Universität Rostock, 2011
Abstract: In this thesis, a comprehensive and unified stability analysis for a class of fast DCT (discrete cosine transform) and DST (discrete sine transform) algorithms is performed, both for fixed-point and floating-point arithmetic. Each of them is based on a factorization of the underlying orthogonal transform matrix into a product of sparse orthogonal matrices. Additionally to worst case analysis, also the average case is considered using stochastic models for the relative and absolute roundoff errors. Particularly with regard to applications in digital image processing, the stochastic analysis of roundoff error is done without assuming the data to be uncorrelated or independent.
doctoral thesis free access