JPH0721827B2 - Image resampling processing method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a resampling processing method for an earth observation image and the like, and more particularly, to an inverse mapping- ¹ for a mapping from an original image to a corrected image without iterative calculation. The present invention relates to an image resampling processing method suitable for obtaining two-dimensional resampling processing with high accuracy and at high speed.

[Prior art]

It is known that images taken by artificial satellites contain fluctuations in shooting, non-linearity due to the characteristics of the sensors mounted on the satellite, or geometric distortion due to the rotation of the earth. Has been. In order to use this image, it is necessary to remove the above distortions. Therefore, various methods have conventionally been provided for removing the strain.

By the way, the method of removing the distortion, that is, the geometric distortion correction is performed by correcting the two-dimensional positional deviation between the reference image and the original image. Such distortion correction is realized using a distortion model.
Therefore, if this distortion model can be expressed with high accuracy, it becomes possible to rearrange each pixel of the original image at a predetermined position of the corrected image at equal intervals. However, even if each pixel (l, p) of the original image is mapped to the corrected image space (X, Y), this generally does not come to a predetermined image grid point.

Therefore, the image information (intensity value) is subjected to resampling processing (spatial interpolation processing).

In such image resampling processing, it is general to calculate the resampling image position for each pixel, calculate the weight corresponding to the position, and interpolate the pixel intensity. However, according to such a resampling processing method, an enormous amount of processing time is required when processing a large amount of image information such as a satellite image.

Therefore, as means for performing high-speed image resampling, (i) a method of dividing the entire image space into blocks of appropriate size and approximating the geometric distortion correction amount in each block by a bilinear model , (Ii) A method of performing processing by decomposing the two-dimensional resampling into two one-dimensional resamplings in the horizontal direction and the vertical direction is generally provided. By the way, the resampling is generally performed in the original image space, not in the corrected image space. For that purpose, it is necessary to obtain the inverse mapping ^-1 of the distortion model. However, since this cannot be obtained analytically, it has to be calculated numerically by performing a convergence calculation, resulting in the disadvantage that the processing time becomes long and the processing accuracy cannot be improved. It was

[Object of the Invention]

The object of the present invention is to eliminate the above-mentioned drawbacks of the prior art and correct the distortion of the Earth observation image, even when the inverse mapping ^-1 to the mapping from the original image to the corrected image is not analytically obtained. An object of the present invention is to provide an image resampling processing method capable of obtaining the inverse mapping ^-1 accurately and at high speed without iterative calculation and thereby performing two-dimensional resampling at high speed.

[Outline of Invention]

In order to achieve the above-mentioned object, the present invention is applied to an original image captured by an image capturing device mounted on a flying body such as an artificial satellite.
In an image processing method that performs a three-dimensional resampling process and creates a corrected image corrected to remove geometric distortion, the original image and the corrected image space are divided into multiple blocks, and the original image is corrected to a corrected image. The corrected image corresponding to each lattice point of each block in each block of the original image by mapping by the forward mapping model created by the motion model of the flying object and the shooting model in order to remove geometric distortion. The position of each grid point in the space is obtained, and an approximate model of the inverse mapping is created based on the position error between each grid point in block units of the original image and each corresponding grid point obtained, and the approximation is performed. The feature is that interpolation processing is performed by the model, and then distortion correction coefficient calculation processing for resampling is performed.

Example of Invention

The preferred embodiments of the present invention will be described below, but before that, the matters that form the basis of the present invention will be described.

FIG. 1 is a block diagram showing an image processing apparatus that executes image distortion processing. In FIG. 1, an image processing device 10 includes a processing device 11 such as an electronic computer, an external storage device 12 such as a magnetic drum for storing processing data and programs of the processing device 11, initial values and other necessary data. Input device 13 for inputting, a magnetic tape device 15 for reproducing the magnetic tape 14 on which original image information such as an observation image and attitude data is recorded and inputting it to the processing device 11, and image distortion processing by the processing device 11. And a magnetic tape device 16 for recording the corrected image on the magnetic tape 17.

