1. ABSTRACT The problem of efficiently compressing a large number, L, of fl dimensional signal vectors is considered. The approach suggested here achieves efficiencies over current preprocessing and Karhunen-LoCve techniques when both L and N are large. Preprocessing and partitioning techniques are first ap plied to the L x hf data matrix 3 to reduce the database to a manageable number of subblocks of lower dimension. Within each subblock an iterative chain approhation is proposed that effects a transform at each stage of the iterative scheme. A particularly appealing transform, using prolate spheroidal sequences, is suggested. To evaluate a reduced dimensionality approximation for the expansion coefficients, the approach used in the orthogonal Procrustes problem solution is combined with an iterative interlacing technique due to Daugavet for factorizing matrices.
[1]
D. Slepian.
Prolate spheroidal wave functions, fourier analysis, and uncertainty — V: the discrete case
,
1978,
The Bell System Technical Journal.
[2]
David Casasent,et al.
KL Techniques For Optimal Processing Of Time Sequential Imagery
,
1989,
Other Conferences.
[3]
Michael Lindenbaum,et al.
Partial eigenvalue decomposition of large images using spatial temporal adaptive method
,
1995,
IEEE Trans. Image Process..
[4]
David Casasent,et al.
Karhunen-Loeve techniques for optimal processing of time sequential imagery
,
1991
.
[5]
Allen Gersho,et al.
Image compression with variable block size segmentation
,
1992,
IEEE Trans. Signal Process..