic mounting equipment operative scenario, the method that reduces picture noise and color is near the mark.

Background technology

The classical noise-reduction method at single image is spatial domain filtering and frequency filtering.Spatial domain filtering is to utilize template directly pixel value to be carried out convolution algorithm at former figure, comprises mean filter, medium filtering, low-pass filtering.Frequency filtering is to utilize Fourier transform that former figure is changed to frequency field from transform of spatial domain, and the image coefficient of adjusting different frequency then removes noise, becomes image again spatial domain from frequency field at last.In addition, utilize methods such as wavelet transform filtering, partial differential equation, the variational method, morphologic filtering to carry out image denoising in addition.

In fact, the noise-reduction method at single image can cause losing of image detail.

Summary of the invention

The objective of the invention is to overcome the deficiencies in the prior art, provide a kind of under precise electronic mounting equipment operative scenario, reduce picture noise and the method that color is near the mark with many camera lenses.

In order to realize goal of the invention, the technical solution used in the present invention is: adopt many lens shootings to obtain two bigger width of cloth printed circuit board images of public domain, synthetic the public domain weighting of two figure as draft with an image then, utilize the randomness of noise to reduce the noise of image, by the color correction matrix color is near the mark at last.Utilize the synthetic scheme that reduces noise of two figure can keep the details of image preferably.

A kind of method that reduces picture noise and color is near the mark, under precise electronic mounting equipment operative scenario, adopt many lens shootings to obtain two bigger width of cloth printed circuit board images of public domain, then so that wherein an image is synthetic the public domain weighting of two width of cloth printed circuit board images as draft, utilize the randomness of noise to reduce the noise of image, by the color correction matrix color is near the mark at last.

Above-mentioned reduction picture noise and the method that color is near the mark is characterized in that specifically comprising the steps:

(1) two width of cloth image public domain weighting is merged: select a image in two width of cloth printed circuit board images as draft, utilize homography matrix another image to be transformed into the image that corresponding element is arranged in the public domain with draft, public domain to two width of cloth images is weighted then, obtain new image, thereby reduce picture noise;

(2) colour correction: take advantage of 3 matrixing by 3, original RGB color space in the new image is mapped to the RGB color space of standard, make the rgb pixel value that the rgb pixel value of image more is near the mark.

Two captured width of cloth printed circuit board images are the image on the same plane of precise electronic mounting equipment operative scenario in the above-mentioned steps (1); Manually or automatically choose n group (being no less than 4 groups) corresponding point on the same plane, obtain the coordinate of corresponding point, utilize direct linear transformation's algorithm to solve the homography matrix of two image correspondences.

In the above-mentioned steps (1), another image is transformed into described draft to be had in the image process of corresponding element in the public domain, in order to improve arithmetic speed, the square frame of per 16 pixel *, 16 pixels uses homography matrix to obtain 4 groups of summit corresponding point coordinates (these 4 groups of corresponding point coordinates are the summit of square frame), and all the other corresponding point coordinates obtain with the method for bilinear interpolation and 4 groups of corresponding point coordinates having asked.

In the above-mentioned steps (1), another image is transformed into described draft to be had in the image process of corresponding element in the public domain, the pixel value of new images will obtain at former figure by the corresponding point coordinate, need utilize bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous coordinates of impact point.

