5.17.1.A. Geometric Corrections

Raw remote sensing data is without any geographic coordinates, and has distortions, mainly caused by the sensor geometry. Therefore, these can’t be used as such for any quantitative measurements on them. Georeferencing is the conversion of image coordinates to ground coordinates by removing the distortions caused by the sensor geometry. Georeferencing is important to deal with various images, create mosaicking and compare various scenes (e.g., change assessment). It is a process of locating an entity/object in real world coordinates, also called geo-rectification or geo-registration.

The direction of satellite motion in the orbit and on-board sensors while taking images is not exactly north-south or west-east, respectively. In addition, there is a rotation of the Earth about its own axis while taking the images, so images are not perfect square but they have somewhat skewed shape. Georeferencing re-orients the image to a coordinate system representing the Earth, and making its geometry same as the Earth. Georeferenced images can be viewed, compared, and analysed with other geographic data.

To do georeferencing, the exact locations of several known points, called Ground Control Points (GCPs), are required. These GCPs are normally selected as prominent objects whose geographical locations can be accurately determined either from the topographic maps or GPS survey. A minimum of four control points are required for georeferencing, however, additional control points would help increasing the accuracy of georeferencing. These GCPs are also identified on the image to be georeferenced. With these two sets of coordinates, polynomial is fitted amongst the GCPs, and rms error is minimized to ±1 pixel size. After georeferencing, each point on the image has real-world coordinates associated. The accuracy of the georeferencing would depend on the number, accuracy, and distribution of the control points and the choice of transformation polynomial. Normally, 2nd or 3rd order polynomial is used.

Georeferencing is the process that aligns an image with real-world coordinates so it can be accurately interpreted and analyzed. Raw remote sensing data often lack geographic information and can appear distorted due to the angle at which they were captured. Georeferencing involves correcting these distortions so that the image can be matched to a map. For effective georeferencing, at least four known geographic points (Ground Control Points) are needed; these points should be prominent features that can be accurately located. The relationship between the image coordinates and the real world is established mathematically to improve the image's accuracy.

Imagine trying to fit a puzzle piece into the correct spot without knowing the shape of the puzzle. Georeferencing is like taking a map of the puzzle and adjusting the pieces to fit perfectly, aligning them so that when you look at your completed puzzle, it matches exactly where things are supposed to be in the real world.

After the georeferencing process, we may find that the pixels have been oriented differently than the way they were present in the original image coordinate system. Resampling is the process of interpolation the new DN values of the displaced pixels (new pixel location) in the new coordinate system. Three methods of resampling are commonly used, as given below:

(a) Nearest Neighbour: In this method, the attribute value of the original pixel nearest to a pixel in the output image is assigned to the corresponding cell.

(b) Bilinear Interpolation: It assigns the value to a pixel in the output image by taking weighted average of the surrounding four pixels in the original grid nearest to it.

(c) Cubic Convolution: It assigns the value to a pixel in the output image by taking weighted average of the surrounding sixteen pixels in the original grid nearest to it.

Among the three methods, nearest neighbour is a preferred method as it doesn’t alter the values of the original grid cells assigned to the resampled grid cells but it produces a blocky image. The cubic convolution on the other side does change the values but is more accurate. It generates a smoother image.

Resampling is necessary after georeferencing because it ensures that the pixels in the corrected image accurately reflect their new positions. The process uses interpolation techniques to assign new values to pixels based on their neighbors in the original image. There are three main methods: Nearest Neighbour, which takes the value from the closest pixel; Bilinear Interpolation, which uses the average of four surrounding pixels to determine a new value; and Cubic Convolution, which looks at sixteen pixels to create an even more refined average. Each method has its pros and cons in terms of accuracy and image quality.

Think of resampling like adjusting a recipe when you change the serving size. If you scale down the recipe, you can either take the closest measurement (nearest neighbour), use a balanced approach by averaging out some ingredients (bilinear), or carefully calculate based on a larger group of ingredients for precision (cubic convolution). Just like the recipe needs accurate proportions to taste right, images require accurate pixel values to look correct.

--