Use of texture filters to improve quality of digital elevation models derived from stereo imagery

Fundamental to automated digital elevation models (DEM) generation from stereo imagery is the image matching process which finds common or conjugate points (pixels) on the two images in a stereo pair. Accurate identification of conjugate points requires that corresponding pixels should have distinct spatial features to be distinguished from other pixels in the surrounding. However, in homogeneous areas such as water bodies or closed canopy forests, it is very difficult to find conjugate pixels and hence, elevation values are interpolated rather than precisely calculated. As a result, most of the erroneous elevation values in the stereo DEM are produced in such homogeneous areas. This paper presents a method for improving the quality of stereo DEM by utilizing texture filters. An entropy filter is applied to one of the images in a SPOT level 1B panchromatic stereo pair to extract water areas before generation of the DEM. Pixels in the extracted water areas are excluded from being candidates in the stereo matching process. Further, a statistical texture filter is applied to the generated elevation values before the interpolation process to remove the remaining anomalous elevation values. The results show that the utilization of texture filters, as pre and post processors to image matching, dramatically improved the quality of the automatically generated DEM, especially in homogeneous areas.