Navigation

Image Processing

The data from flying or orbiting sensors are initially uncorrected for radiometric and geometric discrepancies; they are considered "raw" (some users prefer that status, so that they can apply corrections to their own specifications). However, most users prefer to have errors and corrections made by the supplier (usually the organization that receives the data stream telemetry or, sometimes, the secondary distributor). The subject of correction is tied to the procedures called pre-processing or image restoration. The treatment of these modifications is extensive and will not be covered in this Appendix, other than to mention the principal actions normally made in adjusting the DN values.

The first group is Geometric: These include correcting for skew - the effect owing to rotation of the Earth, and hence the ground target that moves progressively underneath the advancing spacecraft - that gives rise to the rhombic shape of a printed image; also, the space (or air) craft is subject to platform movements, called pitch, roll, and yaw that cause the straight-down line of sight to deviate from the vertical; the pixels acquired off nadir (vertical) along the line of mirror traverse are progressively elongated depending on the look angle; the flat image of the surface is distorted by the natural curvature of that surface so that some projection compensation is needed - the data require reference to a coordinate system and those most commonly applied are the Universal Transverse Mercator and Space Oblique Mercator projections. Once the various corrections are made, the result is usually a shift in position of any given pixel into its new framework, such that it does not necessarily have the same DN values that it had in the original (distorted) position; a new set of values can be calculated by resampling, a mathematical process involving interpolation of values using some algorithm such as Nearest-Neighbor, Bilinear, or Cubic Convolution.

Corrections can be made that affect undue Radiometric variations, i.e., changes in the measured radiances owing to a variety of factors, conveniently divided into natural and instrumental. One natural condition relates to day to day and place to place differences in atmospheric conditions. For example, presence of water vapor influences the radiance, so that its values would not be the same for the same target type if it were present in the tropics as compared with the dry arctic. Another, not always applied, correction takes into account the changes in sun angle (seasonal elevation; time of day). Instrument corrections involve variations in detector response and electronic perturbations. Most common are systematic differences in one or more detectors (such as in the MSS) or in individual CCD chips. This can induce such effects as variable line darkening (one detector may produce a line that is brighter or darker than its neighbors), line drop out (a fluctuation may cause all or part of a line to be missing), and random noise (speckling). Procedures are available to apply computer-generated corrections for any of these, improving overal image quality.

With the advent of micro-computers, the cost of image processing has greatly dropped while quality of the output may be superior to older, large machine methods. Data can now be supplied on 5.25 or 3.5 inch floppy discs, CDs, and other storage devices. Commonly, one or more discs are needed to provide the complete set of data; a full Landsat scene can be included but requires systematic dropping of multiple pixels (thus, say, 1 of every 4 in the sequence is retained); CDs or magnetic tapes (Jaz; ZIP) have the capacity to contain a full scene. But, typically these storage media contain subscenes, such as the 512 x 512 subsets (in byte binary format) we use in PIT .

Armed with this overview, we are now ready to get you directly involved in the procedures of space image processing.

Navigation


Nicholas M. Short, Sr. email: [email protected]
Jeff Love, PIT Developer ([email protected] Next Previous Next Table of Contents Previous