BibTeX
@MISC{Dionysian94variable-precisionarithmetic,
author = {Raffi Dionysian},
title = {Variable-Precision Arithmetic for Vector Quantization},
year = {1994}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
This research proposes and investigates a method for the storage and computation in Vector Quantization (VQ) -- a promising technique for image/speech compression. The improvement is in the representation and arithmetic algorithm; the idea is independent of the technology and accommodates different search algorithms. Specifically, with simple lossless compression, the codebook storage in tree searched VQ is reduced more than 20%. For large codebooks, the simulations predict that the compression would be more than 40%. The compression of codevectors is achieved with Variable-Precision Representation (VPR), where we eliminate the sign extension bits. By categorizing vectors, VPR uses non-stationary nature of codevectors. Entropy measure shows that VPR compresses at least 75% as well as Huffman coding of vector elements. In conjunction
Keyphrases
vector quantization variable-precision arithmetic codebook storage variable-precision representation accommodates different search algorithm simple lossless compression vector element large codebooks arithmetic algorithm non-stationary nature huffman coding entropy measure image speech compression promising technique sign extension bit
