CN110956588B - Image high-precision geometric correction method based on shortest distance of encrypted points - Google Patents

Abstract

The invention provides an image high-precision geometric correction method based on the shortest distance of an encryption point, which comprises the steps of firstly determining object coordinates according to a coordinate forward calculation function, then determining the object coordinates of each pixel on a corrected image according to the object coordinate range and a resampling interval, extracting the maximum value of the number of pixels on the corrected image corresponding to 1 pixel on an original image, and recording the maximum value as N; setting the geometric correction control precision as e pixels, encrypting each pixel of an original image into a point of [ N/e ] x [ N/e ], calculating object coordinates of an encrypted point by using a coordinate forward calculation function, calculating image coordinates of the encrypted point on the corrected image one by one, and calculating the nearest integer pixel position; for each whole pixel position on the corrected image, if a plurality of corresponding encrypted point coordinates are found, determining a corresponding relation by comparing distance errors; and (4) interpolating gray scale from the original image according to the corresponding encrypted point coordinates of the position of each point of the corrected image to obtain a complete output image.

Description

Image high-precision geometric correction method based on the shortest distance of encrypted points

technical field

The invention relates to the technical field of image geometric correction, in particular to the direct method geometric correction, and relates to an image high-precision geometric correction method based on the shortest distance of encrypted points.

Background technique

Optical sensors using linear array pendulum-scan imaging can be applied to the infrared band, with a large imaging field of view and a wide range of image coverage. For example, a marine satellite aqua scanner can cover the world once a day, and has great advantages in dynamic monitoring. Advantage. However, its geometric imaging model is more complicated than linear array push-broom imaging, mainly because: during linear array push-broom, each row of pixels is imaged at the same time, so it has the same orbit, attitude and other parameters; Pixels are imaged at different times and have different orbital pose parameters. This leads to different subsequent geometric processing methods: when linear array push-broom is used, efficient indirect method geometric correction can be achieved by establishing a rational polynomial model equivalent to the collinear equation model; while linear array swing sweep can only be achieved by the general direct method Or the direct method and the indirect method of local approximation are combined for geometric correction. Whether it is corrected by the general direct method or the combination of the direct method and the indirect method of local approximation, there is a problem of insufficient precision.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned problems of the prior art, the present invention proposes a high-precision geometric correction method for images based on the shortest distance of encrypted points.

The technical scheme of the present invention provides a kind of image high-precision geometric correction method based on the shortest distance of the encrypted point, comprising the following steps:

1) First, determine the object coordinates according to the coordinate positive calculation function, and then determine the object coordinates of each pixel on the corrected image according to the object coordinate range and resampling interval, and extract 1 pixel on the original image corresponding to the corrected image. The maximum value of the number of pixels above is recorded as N;

2) Set the geometric correction control accuracy to e pixels, encrypt each pixel of the original image into a point of [N/e]×[N/e], and use the coordinate positive calculation function to obtain the encrypted point (sx, sy). Object coordinates (lon,lat), where [] means rounded up;

Calculate the image coordinates (X, Y) of the encrypted points on the corrected image one by one, and find the nearest integer pixel position (X0, Y0);

For each integer pixel position (X0, Y0) on the corrected image, if it is found that there are multiple corresponding encrypted point coordinates, the corresponding relationship is determined by comparing the distance error;

3) For each point of the corrected image, according to the coordinates of the corresponding encrypted point at the position (X0, Y0), the grayscale is interpolated from the original image to obtain a complete output image.

Moreover, the coordinate positive calculation function is as follows,

It is known that the image coordinate of a certain point P is (s, l). According to the camera pointing angle model, the three-dimensional vector of point P in the camera coordinate system is obtained as v _cam =[x _C (s) y(s) 1] ′, where a ₀ , a ₁ , a ₂ , b ₀ , b ₁ , b ₂ are the camera pointing angle model coefficients,

According to the principle of linear sweep imaging, the imaging time of the P point is obtained as T=Tl ₀ +Δt×s, where Tl ₀ is the imaging time of the pixel in the 1st row and the 0th column, and Δt is the exposure time interval of each column of pixels , obtain the satellite's position [X _T Y _T Z _T ]', velocity and attitude according to T interpolation, and obtain the transformation matrix R T from the J2000 coordinate system to the WGS84 coordinate system and from the body coordinate system to the J2000 coordinate system at the time _T , _RGFFB ;

Assuming that R _BS is the transformation matrix from the camera coordinate system to the body coordinate system, the coordinate positive calculation function from the image coordinate (s, l) to the object coordinate vector v _obj = [XYZ]' is as follows,

The object coordinates (X, Y, Z) also satisfy the WGS84 ellipsoid equation,

Among them, A and B are the major and minor semi-axes (known constants) of the WGS84 ellipsoid, and H is the elevation of the object coordinate (X, Y, Z) on the WGS84 ellipsoid.

