Degradation of picture quality by speckle in coherent mapping systems

Citation

Korwar, Vijaya Narayan (1980) Degradation of picture quality by speckle in coherent mapping systems. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/edk6-jc81. https://resolver.caltech.edu/CaltechETD:etd-12082006-144128

Abstract

When a coherent imaging system is used to map a rough surface, for instance when a synthetic aperture radar (SAR) system is used to map the earth, the resulting picture of the surface is degraded by random intensity fluctuations called speckle. Speckle obscures the intensity variations caused by the inherent reflectivity differences that identify various features in the picture. The problem of specifying the extent of the degradation caused by speckle in pictures meant to be examined by a human observer was investigated. In particular, the problems considered were those in which an observer has to (a) detect a small feature immersed in a somewhat darker background; (b) detect a grating consisting of alternating bright and somewhat darker lines; (c) distinguish between two or four specified geometrical forms. In each case the picture was corrupted by speckle.

In investigating each of these problems, a plausible theoretical model was developed for the decision process used by the observer in his detection or discrimination task. This model was used to relate the probability of his making a correct decision to the relevant picture parameters such as contrast ratio between the reflectivities of various parts of the picture, number of looks per pixel, picture size, and dimension of the features or lines. These calculations were verified by experiments in which the decisions made by an observer examining computer simulations of speckle-corrupted pictures were noted.

Results of calculations for the reasonable SAP parameters of 1 dB contrast ratio and 12 looks per pixel showed that, in order to achieve a probability of correct decision of 0.95, (a) a small square in a 100 by 100 pixel background needs to be about 7 pixels on a side; (b) a 100 by 100 pixel grating of line pairs needs to have lines about 2 pixels wide; (c) a simple geometrical form (a specific one) needs to be at least 12 pixels on a side to be distinguished from another (specific) form of the same size, when these two forms are the only possible alternatives. These results illustrate that detectability of line pairs is a poor criterion for characterizing picture quality, while form discrimination imposes the most stringent requirements on the imaging system.

Thesis Files