Described direct linear transformation's algorithm comprises: with two groups of corresponding point coordinate X _lAnd X _rProcess $T_l=\left(\begin{array}{ccc}1/D\left(x_l\right)& 0& -E\left(x_l\right)\\ 0& 1/D\left(y_l\right)& -E\left(y_l\right)\\ 0& 0& 1\end{array}\right)$ With $T_r=\left(\begin{array}{ccc}1/D\left(x_r\right)& 0& -E\left(x_r\right)\\ 0& 1/D\left(y_r\right)& -E\left(y_r\right)\\ 0& 0& 1\end{array}\right)$ Normalization obtains X _LgAnd X _Rg, (subscript l and r represent left figure and right figure, X respectively _lAnd X _rRepresent the original corresponding point coordinate of Zuo Tu and right figure respectively, X _LgAnd X _RgRepresent corresponding point coordinate after the normalization of Zuo Tu and right figure respectively.x _lAnd x _rRepresent the x coordinate of Zuo Tu and right figure respectively, y _lAnd y _rRepresent the x coordinate of Zuo Tu and right figure respectively, E (x _l) and E (x _r) be respectively the expectation of x coordinate before left figure and the right figure normalization, E (y _l) and E (y _r) be the expectation of y coordinate before left figure and the right figure normalization, D (x _l) and D (x _r) be respectively the variance of x coordinate before left figure and the right figure normalization, D (y _l) and D (y _r) be respectively the variance of y coordinate before left figure and the right figure normalization); Utilize X _i' * HX _i=0 is organized into Ah=0 gets $\left[\begin{array}{ccc}{0}^T& {-X}_i^T& y_i^{\′}{X}_i^T\\ X_i^T& 0^T& {-x}_i^{\′}{X}_i^T\\ {-y}_i^{\′}{X}_i^T& x_i^{\′}{X}_i^T& 0^T\end{array}\right]\left[\begin{array}{c}{h}^1\\ h^2\\ h^3\end{array}\right]=0,$ H wherein ⁱRepresent the 3 capable transposition of i of taking advantage of 3 homography matrixes, $X_i^{\′}=\left[\begin{array}{c}{x}_i^{\′}\\ y_i^{\′}\\ 1\end{array}\right],$ $X_i=\left[\begin{array}{c}{x}_i\\ y_i\\ 1\end{array}\right];$ 3n taken advantage of 9 A matrix (n refers to corresponding point group number) carry out svd A=UDV ^T, homography matrix H after the normalization _gBe last row of V, anti-normalized homography matrix

Described bilinear interpolation method comprises: interpolation is carried out one by one according to the square frame of 16 pixel *, 16 pixels, f (1), and f (2), f (3), f (4) refers to the mapping point of four vertex positions of square frame respectively, in each square frame inside, utilizes formula f ₁₂(k, 1)=(f (1) * (16-k)+f (2) * k)〉〉 2, f ₂₃(16, k)=(f (2) * (16-k)+f (3) * k)〉〉 2, f ₃₄(k, 16)=(f (3) * (16-k)+f (4) * k)〉〉 3, f ₄₁(1, k)=(f (4) * (16-k)+f (1) * k)〉〉 3 first interpolation four edges, utilize then formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 interior zone of 12 interpolation square frames; Wherein, subscript represents the mark on first summit and second summit on interpolation limit, k representative be positioned on the interpolation limit and with this interpolation limit on first vertex distance be the coordinate of k-1 pixel, the k value is 2～15; X, y represent the coordinate in the square frame, and value is 2～15.

Described bilinear interpolation method utilizes bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous corresponding point coordinates of impact point; Utilize formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 12, f (1), f (2), f (3), f (4) refers to the pixel value of 4 the most contiguous coordinates respectively.

Described step (2) comprising: only need do asking for of a color correction matrix in this scene, just use this matrix as color correction later on always; Two video cameras are taken 24 colour standard palettes earlier under operative scenario, palette will utilize the described noise-reduction method of step (1) to obtain composograph in the public domain of two width of cloth figure then.

The method that in the step (2) color is near the mark comprises: as input, the standard pixel value of the corresponding color lump of palette is as output, formula with 5 pixel average reading 24 color lumps on the composograph $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccc}{c}_1& c_2& c_3\\ c_4& c_5& c_6\\ c_7& c_8& c_9\end{array}\right]*\left[\begin{array}{c}{R}_{\mathrm{in}}\\ G_{\mathrm{in}}\\ B_{\mathrm{in}}\end{array}\right]$ Be organized into the form (R of Ax=b _In, G _In, B _InBe respectively redness, the green of input, the color value of blue channel matrix, R _In, G _In, B _InBe respectively redness, the green of standard color block, the color value of blue channel matrix), get formula $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccccccccc}{R}_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}\end{array}\right]*\left[\begin{array}{c}{c}^1\\ c^2\\ c^3\end{array}\right]$ (namely $b=A*\left[\begin{array}{c}{c}^1\\ c^2\\ c^3\end{array}\right]$ ), wherein, c ⁱBe to represent 3 to take advantage of the capable transposition of 3 color correction matrix i, the matrix A size is 72*9, and the size of vectorial b is 72*1; Utilize least square method x=(A ^T* A) ^-1* A ^TB or svd can obtain needed transformation matrix.

Compared with prior art, the present invention has following advantage and effect:

The multi-cam image co-registration reduction picture noise that the present invention adopts can reduce classical single image filtering denoising and lose details, utilizes color correction matrix that color is near the mark, and actual production meaning is arranged.

