A rotational clutter metric

مرکز و کتابخانه مطالعات اسلامی به زبان های اروپایی

منو

" A rotational clutter metric "
Salem Mohammad Salem C. Halford

Document Type	:	Latin Dissertation
Language of Document	:	English
Record Number	:	54359
Doc. No	:	TL24313
Call number	:	‭3293773‬
Main Entry	:	Salem Mohammad Salem
Title & Author	:	A rotational clutter metric\ Salem Mohammad Salem
College	:	The University of Memphis
Date	:	2007
Degree	:	Ph.D.
student score	:	2007
Page No	:	55
Abstract	:	A new clutter metric is proposed that is based on the Log-Gabor filter hank response of the clutter images. Increased presence of clutter in images triggers a higher response in some LG filters more than in others. The clutter problem considered is a classification issue. Thus, our goal is to find a set of LG filters that are best suited to classify images into clutter levels. The clutter metric does not make any assumptions about what constitutes clutter, but rather teams from human input. A human expert observer labels a set of training images into discrete clutter levels. The clutter metric relies on the labeling to learn what constitutes clutter. An extension of the metric is proposed where the image labels are extracted from the results of human perception experiments. The metric converts human response into a set of weights using fuzzy logic. The weights bias the metric output to maximize correlation with the human response. Linear Discriminant Analysis (LDA) is applied as a pre-processing step to classification. LDA is an excellent tool for feature reduction based on class separability, but it suffers from a few limitations. A new approach to LDA, called Rotational LDA (RLDA), is proposed using orthonormal geometric rotations. Each rotation brings the LG feature space closer to the LDA solution. As target size changes, what constitutes clutter also changes. To understand the effects of target size on LG response, RLDA was applied at different target sizes. The outputs are easily related since they are functions of orthonormal rotation angles. Finally, Bayesian decision theory was used to team class boundaries for clutter levels at different target sizes.
Subject	:	Applied sciences; Clutter metric; Log-Gabor filters; Rotational LDA; Electrical engineering; 0544:Electrical engineering
Added Entry	:	C. Halford
Added Entry	:	The University of Memphis