With the image processing apparatus 10 configured as described above, the original image information from the magnetic tape 14 is converted into a corrected image and the magnetic tape 16 is used.
That is, the image distortion processing to be output to is executed. Hereinafter, the overall outline of the image distortion process will be described in more detail with reference to FIG.

FIG. 2 is roughly divided into three processes. (1) Input / radiometer distortion calculation (steps S3 to S6), (2) Geometric distortion calculation (step S7), (3) Distortion correction and corrected image output (steps S8 and S9)
It consists of and.

First, regarding the input / radiometer distortion calculation, in the input / radiometer distortion calculation, the trajectory is calculated in step S3 with the initial value input from the input device 3 in step S1, and the step S2 is calculated.
At step S4, the posture is calculated based on the image information and the posture data from the magnetic tape device 15, and at step S5
The observation image information is input at and the distortion correction function of the radiometer is calculated at step S6.

Next, regarding the geometric distortion calculation, a geometric distortion correction function is calculated in step S7 based on the value calculated by the trajectory calculation in step S3 and the posture calculation result calculated in step S4.

Furthermore, when describing the distortion correction and the output of the corrected image,
The observation image information entered in step S5 and step S6
The radiometer distortion correction function calculated in step S7 and the geometric correction function calculated in step S7 are imported, the resampling process is performed in step S8, the corrected image is output in step S9, and the corrected image is output in step S10. Record at 16.

The overall outline of the image distortion processing is as described above.

By the way, it has already been explained that the image captured by the artificial satellite contains geometric distortion. In order to remove this distortion, it has been explained that if a distortion model is used and the distortion model can be expressed with high accuracy, then the distortion of the image can be removed and the images can be rearranged at correct positions. This point will be further described with reference to FIGS. 3 and 4.

FIG. 3 is an explanatory diagram using a satellite model used for image correction. Further, FIG. 4 is a flow chart showing the outline of the mapping used for image correction, and showing the coordinate conversion processing from the observed image to the corrected image coordinate system.

In FIG. 3, when the point i on the surface 30 is observed by the satellite 31 at the time t _i and the pixel (l _i , p _i ) is obtained, the point calculated using the satellite model G is J It becomes the street.

J = G {x (t), θ (t), β (t)} where G is the satellite model x (t) is the position of the satellite in orbit θ (t) is the attitude angle β (t) of the satellite Is the scan angle when shooting with a satellite.

Further, y (t) is the true position of the point i, and the point K shows the position. If the satellite model G cannot be calculated accurately, an estimation error of the satellite model e will occur. The reference numeral 32 is the center of the earth.

FIG. 4 shows a flow for obtaining the corrected image in step S24 by using the mapping of the observed image in step S20 as described above. Observation image (Step S2
0) Satellite model G from Earth center rotation system (ECR: Ear
th Center Rotating) conversion processing (steps S21 and S22) is performed with mapping ₀ (step S23).
Mapping to the corrected image is performed using mapping ₂ (step S24).

Then, the mapping has the relation of = ₀ , ₁ and ₂ . By the way, ₁ and ₂ can be calculated accurately according to the shape model of the earth and the map projection method, regardless of the state of satellites and sensors. Was but connexion, satellite image-specific strain, mapping ₀
It is necessary to model this mapping ₀ as precisely as possible, as it is included in. By doing so, the mapping can be expressed with high accuracy, and the distortion can be removed.

However, even if each pixel (l, p) of the original image is mapped to the corrected image space (X, Y), this generally does not come to a predetermined pixel grid point. Therefore, it is necessary to resample the image information.

To resample an image, the resampling image position is calculated for each pixel, the weight corresponding to the position is calculated, and the pixel intensity is interpolated. When processing, the processing time is enormous.

Therefore, in order to perform the resampling processing at high speed, (i) a method of approximating the geometric distortion correction amount by a straight line model, and (ii) two-dimensional image resampling in the horizontal and vertical directions.
A method of decomposing into one one-dimensional resampling and processing is adopted. With respect to the above item (ii), since the one-dimensional resampling is performed in the vertical and horizontal directions, the space defining the distortion correction amount naturally differs in the horizontal direction and the vertical direction.