Description of drawings

Fig. 1 is that the present invention reduces noise and makes the color general flow chart that is near the mark

Fig. 2 asks for the homography matrix flow process

Fig. 3 adds full synthetic schemes with an image conversion and with another image

Fig. 4 asks for color correction matrix process flow diagram

Embodiment

Below in conjunction with accompanying drawing enforcement of the present invention is described further, but enforcement of the present invention and protection domain are not limited thereto.

A kind of method that reduces picture noise and color is near the mark specifically comprises following five key steps:

(1) manually or automatically chooses and be no less than 4 groups of corresponding point on the same plane, obtain the coordinate of corresponding point, utilize direct linear transformation's algorithm to solve the homography matrix of two image correspondences.

Direct linear transformation's algorithm: with two groups of corresponding point coordinate X _lAnd X _rProcess $T_l=\left(\begin{array}{ccc}1/D\left(x_l\right)& 0& -E\left(x_l\right)\\ 0& 1/D\left(y_l\right)& -E\left(y_l\right)\\ 0& 0& 1\end{array}\right)$ With $T_r=\left(\begin{array}{ccc}1/D\left(x_r\right)& 0& -E\left(x_r\right)\\ 0& 1/D\left(y_r\right)& -E\left(y_r\right)\\ 0& 0& 1\end{array}\right)$ Normalization obtains X _LgAnd X _Rg, E (x) is the expectation of coordinate before the normalization, D (x) is the variance of coordinate before the normalization.Utilize X _i' * HX _i=0 is organized into Ah=0 gets $\left[\begin{array}{ccc}{0}^T& {-X}_i^T& y_i^{\′}{X}_i^T\\ X_i^T& 0^T& {-x}_i^{\′}{X}_i^T\\ {-y}_i^{\′}{X}_i^T& x_i^{\′}{X}_i^T& 0^T\end{array}\right]\left[\begin{array}{c}{h}^1\\ h^2\\ h^3\end{array}\right]=0,$ H wherein ⁱRepresent the 3 capable transposition of i of taking advantage of 3 homography matrixes, $X_i^{\′}=\left[\begin{array}{c}{x}_i^{\′}\\ y_i^{\′}\\ 1\end{array}\right],$ $X_i=\left[\begin{array}{c}{x}_i\\ y_i\\ 1\end{array}\right].$ 3n taken advantage of 9 A matrix carry out svd A=UDV ^T, homography matrix H after the normalization _gBe last row of V, anti-normalized homography matrix

(2) corresponding point of image in the calculating public domain, in order to improve arithmetic speed, just the square frame of per 16 pixel *, 16 pixels uses homography matrix to obtain 4 groups of corresponding point coordinates, all the other corresponding point coordinates obtain with method and 4 groups of corresponding point coordinates of bilinear interpolation.

Bilinear interpolation method carries out one by one according to the square frame of 16 pixel *, 16 pixels, f (1), and f (2), f (3), f (4) refers to the mapping point of four vertex positions of square frame respectively.In each square frame inside, utilize formula f ₁₂(k, 1)=(f (1) * (16-k)+f (2) * k)〉〉 2, f ₂₃(16, k)=(f (2) * (16-k)+f (3) * k)〉〉 2, f ₃₄(k, 16)=(f (3) * (16-k)+f (4) * k)〉〉 3, f ₄₁(1, k)=(f (4) * (16-k)+f (1) * k)〉〉 3 first interpolation four edges, utilize then formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 interior zone of 12 interpolation square frames.Subscript represents wherein two summits of square frame, and as subscript 12 representative with first summit 1 on the interpolation limit that is end points, these two summits and the second summit 2(and for example, subscript 41 represents first summit 4 and second summit 1 on interpolation limit), f ₁₂(k, 1) representative is the mapping point of the point of k-1 pixel with the distance on first summit 1 with first summit 1 and second summit 2 on the interpolation limit that be end points formation; f _Ab(k, 1) representative is being to be the mapping point of the point of k-1 pixel with the distance of the first summit a on the interpolation limit that constitutes of end points with the first summit a and the second summit b, k representative be positioned on the interpolation limit and with this interpolation limit on the distance of first summit a be that (k is coordinate, f for the coordinate of k-1 pixel _Ab(k, 1) is mapping point, and coordinate is the coordinate system of setting up in square frame, and mapping point is the coordinate system of setting up in complete big figure), value is 2～15; X, y represent the coordinate in the square frame, and value is 2～15.

(3) another image is transformed into the image that corresponding element is arranged in the public domain with draft, then that two figure weightings are synthetic.The pixel value of new figure will obtain at former figure by the corresponding point coordinate, because the corresponding point coordinate may have decimal, so need utilize bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous coordinates of impact point.