Moreover, the implementation of computing N is as follows,

求出所有d的最大值，并向上取整得到N。For two adjacent integer points on the original image, the corresponding object coordinates are (Xs, Ys), (Xt, Yt), respectively. If the resampling interval of the corrected image is GSD, then the two points in the corrected image The pixel distance on is

Find the maximum value of all d and round up to get N.

Moreover, according to the object coordinates (lon, lat) of the encrypted point (sx, sy), the image coordinates (TX, TY) of the encrypted point on the corrected image are calculated as follows,

TX=(lon-Xmin)/GSD, TY=(Ymax-lat)/GSD

Among them, (Xmin, Ymax) is the object coordinate of the upper left corner of the output corrected image.

Furthermore, steps 1) to 3) are executed in parallel in an image block manner.

Compared with the prior art, the method proposed in the present invention has controllable geometric correction accuracy, supports rapid processing through parallel computing, and can also be widely applied to coordinate inverse calculation problems of any known coordinate forward calculation function, such as aviation and aerospace. The linear array push-broom imaging method is also applicable, the aviation linear array swing scanning method is also applicable, and the area array imaging method is also applicable.

Description of drawings

FIG. 1 is a schematic flowchart of an embodiment of the present invention.

Detailed ways

In order to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

Referring to Fig. 1, an embodiment of the present invention proposes a high-precision geometric correction method for images based on the shortest distance of encrypted points, comprising the following steps:

1) First determine the object coordinate range of the original image according to the coordinate positive calculation function, and then according to the object coordinate range and resampling interval (which is a preset value, usually the ground resolution corresponding to the center point of the original image), determine the corrected image. The object coordinate of each pixel on the image, and the calculated 1 pixel on the original image corresponds to the maximum number of pixels on the corrected image and is recorded as N (rounded up).

The coordinate positive calculation function is as follows:

It is known that the image coordinate of a certain point P is (s, l). According to the camera pointing angle model, the three-dimensional vector of point P in the camera coordinate system is obtained as v _cam =[x _C (s) y(s) 1] ', where a ₀ , a ₁ , a ₂ , b ₀ , b ₁ , b ₂ are camera pointing angle model coefficients (known).

According to the principle of linear sweep imaging, the imaging time of the P point is obtained as T=Tl ₀ +Δt×s, where Tl ₀ is the imaging time of the pixel in the 1st row and the 0th column, and Δt is the exposure time interval of each column of pixels . According to T, the satellite's position [X _T Y _T Z _r ]', speed, attitude and other parameters can be interpolated, so as to obtain the transformation matrix from the J2000 coordinate system to the WGS84 coordinate system and from the body coordinate system to the J2000 coordinate system at the time of T R _T , R _GFFB .

Let R _BS be the transformation matrix (known value) from the camera coordinate system to the body coordinate system, then the coordinate positive calculation function from the image coordinate (s, l) to the object coordinate vector v _obj = [XYZ]' is as follows :

Because the object coordinates (X, Y, Z) also satisfy the WGS84 ellipsoid equation:

Where A and B are the major and minor semi-axes of the WGS84 ellipsoid (known constants), and H is the elevation of the object coordinate (X, Y, Z) on the WGS84 ellipsoid.

According to the above coordinate positive calculation function, the process of solving the object coordinate (X, Y, Z) from the image coordinate (s, l) is as follows:

a) First assume that the elevation H = 0, let

The three unknowns (XYZ) in Equation 2 are expressed as linear expressions about the parameter m to be solved:

Substitute (Formula 4) into (Formula 3), solve the quadratic equation of one variable about m, (remove the larger value) to obtain m. Then substitute m into (Formula 2) to solve the object coordinate (X, Y, Z).

b) According to the known DEM, interpolate from the object coordinates (X, Y, Z) to obtain a more accurate elevation H, repeat the above process 1), and solve the more accurate object coordinates (X, Y, Z) ).

c) Stop iteration when the difference between two adjacent calculation results of object coordinates (X, Y, Z) is less than a preset threshold. During specific implementation, the threshold value can be preset according to the need of precision.

The object coordinate range of the original image is calculated as follows:

Knowing the size of the original image, set the original image width w and height h, the four corner coordinates (0, 0), (w-1, 0), (0, h-1), (w-1, h- 1) Substitute into the coordinate positive transformation function respectively, and obtain the object coordinates (X1, Y1), (X2, Y2), (X3, Y3), (X4, Y4) of the corresponding four corner points, and find the four corners The maximum and minimum coordinate ranges (Xmin～Xmax, Ymin～Ymax) of the point are the object coordinate range of the original image.