Resampling is generally performed in the original image space, not the corrected image space. For that purpose, it is necessary to obtain the inverse mapping ^-1 of the distortion model. However, since this cannot be obtained analytically, it is obtained numerically as follows.

As shown in FIG. 5 (I), the corrected image space is divided into two at equal intervals.
Divide into a dimensional grid. Each grid point is displayed in black. For each grid point P _i (X, Y), the corresponding point Q _i (l, p) on the original image is obtained. The grid points of this are shown in white.

For this, the convergence calculation is performed by repeatedly using the equation (2-1-1).

(X, Y) = (p, l) ... (2-1-1) From the inverse distortion amount at this lattice point, the inverse distortion amount at each pixel of the corrected image is calculated by the following (2-1-1) ) It approximates with the function shown in a formula.

(L, p) ＝ ^-1 (X, Y) ………… (2-1-2) Normally, the number of grids is set to be sufficiently large, and ^-1 of each grid section (block) is set to (X, Y). Expressed as a linear function.
Incidentally, the space between the sample points is interpolated as shown in FIG.

From the above, the distortion correction amount δX in the horizontal direction and the distortion correction amount δY in the vertical direction are defined as follows.

(I) First, the definition of the lateral distortion correction amount δX will be described.

The amount of image distortion correction is first performed in the order of horizontal distortion correction and then vertical distortion correction. Therefore, the distortion correction amount δX in the horizontal direction is defined between the original image and the intermediate image in which the distortion correction is performed only in the horizontal direction. Further, it is necessary to represent δX by a piecewise-straight line model. Therefore, the original image and the intermediate image are divided into blocks. Let (u, v) be the in-block relative position of the pixels of the original image, and (X, v) be the in-block relative position of the pixels of the intermediate image. The lateral distortion correction amount δX in the block is defined as δX = u−X = a ₀ + a ₁ X + a ₂ v + a ₃ Xv ... (2-1-3). Let a ₀ , a ₁ , a ₂ , and a ₃ be called distortion correction coefficients in the horizontal direction.

(II) Next, the definition of the vertical distortion correction amount δY will be described.

The vertical distortion correction amount δY is defined between the intermediate image in which only the horizontal direction is distortion-corrected and the output image. Similarly, the intermediate image and the output image are divided into blocks.

(X, Y) is the relative position (X, Y) in the block of the pixel of the intermediate image
Is the relative position within the block of the pixels of the output image. The vertical distortion correction amount δY in the block is defined as δY = V−Y = b ₀ + b ₁ X + b ₂ Y + b ₃ XY ... (2-1-4). Let b ₀ , b ₁ , b ₂ and b ₃ be called distortion correction coefficients in the vertical direction.

(III) Furthermore, the definition of the regular lattice will be described.

The geometrical distortion correction amount is obtained by dividing one image into N × N section blocks and calculating the distortion in the section blocks by a straight line model. It is necessary to define a regular lattice to form this N × N partition block.

First, the number of lines and the number of pixels included in the division block are defined as follows.

(A) the number of lines in one block of the original image is b _l , (b) the number of pixels in one block of the original image is b _p , (c) the number of lines in one block of the output image is B _l , (d ) The number of pixels in one block of the output image is B _p , the image size of the original image and the output image are (e) the horizontal pixel size of the original image is d _p , (f) the vertical pixel of the original image The size is d _l , (g) the horizontal pixel size of the output image is D _p , and (h) the vertical pixel size of the output image is D _l . The size of the corrected image that defines the distortion correction coefficient is
(B _l + N + 1) × (B _p × N + 1), and the size of the corresponding original image is (b _l × N + 1) × (b _p × N + 1)
Is.

The coordinates of grid points in each coordinate system can be calculated as follows.

(B) Normal grid point coordinate values in the output space, Here, 1iN + 1,1jN + 1, X _s (1) = X _s ^c − (B _p × D _p × N) / 2 Y _s (1) = Y _s ^c − (B _p × D _p × N) / 2 is there.