Utilize bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous corresponding point coordinates of impact point.Utilize formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 12, f (1), f (2), f (3), f (4) refers to the pixel value of 4 the most contiguous coordinates respectively.

(4) only do asking for of a color correction matrix in operative scenario, just use this matrix as color correction later on always.Two video cameras are taken 24 colour standard palettes earlier under operative scenario, palette will utilize the above-mentioned steps composograph then in the public domain of two width of cloth figure.

As input, the standard pixel value of the corresponding color lump of palette is as output, formula with 5 pixel average reading 24 color lumps on the figure $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccc}{c}_1& c_2& c_3\\ c_4& c_5& c_6\\ c_7& c_8& c_9\end{array}\right]*\left[\begin{array}{c}{R}_{\mathrm{in}}\\ G_{\mathrm{in}}\\ B_{\mathrm{in}}\end{array}\right]$ Be organized into the form of Ax=b, get formula $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccccccccc}{R}_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}\end{array}\right]*\left[\begin{array}{c}{c}^1\\ c^2\\ c^3\end{array}\right].$ Wherein, c ⁱBe to represent 3 to take advantage of the capable transposition of 3 color correction matrix i, the matrix A size is 72*9, and the size of vectorial b is 72*1.Utilize least square method x=(A ^T* A) ^-1* A ^TB or svd can obtain needed transformation matrix.

(5) composograph that obtains each time obtains new image through color correction.

As Fig. 1, for reduction noise of the present invention with make the color process flow diagram that is near the mark.Be specially and take two chip mounter operative scenario images that the public domain is bigger earlier, utilize direct linear transformation's algorithm to ask for homography matrix, obtaining the image after the wherein figure conversion again, is that draft is synthetic with two figure weightings with another figure, obtains final image by color correction at last.The color correction matrix only need just once be found the solution and can be obtained.

Be illustrated in figure 2 as and find the solution the homography matrix process flow diagram.The first step is to be no less than 4 groups of corresponding point by automatic or manual choosing, and corresponding point preferably can be evenly distributed on the image each several part; Second step was that corresponding point are carried out normalization, in order to improve the precision of finding the solution homography matrix; The 3rd step was the homography matrix that utilizes after svd obtains normalization; The 4th step was that the matrix to previous step carries out anti-normalization, obtained the homography matrix of original respective coordinates.

Be illustrated in figure 3 as with an image conversion and with another image and add full synthetic schemes.The first step is the square frame that image is divided into a plurality of 16 pixel *, 16 pixels, utilizes homography matrix to calculate the corresponding point coordinate at all 4 angles of square frame; Second step was to utilize the method for bilinear interpolation to ask for all corresponding point coordinates of public domain; The 3rd step was the image that utilizes after corresponding point coordinate and bilinear interpolation obtain a wherein figure conversion; The 4th step was to be draft with another image, and public domain image weighting is synthetic, obtained the image of noise reduction.

As shown in Figure 4 to asking for color correction matrix process flow diagram.The first step is to take two images that palette is arranged in the public domain earlier; Second step was with said method that two image weightings are synthetic; The 3rd the step be the rgb pixel value of 24 color lumps of composograph as input, wherein each color lump is got 5 mean value, the standard rgb pixel value of palette as output; The 4th step was to utilize least square method or svd to obtain the color correction matrix.

This example is taken action under bright environment image is as standard, obtains identical image and the image of palette then under the environment of dark (being that noise is more) slightly, and uses this method to handle.

The evaluation function of noise is: $\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}[{(s_R^i-t_R^i)}^2+{(s_G^i-t_G^i)}^2+{(s_B^i-t_B^i)}^2].$ Wherein, s and t represent the standard noise-free picture respectively and the image of making an uproar, subscript i=1, and 2 ..., N, N are the image roads.If MSE is more little, then noise is more little.

The evaluation function of color standard degree is: $q=\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}[{(R_{\mathrm{in}}^j-R_{\mathrm{out}}^j)}^2+{(G_{\mathrm{in}}^j-G_{\mathrm{out}}^j)}^2+{(B_{\mathrm{in}}^j-B_{\mathrm{out}}^j)}^2].$ Wherein, R, G, B be three passages of representative image respectively, and subscript in and out represent standard value and measured value to be checked respectively, and subscript j represents j color lump, and the sum of m representative color piece gets 24 herein.If q is more little, color more is near the mark.

