Date of Award

12-1991

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Operational Sciences

First Advisor

David G. Robinson, PhD

Second Advisor

Thomas S. Kelso, PhD

Abstract

Imagery analysts are always looking for improved methods of analyzing digital satellite imagery. The resolution of satellite imagery can be improved by enlarging the images since the result will be a higher degree of discernable detail. Currently, nearest-neighbor, bilinear interpolation, and cubic convolution techniques are used for this purpose. The nearest-neighbor technique produces block-like images. The latter two methods produce sharp imagery, but the original information contained in the pixel values is changed in the process of convolving the image. These techniques cannot, therefore, be considered true representations of the original image. Kriging is a statistical technique which can be applied to enlarging satellite imagery. Specifically, it is a method of best linear unbiased prediction of spatial data. One of the benefits of kriging is that it is an exact interpolator: the original pixel values will not be modified in the resulting kriged images. This thesis develops the application of universal punctual kriging to the analysis of digital satellite imagery. Current convolution techniques and kriging are used to produce enlarged images and comparisons are made. Images are also sub-sampled and enlarged back to the original size using convolution and statistical methods. This allows the products of cubic convolution and kriging to be subtracted from the original image. This procedure provides an additional quantitative comparison of kriging and cubic convolution. Results indicate that kriging performs as well as or better than cubic convolution when used to enlarge images.

AFIT Designator

AFIT-GSO-ENS-91D-13

DTIC Accession Number

ADA244014

Comments

The author's Vita page is omitted.

Download

Share

COinS