(B) Normal grid point coordinate values on the corrected image Similarly, 1iN + 1,1jN + 1 (c) Pseudo-resor point coordinate values on the original image A grid having the same size as the normal grid stretched on the corrected image is stretched on the original image to be called a pseudo-resolve. The pseudo resolution coordinate value is Here, 1iN + 1,1jN + 1 u (1) = u _c − (N × b _p ) / 2 v (1) = v _c − (N × b _l ) / 2. u _c, v _c is the framing center point in the original image, X _s ^c, Y _s ^c is the framing center point in the output space. That is, (X _s ^c , Y _s ^c ) = (u _c , v _c ) ... (2-1-8).

(IV) Furthermore, the lateral distortion correction coefficient (a ₀ , a ₁ , a ₂ , a ₃ )
Calculation of will be described with reference to FIG.

In FIG. 7, first, in step S30, a regular lattice is defined. This is the processing of the above (III). Then, steps S31 to S35 are performed. That is, the coefficient a ₀ , a of the straight line model for obtaining the lateral distortion correction amount δX on the block (i, j)
Calculation of ₁ , a ₂ and a ₃ is performed by the following procedure.

(I) The distortion correction amount δX in the horizontal direction has been described above, but is defined between the original image and the intermediate image that has been subjected to the distortion correction only in the horizontal direction.

As shown in FIG. 6 (I), in the block (i, j), the line values v (j), v (j + 1) are fixed and correspond to the grid points A, B, C, D on the intermediate image. Grid point A ′ of the original image
Find B ', C', D '.

The following relation holds between A, B, C, D and A ', B', C ', D'.

Here, ▲ ^-1 _p ▼ is a reverse mapping in the lateral direction, but it is generally not analytically obtained.

(Ii) Calculation of ▲ ^-1 _p ▼, that is, (X (i), v
The calculation of the point of (u ′ _j (i), v (j)) corresponding to (j)) is performed by the following convergence calculation. First, the step
Initial setting processing is performed in S31.

Next, the processing is performed as follows.

(A) Map the original image points to the output space as follows (step 32).

^{_{(X s n (i),}} Y s n (j)) = (u j n (i), v (j)) ... (2-1-10) (b) the following formula to calculate the distance between the lattice points (Step S33).

ΔX ⁿ = X _s (i) -X _s ⁿ (i) (2-1-11) (c) At step S34, the following formula is determined.

When │ΔX ⁿ │ε, the following equation is stored as " ^sum ", and the process proceeds to step S36.

u _j ′ (i) = u _j ⁿ (i) ……… (2-1-12) Also, when | ΔX ⁿ |> ε, it is determined as “no” and step S3
Move to 5 and update the position of the grid point as shown in the following formula.

u _j ^{n + 1} (i) = u _j ⁿ (i) + ΔX ⁿ / d _p ... (2-1-13) Then, the process from (a) is repeated. That is, the process is repeated from step S32.

Here, as the initial value of the convergence calculation, u _j ⁰ (i) = u (j), that is, the value of the pseudo resolution point is used.

(Iii) Further, in step S36, the lateral distortion correction amount at each grid point is calculated by the following equation (2-1-14).

(Iv) Similarly, in step S36, the geometric distortion correction coefficient (a ₀ , a ₁ , a ₂ , a ₃ ) is calculated using the value obtained in (iii) above.

When (X, v) is the relative position in the block of the point to be obtained, δX is calculated by using δX = u−X = a ₀ + a ₁ X + a ₂ v + a ₃ Xv) It can be represented by (2-1-15). At this time, the distortion correction coefficient is Is. When the calculation is made in step S36 in this way, the process proceeds to step S37.

The vertical distortion correction coefficients (b ₀ , b ₁ , b ₂ , b ₃ ) are calculated in steps S37 to S42.

The calculation method of the coefficients b ₀ , b ₁ , b ₂ , b ₃ of the straight line model of the distortion correction amount δY in the vertical direction of the block (i, j) shown in FIG. 6 (II) is the same as the calculation method in the horizontal direction. Is. First, in step S37, initial settings are made. This performs the following processing.

(I) The vertical distortion correction amount δY is defined between the intermediate image and the corrected image. The pixel values X (j), X (j + 1) are fixed at the block (i, j), and the points E ', F'on the intermediate image corresponding to the regular grid points E, F, G, H on the corrected image are fixed. , G ′,
Find H ′.