The color standard degree of original shooting palette image is 17171.4 in this example, adjust the standardized pallet image (the color standard degree is 0) that obtains with PS, proofread and correct the palette image (the color standard degree is 5618.4) that obtains with asking color correction matrix, account for color more is near the mark.

The MSE of captured left figure is 7463.9 in the example, and the MSE of the right figure of shooting is 4433.5, and the MSE of synthetic noise reduction image is 1338.5.

Claims (9)

1. one kind is reduced picture noise and the method that color is near the mark, it is characterized in that under precise electronic mounting equipment operative scenario, adopt many lens shootings to obtain two bigger width of cloth printed circuit board images of public domain, then so that wherein an image is synthetic the public domain weighting of two width of cloth printed circuit board images as draft, utilize the randomness of noise to reduce the noise of image, make the color of image color that more is near the mark by the color correction matrix at last.

2. the method for reduction picture noise according to claim 1 and raising color sharpness is characterized in that specifically comprising the steps:

(1) two width of cloth image public domain weighting is merged: select a image in two width of cloth printed circuit board images as draft, utilize homography matrix another image to be transformed into the image that corresponding element is arranged in the public domain with draft, public domain to two width of cloth images is weighted then, obtain new image, thereby reduce picture noise;

(2) colour correction: take advantage of 3 matrixing by 3, original RGB color space in the new image is mapped to the RGB color space of standard, make the rgb pixel value that the rgb pixel value of image more is near the mark.

3. reduction picture noise according to claim 2 and the method that color is near the mark is characterized in that two captured in the step (1) width of cloth printed circuit board images are the image on the same plane of precise electronic mounting equipment operative scenario; Manually or automatically choose on the same plane n group corresponding point, obtain the coordinate of corresponding point, utilize direct linear transformation's algorithm to solve the homography matrix of two image correspondences, n 〉=4.

4. reduction picture noise according to claim 2 and method that color is near the mark, it is characterized in that in the step (1), another image is transformed into described draft to be had in the image process of corresponding element in the public domain, in order to improve arithmetic speed, image is drawn the square frame that divides a plurality of 16 pixel *, 16 pixels, the square frame of per 16 pixel *, 16 pixels uses homography matrix to obtain 4 prescription frame summit corresponding point coordinates, and all the other corresponding point coordinates obtain with the method for bilinear interpolation and 4 groups of corresponding point coordinates having asked.

5. reduction picture noise according to claim 2 and method that color is near the mark, it is characterized in that in the described step (1), another image is transformed into described draft to be had in the image process of corresponding element in the public domain, the pixel value of new images will obtain at former figure by the corresponding point coordinate, need utilize bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous coordinates of impact point.

6. reduction picture noise according to claim 3 and the method that color is near the mark is characterized in that described direct linear transformation's algorithm comprises: with two groups of corresponding point coordinate X _lAnd X _rProcess $T_l=\left(\begin{array}{ccc}1/D\left(x_l\right)& 0& -E\left(x_l\right)\\ 0& 1/D\left(y_l\right)& -E\left(y_l\right)\\ 0& 0& 1\end{array}\right)$ With $T_r=\left(\begin{array}{ccc}1/D\left(x_r\right)& 0& -E\left(x_r\right)\\ 0& 1/D\left(y_r\right)& -E\left(y_r\right)\\ 0& 0& 1\end{array}\right)$ Normalization obtains X _LgAnd X _Rg, wherein subscript l and r represent captured left figure and right figure, X respectively _lAnd X _rRepresent the original corresponding point coordinate of Zuo Tu and right figure respectively, X _LgAnd X _RgRepresent corresponding point coordinate after the normalization of Zuo Tu and right figure respectively; x _lAnd x _rRepresent the x coordinate of Zuo Tu and right figure respectively, y _lAnd y _rRepresent the x coordinate of Zuo Tu and right figure respectively, E (x _l) and E (x _r) be respectively the expectation of x coordinate before left figure and the right figure normalization, E (y _l) and E (y _r) be the expectation of y coordinate before left figure and the right figure normalization, D (x _l) and D (x _r) be respectively the variance of x coordinate before left figure and the right figure normalization, D (y _l) and D (y _r) be respectively the variance of y coordinate before left figure and the right figure normalization; Utilize X _i' * HX _i=0 is organized into Ah=0 gets $\left[\begin{array}{ccc}{0}^T& {-X}_i^T& y_i^{\′}{X}_i^T\\ X_i^T& 0^T& {-x}_i^{\′}{X}_i^T\\ {-y}_i^{\′}{X}_i^T& x_i^{\′}{X}_i^T& 0^T\end{array}\right]\left[\begin{array}{c}{h}^1\\ h^2\\ h^3\end{array}\right]=0,$ H wherein ⁱRepresent the 3 capable transposition of i of taking advantage of 3 homography matrixes, $X_i^{\′}=\left[\begin{array}{c}{x}_i^{\′}\\ y_i^{\′}\\ 1\end{array}\right],$ $X_i=\left[\begin{array}{c}{x}_i\\ y_i\\ 1\end{array}\right];$ 3n is taken advantage of 9 A matrix, and n refers to corresponding point group number, carries out svd A=UDV ^T, homography matrix H after the normalization _gBe last row of V, anti-normalized homography matrix

