PLT versus KLT [transforms]

Author

Phoong, See-May ; Lin, Yuan-Pei

Author_Institution

Dept. of Electr. Eng. & Inst. of Commun. Eng., Nat. Taiwan Univ., Taipei, Taiwan

Volume

3

fYear

1999

fDate

36342

Firstpage

516

Abstract

In this paper a new class of transforms named Prediction-based Lower triangular Transform (PLT) is introduced for signal compression. The PLT is a nonunitary transform that yields the same coding gain as the Kahunen-Loeve transform (KLT). Unlike KLT, the derivation of PLT does not involve any eigen problem. An M-dimensional PLT can be solved by the Levinson-Durbin recursion formula in O(M²). Moreover the complexity of PLT is less than one half of KLT. For AR(1) inputs, the PLT has a closed form expression and needs only (M-1) multiplications

Keywords

data compression; prediction theory; transform coding; transforms; Kahunen-Loeve transform; Levinson-Durbin recursion formula; M-dimensional PLT; closed form expression; coding gain; complexity; four-channel PLT coder; ladder-based implementation; nonunitary transform; prediction-based lower triangular transform; signal compression; Costs; Decoding; Decorrelation; Discrete cosine transforms; Discrete transforms; Karhunen-Loeve transforms; Matrix decomposition; Signal processing; Statistics; Transform coding;

fLanguage

English

Publisher

ieee

Conference_Titel

Circuits and Systems, 1999. ISCAS '99. Proceedings of the 1999 IEEE International Symposium on

Conference_Location

Orlando, FL

Print_ISBN

0-7803-5471-0

Type

conf

DOI

10.1109/ISCAS.1999.778896

Filename

778896