1 Geometric Distortion
In the process of remote sensing imaging, the image pixels generated by the sensor are squeezed, stretched, distorted and offset relative to the actual position of the ground target.
The reasons for geometric distortion are: 1) internal factors of the sensor; 2) remote sensing platform factors; 3) earth factors.
Because of the existence of geometric distortion, we propose a coping strategy of geometric correction.
2 Geometric Correction
2.1 The necessity of geometric correction
Q: Why is geometric correction necessary for the existence of geometric distortion?
This is because geometric distortion will bring errors to the quantitative analysis, change detection, image fusion, map measurement or update processing based on remote sensing images , so it is necessary to correct the geometric distortion of the image, that is, geometric correction.
————————————————————————————————————
After all, geometric correction is to make the pixel points in the distorted image The position and the pixel corresponding to the reference image are in the same position, which is ideal. However, there are generally errors in this process, that is, the distorted image can only be found to be roughly the same as the position of the corresponding pixel in the reference image.
2.2 Types of geometric corrections
Geometric fine correction requires the use of ground control points for fine correction.
The geometric correction of remote sensing images is divided into geometric rough correction and geometric fine correction.
The rough geometric correction is based on the cause of the distortion, using the spatial position change relationship, the correction using the calculation formula and the obtained auxiliary parameters, also known as the system geometric correction.
Geometric fine-tuning refers to fine-tuning with ground control points. Geometric fine correction does not consider the causes of distortion, and directly uses ground control points to establish a mathematical model between pixel coordinates and target geographic coordinates to realize the transformation of pixel positions in different coordinate systems.
2.2.1
According to the different sources of ground control points, geometric fine correction can be divided into image-to-image geometric correction, image-to-map geometric correction and geometric correction with known geometric information.
Among them, image-to-image geometric correction is the most commonly used method. The picture shows the selection of ground control points before image-to-image geometric correction. The ground control points should be selected in the same place on two different images.
Just as the purpose of geometric correction described above, after selecting the ground control point here, from the actual point of view of the figure, it is to rotate the distorted image to the same as the reference image.
2.3 The process of geometric correction
Geometric correction involves two processes: 1) transformation of spatial position (coordinate transformation); 2) resampling of gray values.
2.3.1 Geometric Correction - Method of Coordinate Transformation
In the coordinate transformation, the indirect method is the most commonly used. The
coordinate transformation includes the direct method and the indirect method. The direct method calculates the coordinates of each pixel in the output image sequentially through the original image. Although the size of the pixel value output by the direct method does not change, the pixel distribution in the output image is not uniform. Therefore, the general forward derivation is not suitable. The schematic diagram of the direct method is as follows:
the indirect method starts from the output image, calculates the position of each pixel in the original image in turn, and then calculates the pixel value of the position in the original image. At this time, it may appear that the position of the calculated pixel point is not an integer, and its gray value can be obtained by using the weighting of its adjacent points. The schematic diagram of the indirect method is as follows:
this method can ensure that the pixels of the corrected image are evenly distributed in space, but requires grayscale resampling, and the indirect method is the most commonly used.
To summarize, the direct method of coordinate transformation is the original image to the output image, and the indirect method is the output image to the original image.
2.3.2 Coordinate Transformation - Resampling
The reason for resampling is as mentioned above using the indirect method, because it is derived from the position of a pixel in the output image (rows and columns are integers) to its position in the original image (rows and columns are not necessarily integers), because there are only integers in the image. Only when the row column is an integer has the grayscale value, which is determined by the nature of the image. And precisely because the point circled in red in the figure appears during the derivation, we need to assign a gray value to it. So there is grayscale resampling to solve this problem.
2.3.3 Selection of Ground Control Points
The selection of ground control points in coordinate transformation: 1) the ground control points should be obvious in the image, 2) the ground objects on the ground control points do not change with time, 3) the control points should be selected on the image without terrain correction. 4) The ground control points should be evenly distributed in the whole image, and there must be a certain number of guarantees.
The results of geometric correction are generally evaluated by the Euclidean distance (RMS) before and after the coordinate transformation of the ground control points.
2.3.4 Resampling method
The nearest neighbor method, bilinear interpolation, and cubic convolution method can produce smoother images in turn, and the visual effect is good, but the preservation of image spectral information will be weakened in turn. Therefore, if the resampling image needs to be used in the future, for example, for classification, vegetation coverage calculation, etc., it is recommended to use the nearest neighbor method to maintain spectral information.
————————————————————————————————————
Specifically ,
the advantage of the nearest neighbor method is that the method is simple, the processing It's fast, doesn't change the value of the original raster, but the processed image isn't smooth enough.
Bilinear interpolation gives smoother results than nearest neighbor, but changes the original grid values and loses some tiny features. It is suitable for continuous data representing the distribution of a certain phenomenon and the terrain surface.
The cubic convolution method can make the image smooth and the visual effect is good, but it will destroy the spectral information of the image. This method can be used when data processing based on spectral analysis is not required, but only for cartographic expression.
Orthocorrection
Orthophoto correction can not only achieve the function of conventional geometric correction, but also eliminate the geometric distortion of the image caused by terrain fluctuations by measuring the elevation point (ie the height of the point) and DEM, and improve the geometric accuracy of the image. Orthorectified images have precise spatial locations.
The orthorectified model is as follows:
