4.6 Two Images - Not One!

Viewing imagery of the Earth's surface in stereo provides image interpreters with more qualitative and quantitative information than a single image. Cartographers, engineers, foresters, geologists, hydrologists and other scientists have traditionally used stereo viewing in their work. A stereo-model can convey information about slopes, shapes of landforms, and elevations much more clearly than a 2-D representation. Furthermore, stereo-models may show relationships between landforms and vegetation, mineral outcrops, or drainage systems that are not obvious from a single image.

A stereo pair also provides more quantitative information than a single image. This is important for mapping applications. In the monoscopic method, a single, two-dimensional image, with complementary data provided by a one-dimensional digital elevation model (DEM), can be rectified to generate an ortho-image. This is an image that is planimetrically correct, has an orthogonal projection, and can be superimposed on a map or other geocoded data. Like a map, X and Y cartographic coordinates can be extracted from ortho-images. However, they do not contain any height, or Z information, which can be obtained directly from the DEM. In contrast, with the stereoscopic method, a stereo pair can be directly used to extract the X, Y, and Z cartographic coordinates of ground features from the "virtual" stereo-model.

2D versus 3D viewing methods

Why can a stereo pair directly provide X, Y and Z coordinates? The answer is found in mathematics. Mathematically, an image is the representation of 3-D space in a 2-D medium. The physical constraint between the two spaces corresponds to the condition of collinearity well known in photogrammetry. This condition states that the exposure centre, the location of a ground point and its image point all lie on the same line. As a result of the collinearity condition, one degree of freedom is lost between 3-D reality and the 2-D image.

Consequently, extra information is necessary to reconstruct the third dimension from 2-D images. A single image can only give the row and line coordinates of image points. This is why terrain elevation information must be present in the ortho-rectification process. The 2-D single image added to the 1-D DEM gives us three pieces of information. In the case of a stereo pair, we have four pieces of information - two sets of row and line coordinates. Consequently, a stereo pair provides supplementary information for extracting the three X, Y and Z coordinates.

The ortho-image rectification process (monoscopic method) is dependent on the accuracy of the DEM. Errors in the DEM will propagate through ortho-image generation and feature extraction processes. As a result, planimetric features extracted from the ortho-image could be less accurate than those extracted from the stereo-model. In the stereoscopic method, planimetric feature extraction is independent of the altimetry.

Another advantage of stereo viewing is that it can be done directly from raw images. Resampling, which can degrade image radiometry and geometry thereby reducing interpretability, is unnecessary. On the other hand, resampling is part of ortho-image generation, adding another source of error to the process.

It has been proven that the ability to identify certain features in mono images is more difficult than in stereo. In thematic applications, stereo display gives a clear perception of terrain to better locate features such as control points, forest cover, clear cuts, river and stream beds.