7. according to claim 4 or 5 described reduction picture noises and the method that color is near the mark, it is characterized in that described bilinear interpolation method comprises: interpolation is carried out one by one according to the square frame of 16 pixel *, 16 pixels, f (1), f (2), f (3), f (4) refers to the mapping point of four vertex positions of square frame respectively in turn, in each square frame inside, utilizes formula f ₁₂(k, 1)=(f (1) * (16-k)+f (2) * k)〉〉 2, f ₂₃(16, k)=(f (2) * (16-k)+f (3) * k)〉〉 2, f ₃₄(k, 16)=(f (3) * (16-k)+f (4) * k)〉〉 3, f ₄₁(1, k)=(f (4) * (16-k)+f (1) * k)〉〉 3 first interpolation four edges, utilize then formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 interior zone of 12 interpolation square frames; Wherein, subscript represents the mark on first summit and second summit on interpolation limit, k representative be positioned on the interpolation limit and with this interpolation limit on first vertex distance be the coordinate of k-1 pixel, the k value is 2～15; X, y represent the coordinate in the square frame, and value is 2～15.

8. according to claim 4 or 5 described reduction picture noises and method that color is near the mark, it is characterized in that described bilinear interpolation method utilizes bilinear interpolation method to obtain pixel value under the needed rounded coordinate at the pixel value of 4 the most contiguous corresponding point coordinates of impact point; Utilize formula f (x, y)=(f (1) * (64-x) * (64-y)+f (2) * (x) * (64-y)+f (3) * (64-x) * (y)+f (4) * x*y)〉〉 12, f (1), f (2), f (3), f (4) refers to the pixel value of 4 the most contiguous coordinates respectively.

9. reduction picture noise according to claim 2 and method that color is near the mark, it is characterized in that, it is characterized in that in the described step (2), only need do asking for of a color correction matrix in this scene, just use this matrix as color correction later on always; Two video cameras are taken 24 colour standard palettes earlier under operative scenario, palette will utilize the described noise-reduction method of step (1) to obtain composograph in the public domain of two width of cloth figure then.10, the method for reduction picture noise according to claim 2 and raising color sharpness, it is characterized in that the method that improves the color sharpness in the step (2) comprises: will read 5 pixel average of 24 color lumps on the composograph as input, the standard pixel value of the corresponding color lump of palette is as output, formula $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccc}{c}_1& c_2& c_3\\ c_4& c_5& c_6\\ c_7& c_8& c_9\end{array}\right]*\left[\begin{array}{c}{R}_{\mathrm{in}}\\ G_{\mathrm{in}}\\ B_{\mathrm{in}}\end{array}\right]$ Be organized into the form of Ax=b, wherein R _In, G _In, B _InBe respectively redness, the green of input, the color value of blue channel matrix, R _In, G _In, B _InBe respectively redness, the green of standard color block, the color value of blue channel matrix), get formula $\left[\begin{array}{c}{R}_{\mathrm{out}}\\ G_{\mathrm{out}}\\ B_{\mathrm{out}}\end{array}\right]=\left[\begin{array}{ccccccccc}{R}_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& R_{\mathrm{in}}& G_{\mathrm{in}}& B_{\mathrm{in}}\end{array}\right]*\left[\begin{array}{c}{c}^1\\ c^2\\ c^3\end{array}\right],$ Namely $b=A*\left[\begin{array}{c}{c}^1\\ c^2\\ c^3\end{array}\right],$ Wherein, c ⁱBe to represent 3 to take advantage of the capable transposition of 3 color correction matrix i, the matrix A size is 72*9, and the size of vectorial b is 72*1; Utilize least square method x=(A ^T* A) ^-1* A ^TB or svd can obtain needed transformation matrix.

Images