The implementation of computing N is as follows:

按此对于原始影像上所有相邻整数点分别求出 像素距离d后，求出所有d的最大值，并向上取整得到N。For two adjacent integer points on the original image, let the pixel coordinates be (sx, sy), (sx+1, sy), or (sx, sy), (sx, sy+1), (where 0≤sx <w-1, 0≤sy<h-1), the corresponding object coordinates are (Xs, Ys), (Xt, Yt), respectively. If the resampling interval of the corrected image is GSD, then these two points The pixel distance on the corrected image is

According to this, after the pixel distance d is calculated for all adjacent integer points on the original image, the maximum value of all d is calculated, and N is obtained by rounding up.

基于最短距离原则，确定各整像素位置(TX0，TY0)相应的加密点原始影像坐标，即对于纠正后图像上某整像素位置(X0，Y0)，如果若发现有多个对应加密点坐标，则通过比较距离误差确定对应关系，最终得到原始坐标查找表和相应误差表。2) Set the preset geometric correction control accuracy to e pixels (where 0<e<1, such as e=0.2), then encrypt each pixel of the original image into [N/e]×[N/e] (where [ ] means rounding up), use the coordinate positive calculation function to obtain the object coordinates (lon, lat) of each encrypted point (sx, sy) on the original image; The image coordinates (TX, TY) on the image, and find the nearest integer pixel position (TX0, TY0), and calculate the distance error

Based on the shortest distance principle, determine the original image coordinates of the encrypted point corresponding to each integer pixel position (TX0, TY0), that is, for a certain integer pixel position (X0, Y0) on the corrected image, if there are multiple corresponding encrypted point coordinates, Then, the corresponding relationship is determined by comparing the distance errors, and finally the original coordinate look-up table and the corresponding error table are obtained.

The implementation of the encryption point operation of the embodiment is as follows:

Interpolate the original image, assuming that there are four integer pixels (sx, sy), (sx+1, sy), (sx, sy+ 1), (sx+1, sy+1) on the original image, using 5 After ×5 encryption, there are a total of 25 encrypted points, and the image coordinates of the encrypted points are:

(sx, sy), (sx+0.25, sy), (sx+0.5, sy), (sx+0.75, sy), (sx+1, sy), (sx, sy+0.25), (sx+0.25 , sy+0.25), (sx+0.5, sy+0.25), (sx+0.75, sy+0.25), (sx+1, sy+0.25), (sx, sy+0.5), (sx+0.25, sy +0.5), (sx+0.5, sy+0.5), (sx+0.75, sy+0.5), (sx+1, sy+0.5) (sx, sy+0.75), (sx+0.25, sy+0.75) , (sx+0.5, sy+0.75), (sx+0.75, sy+0.75), (sx+1, sy+0.75), (sx, sy+1), (sx+0.25, sy+1), ( sx+0.5, sy+1), (sx+0.75, sy+1), (sx+1, sy+1)

The two-dimensional object coordinate (lon, lat) corresponding to the encrypted point is still calculated using the previous positive coordinate transformation function.

Calculate the image coordinates (TX, TY) of the encrypted point on the corrected image as follows:

TX=(lon-Xmin)/GSD, TY=(Ymax-lat)/GSD

in,

(Xmin, Ymax) is the object coordinate of the upper left corner of the output corrected image, that is, the minimum value of lon and the maximum value of lat;

(lon, lat) is the object coordinate of the encrypted point on the output corrected image, generally the east of the horizontal axis is positive, and the north of the vertical axis is positive;

(TX, TY) are the image coordinates of the encrypted point on the output corrected image. Generally, the horizontal axis is positive to the right, the vertical axis is positive downward, and the upper left corner is the origin of the image coordinate system.

Find the nearest integer pixel position (TX0, TY0) of the encrypted point on the corrected image:

TX0=[TX], TY0=[TY], where [] means to take the nearest integer

By looping through the above method, for each encrypted point (sx, sy) on the original image, the nearest integer pixel position (TX0, TY0) and the distance error err on the corrected image can be obtained.

Conversely, at the integer pixel position (TX0, TY0) on the corrected image, the original image coordinates (sx, sy) of the encrypted points are recorded one by one, and the distance error err is recorded at the same time.

Since the distance between the encryption points is very small, the nearest integer pixel position on the corrected image calculated by the adjacent encryption points may be the same coordinate value (TX0, TY0). At this time, it is necessary to compare the distance error err:

In specific implementation, space Q can be pre-allocated to the corrected image, and a specific value, such as -9999, -999999, etc., which are impossible to appear, can be initialized, indicating that the space has not been assigned. Each encrypted point calculates the corresponding integer pixel position (TX0, TY0) and distance error err on the corresponding corrected image in parallel, and assigns (sx, sy) to the integer pixel position (TX0, TY0) of the pre-allocated space Q. When assigning value, if it is found that the corresponding encrypted point image coordinates (sx, sy) obtained before are already at the integer pixel position (TX0, TY0), then compare the distance error err, and use the encrypted point with the smaller distance error based on the shortest distance principle. The image coordinates are used to replace the image coordinates of the encrypted point with a large distance error; when the distance error is exactly the same, either one can be chosen, and the average value of the coordinates of the two encrypted points can also be recorded during the specific implementation. Finally, the original coordinate lookup table and error table are obtained.

