Geometric Correction

Sources of Geometric Distortions

Geometric distortions are typically classified into (a) internal distortion resulting from the geometry of the sensor and Earth rotation or curvature characteristics and (b) external distortions resulting from the attitude of the sensor or the shape of the object. Geometric distortions can be divided into systematic (predictable) source of the error or nonsystematic (random). Internal errors include errors of skew, scanning system, and relief displacement. External errors include altitude alterations and attitude changes (roll, pitch, and yaw).

Methods and Procedure

Some internal distortions are predictable or systematic and are generally corrected at the ground station or by image vendors. There are two types of correction: (1) image-to-map rectification and (2) image-to-image rectification. In the image-to-map rectification, a map (with correct projection) is used to correct images and convert them into a planimetric system. This method should be used when area, direction, or distance measurements are needed after correction. Image-to-image correction involves matching the coordinate systems or column-and-row systems of two digital images. In this case, both images should cover the same geographic area.

The general procedure involves (a) registration: selecting ground control points (GCP) and image pixel coordinates (row and column) with their map coordinate counterparts (e.g., meters northing and easting in a UTM [Universal Transverse Mercator] map projection); (b) spatial interpolation (also called transformation): identifying the geometric relationship between input coordinates and output coordinates; and (c) intensity (or pixel value) interpolation (also called resampling): determining the pixel value of the output image from the input image. For the registration, GCPs can be easily identified on the imagery and located accurately on a map (e.g., road intersections, lake edges, river junctions). For spatial interpolation, the paired coordinates (i, j and x, y) from many GCPs (e.g., 20) can be modeled to derive geometric transformation coefficients. These coefficients [Page 1240]are then used in polynomial equations for computing root mean square error (RMSE). RMSE ideally is to be within 1 pixel. For intensity interpolation, new values are assigned in one of three ways: (1) nearest neighbor, (2) bilinear, and (3) cubic.