Spatial enhancement of digital terrain models using shape from shading with single satellite imagery.

摘要

In most of the geoscientific, environmental, engineering and military related research and applications, among others, Digital Terrain Models (DTMs) play a ubiquitous role. Depending on the region of interest, one may have access to a DTM with a reasonable accuracy and resolution or not. Regardless of what is available, there is always a demand for a denser, hence more accurate DTM.; This research work is an attempt to use Shape from Shading (SFS) with a single (as opposed to stereo) satellite imagery to enhance the interpolation accuracy of DTMs. The motivation is availability of relatively inexpensive yet globally available multiresolution, multispectral single satellite imageries.; Different SFS formulations as well as their corresponding solutions are studied and reviewed. Based on the characteristics of the specific problem of this investigation, the SFS is formulated and the general variational approach is selected as the solution method.; The main deficiency of the standard variational technique is its tendency to over smooth the recovered surface. The over smoothing is due to the smoothing constraint in the formulation of the SFS solution, presence of noise in the image and sudden discontinuities in the surface heights. A more intelligent way of handling the smoothness constraint, which can also distinguish the noise from sudden surface discontinuities, is to use robust statistics.; Using robust statistics, the smoothness constraint in the SFS is reformulated. Moreover, the brightness constraint is applied using the ambiguity cone concept. In addition to the numerical stability, this makes the SFS solution independent of the choice of the regularization factor. Furthermore, calibration of the albedo factor in the SFS formulation is done through classification of pixels using multispectral imageries. It is shown that the new formulation retrieves the original shape of the object much better than the standard variational method, especially in the presence of noise in the image and sudden discontinuities in the surface heights.; This research is of both theoretical and practical value in the context of its topic, as it not only develops a framework for the SFS formulation and solution, but also provides some valuable practical considerations.