The following relationship is established between E, F, G, H and E ', F', G ', H'.

Here, ^{−1 −1} _l is an inverse mapping in the vertical direction, but as in the case of the horizontal direction, the convergence calculation is performed in steps S38 to S41.

(Ii) E ′ (X corresponding to E (X (i), Y (j))
The points (i), v ′ _i (j)) are calculated by the following convergence calculation.

(A) an intermediate image _{(X (i), v i} n (j)) from the original image coordinate values
(u ⁿ (i), v _i ⁿ (j)) was calculated. This is because the mapping is defined between the output space from the input ^{image, u n (i) = X} (i) + a 0 + a 3 × (V i n (j) -V (j ')) ...
(2-1-18) j ': the corresponding block (b) input image points u _i ⁿ (j) using the following equation maps the output space (step S38).

(X _s (i), Y _s ⁿ (j)) ＝ (u ⁿ (i), v _i ⁿ (j)) (c) The distance from the regular grid point is calculated using the following equation (2-1-19). Calculate with (step S39).

ΔY ⁿ = Y _s (j) -Y _s ⁿ (j) ……… (2-1-19) (d) Next, if | ΔY ⁿ | ε at step S40, it is regarded as “ ^sum ” and the following (2-1) -20) is memorized and it moves to step S42.

_{v 'i (j) = v} i n (j) ......... (2-1-20) , however, | ΔY ⁿ |> ε When the proceeds to step S41 as "not", the following (2-1-21 ) Equation is used to update the grid point position.

^{_{v i n + 1 (j)}} = v i n (j) -ΔY n / d l ......... (2-1-21) is then repeated from the (a). That is, step S3
Then, the process goes to step S38 and thereafter.

Here, the initial value of the convergence calculation is v _i ⁰ (j) = v (j) (set in step S37).

(Iii) Then, in step S42, first, the distortion correction amount in the vertical direction at the regular grid point is calculated by the following equation (2-1-22).

(Iv) Similarly, in step S42, the geometric distortion correction coefficient (b ₀ , b ₁ , b ₂ , b ₃ ) is calculated using the following equation (2-1-23).

When (X, Y) is the relative position in the block of the point to be obtained, δY is calculated by using X and Y. δY = v−Y = b ₀ + b ₁ X + b ₂ Y + b ₃ XY) It is expressed as (2-1-23).

At this time, the distortion correction coefficient is Is.

Although it was obtained in this way, since the above processing involves convergence calculation, it took a long processing time and the accuracy was poor.

The present invention solves the above problems.

FIG. 8 is a flow chart showing an embodiment of the image resampling processing method according to the present invention. The following definitions used in the embodiments of the present invention may be the same as those in the processing method described above.

(A) Definition of the lateral distortion correction amount δX, (b) Definition of the vertical distortion correction amount δY, (c) Definition of the regular lattice, that is, the processing method described in the above-mentioned basic matters. The difference is that, as shown in FIG.
The point is that S50 to step S53 are provided to prevent the above convergence calculation. Therefore, the same steps as those in the processing method shown in FIG. 7 are designated by the same step numbers and the description thereof will be omitted.

The step S50 of FIG. 8 will be described below.

Using the mapping from the original image to the corrected image, the corrected grid points A ′, B ′, C ′, corresponding to the respective grid points A, B, C, D of the original image,
Find D ′.

The following relationship holds between A, B, C, D and A ', B', C ', D'.

Here, interpolation processing is performed. First, when the horizontal and vertical distortion models are represented by a straight line model, δu = X−u = a ₀ ′ + a ₁ ′ x
+ A ₂ ′ y + a ₃ ′ xy (4-1-2) δv = Y−v = b ₀ ′ + b ₁ ′ x + b ₂ ′ y + b ₃ ′ xy (4
-1-3).

Next, when each distortion correction coefficient is calculated, Becomes That is, in step S50, the processing is performed as described above, and the process proceeds to step S51.

Next, the inverse mapping approximation model of step S51 is calculated as follows.

Using the equations (4-1-2) and (4-1-3), (X,
(X, y) corresponding to Y) can be obtained.

