Continuous Piecewise Geometric Rectification for Airborne Multispectral Scanner Imagery

Geometric rectification of airborne multispectral scanner image data using traditional polynomial functions often cannot provide satisfactory RMSE accuracy due to the complex nature of geometric distortions in the data. The discrete approach to rectifying these data generates segmented pieces that may cause edge-matching problems after they are reassembled. To improve the rectification accuracy while retaining the continuity of the rectified strip, a continuous piecewise geometric rectification approach is introduced. Using logical divisions of a strip and the concept of overlapping area anchor ground control points, the approach localizes the complex distortion and greatly improves the edge-match between pieces. A description of the procedure is presented along with two case studies, each having a diffrent set of sensor and terrain characteristics, to illustrate the advantages of this approach versus traditional techniques. Introduction In meeting today's pressing needs for comprehensive evaluation of natural resources and environmental conditions, more and more airborne multispectral and hyperspectral data will be acquired for integration into geographic information systems (GIS). These data are often subject to complex geometric distortions and in need of image-to-map georectification before they can be incorporated into a GIS database (Lunetta et al., 1991; Estes and Star, 1993). Bivariate polynomial functions are a common method for modeling image distortion in such a rectification process (Markarian et al., 1973; Novak, 1992; Ehlers and Fogel, 1994; Jensen, 1996). Few complaints are found in the literature regarding the use of bivariate polynomials for the rectification of spaceborne imagery. This is because satellite platforms remain relatively stable in relation to their orbital altitude; any deviation of attitude from normal is usually minor and systematically spread over the entire scene (Christensen et al., 1988). In addition, the high orbital altitude and the small field-of-view hold the usual tangential scale distortion of the scan line to rninimum (Ehlers and Fogel, 1994). The combined effects characterize satellite imagery as having minimal linear geometric distortion that can be adequately modeled by first-order polynomials and usually rectified to within a -+ 1 pixel root-meansquare error(Rh4SE) (Adeniyi, 1985; Novak, 1992). In contrast, airborne remotely sensed images usually suffer more severe and complex geometric distortion due to low platform altitude and large field-of-view, wind and turbulence conditions in the lower atmosphere, and variation of altitude and velocity of the aircraft during data collection. Previous studies document the difficulty and frustration of rectifying airborne scanner data. Gillespie and Kahle (1977) noticed large residual geometric errors after rectifying airborne digital thermal inertia images and employed an additional registration step to reduce the error to an acceptable level. Otepka (1978) resorted to a commercial orthophoto system to differentially rectify the airborne scanner data to a line map or orthophotograph using amethod developed by Kraus (1978). Jensen et al. (1983) found attitude motions of the aircraft platform made the polynomial-based geocorrection acceptable on only small areas. Christensen et al. (1988) investigated a discrete approach to tackle the problem of airborne image-to-image registration for wetland change detection. In these published cases, geometric correction is both labor and computationally intensive. This classic problem recently received a recurrent attention, and most of the research was aimed at providing rigorous solutions. Zhang et al. (1994) developed an orthoimage procedure to model the dynamics of aircraft motion and terrain effects using flight attitude parameters and a digital elevation model. Ehlers and Fogel (1994) investigated the multiquadric method and radial basis functions for the adaptive modeling of complex distortions of airborne scanner imagery. An artificial neural network approach was developed by NASA staff to segment the airborne scanner image strip for piecewise training and rectification (Kiang, 1997). These methods either require on-board flight parameters for model construction or are not readily available to the general users of airborne scanner imagery. This paper introduces a practical approach known as Continuous Piecewise Geometric Rectification to improve the bivariate polynomials-based georectification of airborne scanner data. The approach is based on the mathematical model most widely available in commercial digital image processing packages. With an adequate number of ground control points (GCPS) and a relatively small amount of effort, it can correct the distortions of an entire airborne image strip to the subpixel level. The approach also retains the continuity of the rectified strip even if the rectification is performed in a piecewise manner. This paper discusses the methodology used to test the approach, and then demonstrates with two case studies that have different terrain characteristics. Methodology Bivariate polynomials belong to the family of approximating functions as opposed to the other family of interpolating functions for image rectification (Fogel, 1996). The advantages of M. Ji is presentIy with the Department of Geography, University of North Texas, P.O. Box 305279, Denton, TX 76203 J.R. Jensen is with the Department of Geography, University of South Carolina, Columbia, SC 29201 ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 66, No. 2, February 2000, pp. 163-171. 0099-1112/00/6602-163$3.00/0