Accession Number : ADA300388
Title : Matched Wavelet Construction and Its Application to Target Detection.
Descriptive Note : Doctoral thesis,
Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH
Personal Author(s) : Chapa, Joseph O.
PDF Url : ADA300388
Report Date : 20 OCT 1995
Pagination or Media Count : 167
Abstract : This dissertation develops a new wavelet design technique that produces a wavelet that matches a desired signal in the least squares sense. The Wavelet Transform has become very popular in signal and image processing over the last 6 years because it is a linear transform with an infinite number of possible basis functions that provides localization in both time (space) and frequency (spatial frequency). The Wavelet Transform is very similar to the matched filter problem, where the wavelet acts as a zero mean matched filter. In pattern recognition applications where the output of the Wavelet Transform is to be maximized, it is necessary to use wavelets that are specifically matched to the signal of interest. In this dissertation, an algorithm for for finding both symmetric and asymmetric matched wavelets is developed. It will be shown that under certain conditions, the matched wavelets generate an orthonormal basis of the Hilbert space containing all finite energy signals. The matched orthonormal wavelets give rise to a pair of Quadrature Mirror Filters (QMF) that can be used in the fast Discrete Wavelet Transform. It will also be shown that as the conditions are relaxed, the algorithm produces dyadic wavelets when used in the Wavelet Transform provides significant redundancy in the transform domain. Finally this dissertation develops a shift, scale and rotation invariant technique for detecting an object in an image using the Wavelet Radon Transform (WRT) and matched wavelets. The detection algorithm consists of two levels. The first level detects the location, rotation and scale of the object, while the second level detects the fine details in the object. Each step of the wavelet matching algorithm and the object detection algorithm is demonstrated with specific examples. (AN)
Descriptors : *IMAGE PROCESSING, *PATTERN RECOGNITION, *TARGET DETECTION, ALGORITHMS, SIGNAL PROCESSING, FOURIER TRANSFORMATION, SPATIAL DISTRIBUTION, OPTIMIZATION, MATRICES(MATHEMATICS), TARGET RECOGNITION, RESOLUTION, CLUTTER, WHITE NOISE, THESES, CORRELATION, MATHEMATICAL FILTERS, LEAST SQUARES METHOD, SCALING FACTOR, DATA COMPRESSION, NORMALIZING(STATISTICS), INVARIANCE, MATCHED FILTERS, MATCHING, BACKGROUND NOISE, HILBERT SPACE, IMAGE RESTORATION, PHASE SHIFT, SPREAD SPECTRUM.
Subject Categories : Cybernetics
Distribution Statement : APPROVED FOR PUBLIC RELEASE