^{(X, y) = -1 (} X, Y) ......... (4-1-6) (4-1-3) from the _{equation, y = (Y-b 0} '-b 1' x) / (b _{2 '+ b 3' x +} 1) ......
(4-1-7) Substituting this into the equation (4-1-2), (a ₁ ′ b ₃ ′ −a ₃ ′ b ₁ ′ + b ₃ ′) x ² + (a ₀ ′ b ₃ ′) + A ₁ ′ b ₂ ′ + a ₁ ′ −a ₂ ′ b ₁ ′ + a ₃ ′ Y−
_{a 3 'b 0' -b 3} 'X + b 2' +1) x + (a 0 'b 2' + a 0 '+ a 2' Y-a 2 'b 0' -b 2 'X-X) = 0 ...... … (4-1-18) If you solve this, Where A ₀ = a ₀ ′ b ₂ ′ + a ₀ ′ + a ₂ ′ Y−a ₂ ′ b ₀ ′ X−X A ₁ = a ₀ ′ b ₃ ′ + a ₁ ′ b ₂ ′ + a ₁ ′ −a ₂ ′ B ₁ ′ + a ₃ ′ Y−
a ₃ ′ b ₀ ′ −b ₃ ′ X + b ₂ ′ +1 A ₂ ＝ a ₁ ′ b ₃ ′ −a ₃ ′ b ₁ ′ + b ₃ ′ x has two roots, but a root satisfying 0xb _p is Choose.

Y is obtained from the equations (4-1-19) and (4-1-7).

Where x and y are And for convenience. The above-described processing is, in short, the processing of solving the forward mapping approximation model and approximating the inverse mapping- ¹ . At step 51, it can be calculated as described above.

Once calculated as described above, the geometric distortion correction coefficient for resampling is calculated in steps S52 and S53 as follows.

(I) In the block (i, j) of the corrected image, the points E ′, F ′, G ′, H ′ on the original image corresponding to the regular grid points E, F, G, H on the corrected image are Ask.

The following relationship is established between E, F, G, H and E ', F', G ', H'.

Here, since ▲ ^-1 _x ▼ and ▲ ^-1 _y ▼ are defined for each block on the original image, E ', F', G ',
If H ′ belong to different blocks, ▲
^-1 _x ▼ and ▲ ^-1 _y ▼ are different.

(Ii) Distortion correction amount at regular grid points (Iii) Calculation of geometric distortion correction coefficient When (X, Y) is the relative position in the block of the point to be obtained, δX and δY are calculated using X and Y, Represents.

At this time, the distortion correction coefficient is And then the vertical direction, Is obtained as.

Both the image resampling processing method of FIG. 7 and the image resampling processing method of the present invention are processed by the image processing apparatus shown in FIG.

The above processing will be summarized again and described with reference to FIG.

First, in step S30, a regular grid is defined in the original image and the output space. That is, the original image and the corrected image space are divided into respective blocks.

Next, in steps S31 and S32, the following processing is performed for each grid point.

The grid point coordinates (u, v) are initialized to the corresponding grid point coordinates on the original image (step S31).

A point (X,
Y) is obtained (step S32).

That is, as shown in the following equation, the analytically given original image is mapped to the corrected image.

((X, Y) = (u, v)) Further, in steps S50 and S51, the following processing is performed for each block.

The above-mentioned analytically given forward mapping in the corresponding block is approximated by using a straight line model, that is, by performing interpolation processing (step S50).

The forward mapping model in the block is solved to obtain the inverse mapping approximation model ▲ ^-1 _x ▼, ▲ ^-1 _y ▼ (step S5
1).

Furthermore, in steps S52 and S53, the following processing is performed for each block in the output space.

The lateral distortion correction coefficient (a ₀ , a ₁ , a ₂ , a ₃ ) is obtained. (Step S52).

Calculate the vertical distortion correction coefficient (b ₀ , b ₁ , b ₂ , b ₃ ) (step S
53).

By doing so, it is not necessary to perform the convergence calculation as in the processing method shown in FIG. 7, and the processing time is improved.

Next, the result of comparison between the processing method (FIG. 7) for performing the above convergence calculation and this embodiment will be described. Therefore, it is defined as follows.