3) For each point of the corrected image, according to its position (TX0, TY0), and according to the corresponding encrypted point image coordinates (sx, sy) determined in 2), the grayscale is interpolated from the original image to obtain a complete output. The image, that is, the corrected image, the geometric positioning accuracy of the image is reflected from the error table.

During specific implementation, commonly used grayscale resampling methods, such as the nearest neighbor method, bilinear interpolation method, and bicubic convolution method, can be used.

During specific implementation, the automatic operation of the process can be realized by means of software. Because it is from the image side to the object side, it can be realized by dividing the original image into blocks and processing each block of images in parallel. For example, after the image is divided into blocks, steps 1) to 3) are performed in parallel for each block. Devices that run the process should also fall within the scope of the present invention.

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention pertains can make various modifications or additions to the described specific embodiments or substitute in similar manners.

Claims (4)

1. an image high-precision geometric correction method based on the shortest distance of the encrypted point, is characterized in that, comprises the following steps: 1) First, determine the object coordinates according to the coordinate positive calculation function, and then determine the object coordinates of each pixel on the corrected image according to the object coordinate range and resampling interval, and extract 1 pixel on the original image corresponding to the corrected image. The maximum value of the number of pixels above is recorded as N; the implementation of calculating N is as follows, For two adjacent integer points on the original image, the corresponding object coordinates are (Xs, Ys), (Xt, Yt) respectively, and if the resampling interval of the corrected image is GSD, then the two points in the corrected image The pixel distance on is

Find the maximum value of all d and round up to get N; 2) Set the geometric correction control accuracy to e pixels, encrypt each pixel of the original image into a point of [N/e]×[N/e], and use the coordinate positive calculation function to obtain the encrypted point (sx, sy). Object coordinates (lon,lat), where [] means rounded up; Calculate the image coordinates (X, Y) of the encrypted points on the corrected image one by one, and find the nearest integer pixel position (X0, Y0); For each integer pixel position (X0, Y0) on the corrected image, if it is found that there are multiple corresponding encrypted point coordinates, the corresponding relationship is determined by comparing the distance error; 3) For each point of the corrected image, according to the coordinates of the corresponding encrypted point at the position (X0, Y0), the grayscale is interpolated from the original image to obtain a complete output image. 2. The image high-precision geometric correction method based on the shortest distance of the encrypted point according to claim 1, is characterized in that: described coordinate positive calculation function is as follows, It is known that the image coordinate of a certain point P is (s, l). According to the camera pointing angle model, the three-dimensional vector of point P in the camera coordinate system is obtained as v _cam =[x _C (s) y _C (s) 1 ]′, where a ₀ , a ₁ , a ₂ , b ₀ , b ₁ , b ₂ are the camera pointing angle model coefficients,

According to the principle of linear sweep imaging, the imaging time of the P point is obtained as T=Tl ₀ +Δt×s, where Tl ₀ is the imaging time of the pixel in the 1st row and the 0th column, and Δt is the exposure time interval of each column of pixels , obtain the satellite's position [X _T Y _T Z _T ]', velocity and attitude by interpolation according to T, and obtain the transformation matrix R T from the J2000 coordinate system to the WGS84 coordinate system and from the body coordinate system to the J2000 coordinate system at the time _T , _RGFFB ; Assuming that R _BS is the transformation matrix from the camera coordinate system to the body coordinate system, the coordinate positive calculation function from the image coordinate (s, l) to the object coordinate vector v _obj = [XYZ]' is as follows,

The object coordinates (X, Y, Z) also satisfy the WGS84 ellipsoid equation,

Among them, A and B are the major and minor semi-axes of the WGS84 ellipsoid, and H is the elevation of the object coordinate (X, Y, Z) on the WGS84 ellipsoid. 3. the image high-precision geometric correction method based on the shortest distance of the encrypted point according to claim 2, is characterized in that: according to the object coordinate (lon, lat) of the encrypted point (sx, sy), calculate the encrypted point after correcting The image coordinates (TX, TY) on the image are as follows, TX=(lon-Xmin)/GSD, TY=(Ymax-lat)/GSD Among them, (Xmin, Ymax) is the object coordinate of the upper left corner of the output corrected image. 4. The high-precision geometric correction method for images based on the shortest distance of encrypted points according to claim 1, 2 or 3, wherein steps 1) to 3) are performed in parallel in an image block manner.

Images