Processing time of forward mapping …………… T ₁ number of blocks …………… N number of lattices ……… (N + 1) * (N + 1) = (N + 1) ² Average number of convergence in convergence calculation …… ... M In FIGS. 7 and 8, it will be described in advance that the other processing times are negligibly shorter than the processing time T ₁ of the forward mapping.

Therefore, the processing time T ₀ in the processing method shown in FIG.
Is T ₀ = (T ₁ * N ² ) * N * 2, and the processing time T _N in this embodiment is T _N = (T ₁ * N ² ), so T _N / T ₀ = 1 / (2 * N ).

In the specific example, For, T ₀ = a 60 [sec] T _N = 10 (sec), the processing time is shortened to _{T N / T 0 = 1/} 6.

The number of blocks is related to the geometric distortion correction accuracy,
The relationships shown in Fig. 10 and Fig. 10 are obtained. Incidentally, FIG. 9 is a special line diagram showing a horizontal straight line modeling error.
FIG. 10 is a characteristic diagram showing a vertical straight line modeling error. In these figures, the horizontal axis represents the block number, and the vertical axis represents the pixel number and the line number.

In the processing method shown in FIG. 7, when the number of blocks N = 4, the processing time was 15 [sec] and the accuracy was 0.3 [pixel] (horizontal direction), but in the present embodiment, the number of blocks N = It was possible to increase the number to 12, and the processing time was 14.4 (sec) and the accuracy was 0.05 [pixel].

From now on, the image area tends to increase more and more, and the distortion characteristic itself tends to be complicated.

Therefore, it is necessary to increase the number N of blocks, but with this, the processing time increases on the order of squares. Therefore, it is effective that the processing time can be shortened as in the present invention.

〔The invention's effect〕

As described above, according to the present invention, there is an effect that the geometric distortion correction accuracy is improved and the processing time can be shortened.

[Brief description of drawings]

FIG. 1 is a block diagram showing an image processing apparatus for executing image distortion correction processing, FIG. 2 is a flow chart showing an outline of image distortion processing executed by the image processing apparatus shown in FIG. 1, and FIG. 3 is image correction. FIG. 4 is an explanatory view showing a satellite model used in FIG.
FIGS. 5 (I) and (II) are explanatory diagrams shown for explaining the resampling method, FIGS. 6 (I) and (II) are relationship explanatory diagrams showing a geometric distortion relation in the block, and FIG. FIG. 8 is a flow chart showing a resampling processing method which is the basis of the present invention, FIG. 8 is a flow chart showing an embodiment of an image resampling processing method according to the present invention, and FIG. 9 is a horizontal straight line model. FIG. 10 is a characteristic diagram showing the conversion error, and FIG. 10 is a characteristic diagram showing the vertical straight line modeling error. 10 ... Image processing device, 11 ... Processing device, 12 ... External storage device, 15 and 17 ... Magnetic tape device, S50 ... Forward mapping approximation processing step, S52 ... Inverse mapping approximation model processing step.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Hideo Ota 5-2-1 Omika-cho, Hitachi-shi, Ibaraki, Hitachi, Ltd. Inside the Omika Plant, Hitachi, Ltd. (56) Reference JP-A-55-59573 (JP, A) JP-A-57-196366 (JP, A) JP-A-54-70073 (JP, A) JP-A-53-115262 (JP, A) Practical application Sho-57-84058 (JP, U)

Claims (1)

[Claims] 1. An image in which a corrected image is created by performing a two-dimensional resampling process on an original image taken by an imaging device mounted on an aircraft such as an artificial satellite to correct a geometric distortion. In the processing method, the original image and the corrected image space are divided into multiple blocks,
Each grid point of the block is mapped to each block of the original image by mapping by the forward mapping model created by the motion model and the shooting model of the flying object in order to remove the geometric distortion from the original image to the corrected image. The position of each grid point in the corrected image space corresponding to the above is calculated, and the inverse mapping is approximated based on the position error between each grid point in block units of the original image and each calculated grid point corresponding thereto. An image resampling processing method characterized in that a model is created, interpolation processing is performed using the approximate model, and then distortion correction coefficient calculation processing for resampling is performed.